Le patologie neurologiche sono responsabili del più alto tasso di disabilità e del secondo più alto tasso di mortalità a livello globale (Feigin et al., 2020). L’imaging medico nel campo della neurologia si basa principalmente su modalità che generano grandi quantità di dati complessi e comprende la risonanza magnetica (RM), la tomografia computerizzata (TC) e l'imaging nucleare. Una grande parte della ricerca sulle applicazioni dell’intelligenza artificiale (IA) in radiologia è stata dedicata alle patologie neurologiche. Infatti, il 29-38% delle applicazioni basate sull’IA per la radiologia disponibili in commercio è destinato all’imaging cerebrale o della colonna vertebrale, una percentuale più elevata che per qualsiasi altra zona anatomica (AI Central).

La maggior parte di queste applicazioni ha la funzione di supportare il radiologo nell’interpretazione delle immagini, ad esempio rendendo questa attività più efficiente o ampliando le capacità del radiologo, ad esempio fornendo una quantificazione più dettagliata dei dati di neuroimaging (Olthof et al., 2020). In questa pubblicazione si descrivono le applicazioni di IA più comuni usate in neuroradiologia e se ne analizzano le evidenze.

Emorragia intracranica

L’emorragia intracranica (Acute Intracranial Hemorrhage, ICH) acuta colpisce circa 3,4 milioni di persone ogni anno in tutto il mondo (World Stroke Organization, 2022). L’ICH è associata a un’elevata morbilità e mortalità e spesso richiede un pronto intervento neurochirurgico o uno stretto follow-up clinico e di imaging (Broderick et al., 2007; van Asch et al., 2010). Individuare un’emorragia intracranica acuta è di fondamentale importanza soprattutto nei pazienti che presentano deficit neurologici acuti e sospetto di ictus, in quanto costituisce una controindicazione assoluta alla trombolisi endovenosa (Fugate & Rabinstein, 2015).

In situazioni di emergenza, i casi sospetti di ICH di solito vengono inizialmente studiati mediante TC dell'encefalo senza mezzo di contrasto, perché la TC è un esame ampiamente disponibile, rapido, altamente sensibile per l’ICH e che presenta relativamente poche controindicazioni (A. Jain et al., 2021). L’alternativa è la RM, che è più sensibile alle emorragie molto piccole e croniche ma è più lenta, meno facilmente disponibile, più costosa e controindicata in alcuni pazienti (Chalela et al., 2007).

In uno studio condotto per determinare i modelli di errore da parte degli specializzandi in radiologia nel rilevare l’ICH, i ricercatori hanno riscontrato discrepanze nel 4,6% degli esami notturni interpretati dagli specializzandi, e di questa percentuale il 13,6% era dovuto a emorragie non incluse o descritte in modo inaccurato nei referti degli specializzandi (Strub et al., 2007). L’ICH può essere classificata come emorragia intraparenchimale, emorragia subdurale, emorragia extradurale ed emorragia subaracnoidea. Di queste, le emorragie subdurali e subaracnoidee sono quelle che più spesso non vengono rilevate, soprattutto se sono molto piccole (Strub et al., 2007).

Inoltre, nei referti degli specializzandi in radiologia vengono spesso scambiati per emorragia intracranica alcuni aspetti della normale anatomia del cervello e gli artefatti delle immagini (Erly et al., 2002).

La stragrande maggioranza delle applicazioni basate sull’IA per rilevare e classificare l’emorragia intracranica utilizzano la TC senza mezzo di contrasto come input e si basano sulle reti neurali convoluzionali (Convolutional Neural Networks, CNN). Con poche eccezioni (Bar et al., 2019; Wang et al., 2021; Ye et al., 2019), per la maggior parte delle applicazioni non sono facilmente disponibili descrizioni molto dettagliate dell'architettura di rete. La quantità e la qualità dei dati utilizzati per addestrare questi algoritmi varia ampiamente da centinaia (Bar et al., 2019; Heit et al., 2021) a migliaia (McLouth et al., 2021; Rava, Seymour, et al., 2021), a decine di migliaia (Chilamkurthy et al., 2018; Gibson et al., 2022; Ginat, 2021) di TC senza mezzo di contrasto.

Oltre alla classificazione della presenza o assenza di ICH, sono state utilizzate applicazioni con algoritmi basati sull'IA anche per classificare i sottotipi di ICH (Chilamkurthy et al., 2018; Gibson et al., 2022; Wang et al., 2021; Ye et al., 2019), per rilevare dati associati come l’effetto massa, lo spostamento della linea mediana e le fratture (Chilamkurthy et al., 2018) e per eseguire la segmentazione e la volumetria dell’emorragia (Bar et al., 2019; Gibson et al., 2022; Heit et al., 2021). Inoltre, un’applicazione basata sull'IA stima anche il grado di incertezza nella decisione dell’algoritmo, aiutando il radiologo a interpretarne l’output (Gibson et al., 2022).

Tra i sottotipi di ICH, le applicazioni basate sull'IA degli studi citati mostrano generalmente la massima sensibilità per l’emorragia intraventricolare (Chilamkurthy et al., 2018; Gibson et al., 2022; McLouth et al., 2021; Wang et al., 2021), molto probabilmente per via della grande differenza nella densità alla TC tra il liquido cerebrospinale e il sangue. Nel caso di tutte le applicazioni, la sensibilità è relativamente bassa per le emorragie subaracnoidee (Gibson et al., 2022; McLouth et al., 2021; Rava, Seymour, et al., 2021; Wang et al., 2021; Ye et al., 2019), probabilmente perché queste tendono ad essere piccole e/o adiacenti a strutture ossee o ad artefatti delle immagini TC iperdensi (ad esempio, nelle cisterne subaracnoidee). Altre applicazioni hanno mostrato anche una sensibilità relativamente bassa per l'emorragia subdurale, soprattutto quando si trova in posizioni meno comuni, come lungo la falce cerebrale (Chilamkurthy et al., 2018; Rao et al., 2021; Wang et al., 2021; Ye et al., 2019). La sensibilità tende ad essere inferiore anche per le emorragie più piccole, definite come <1,5 ml o <5 ml a seconda dello studio (Heit et al., 2021; McLouth et al., 2021; Rava, Seymour, et al., 2021). Solo in uno degli studi citati sono state studiate sistematicamente le differenze tra i marchi di macchinari e i parametri di scansione sulle prestazioni diagnostiche delle applicazioni basate sull'IA per il rilevamento dell'ICH (McLouth et al., 2021).

Alcuni studi hanno confrontato direttamente le prestazioni delle applicazioni basate sull'IA con quelle degli esperti. In uno studio su 160 TC senza mezzo di contrasto (49% con ICH) che ha utilizzato la valutazione di un consulente neuroradiologo come verità di base, una CNN U-Net ha mostrato sensibilità (91%) e specificità (89%) inferiori rispetto alle valutazioni di due specializzandi di neuroradiologia (sensibilità del 99-100% e specificità del 98%) (Schmitt et al., 2022). In un altro studio, le interpretazioni di un’applicazione basata sull’IA approvata dalla Food and Drug Administration (FDA) statunitense e dotata di marchio CE sono state confrontate con le letture di un gruppo di tre neuroradiologi clinici che hanno definito la verità di base.

L'applicazione basata sull'IA ha dimostrato la stessa sensibilità di uno specializzando di neuroradiologia che ha completato il training di specializzazione (91,9%), anche se la specificità dell'applicazione era sostanzialmente inferiore (applicazione: 84,4%; specializzando: 99,6%) (Eldaya et al., 2022). Un’altra applicazione basata sull’IA ha mostrato una sensibilità più elevata e una specificità leggermente inferiore per l’ICH rispetto ai tirocinanti in radiologia (Ye et al., 2019). L'ispessimento durale, le calcificazioni durali e intraparenchimali e gli artefatti da movimento o da striature hanno maggiori probabilità di essere scambiati per ICH dalle applicazioni basate sull'IA (Bar et al., 2019; Eldaya et al., 2022; Rao et al., 2021).

Molti studi hanno analizzato l’accuratezza diagnostica delle applicazioni basate sull’IA per il rilevamento dell’ICH, anche se un altro potenziale vantaggio dello screening per l’ICH basato sull’IA è che gli esami possono essere letti più velocemente, il che può portare a una gestione più rapida dei pazienti. Sebbene un numero minore di studi abbia valutato l’impatto dello screening basato sull’IA sui tempi, alcuni studi hanno dimostrato tempi di lettura più rapidi. In uno studio su 620 TC senza mezzo di contrasto, il tempo trascorso dalla fine dell’esame alla compilazione del referto è stato di 73 minuti quando l’IA ha segnalato al lettore umano di aver trovato qualcosa, rispetto ai 132 minuti trascorsi in assenza di tale segnalazione (Wismüller & Stockmaster, 2020). In un altro studio, l’utilizzo dell’applicazione basata sull’IA è stato associato a degenze più brevi dei pazienti nel pronto soccorso (561 minuti vs. 781 minuti senza l’IA) (Chien et al., 2022).

Ictus ischemico acuto

Occlusione dei grandi vasi

Nei pazienti con ictus ischemico acuto, l’identificazione rapida delle occlusioni dei grandi vasi cerebrali è essenziale per un trattamento tempestivo. In generale, il termine “occlusione dei grandi vasi” (Large Vessel Occlusion, LVO) si riferisce all’occlusione di arterie sufficientemente grandi da permettere la trombectomia meccanica. Attualmente si definiscono tali l'arteria carotide interna (Internal Carotid Artery, ICA), le parti prossimali dell’arteria cerebrale media (M1 e M2), anteriore (A1) e posteriore (P1), nonché l'arteria basilare (Mokin et al., 2019 ; Pirson et al., 2022).

Le occlusioni dei grandi vasi vengono rilevate direttamente mediante angiografia a sottrazione digitale, angio-TAC o angio-RM, oppure indirettamente con tecniche non angiografiche. All’angiografia, le occlusioni vascolari appaiono come un’improvvisa interruzione o del riempimento di un’arteria da parte del mezzo di contrasto (nell’angiografia con mezzo di contrasto) o del segnale di flusso (nelle tecniche senza mezzo di contrasto, come l’angio-RM TOF [timeof- flight]). Esse possono verificarsi con o senza la presenza di un riempimento del contrasto o di segnale di flusso distale al sito di occlusione. I segni indiretti di LVO all’imaging nelle tecniche non angiografiche possono essere un vaso iperdenso alla TC senza mezzo di contrasto (che rappresenta il trombo occlusivo) (Gács et al., 1983) e un segno di suscettibilità alla formazione di trombi su immagini di RM pesate in T2^* o per la suscettibilità (Flacke et al. , 2000).

La maggior parte delle soluzioni basate sull'IA per il rilevamento delle LVO utilizzano l’angio-TAC (Amukotuwa et al., 2019; Murray et al., 2020; Rava, Peterson, et al., 2021; Wardlaw et al., 2022; Yahav- Dovrat et al., 2021), mentre altre utilizzano la TC senza mezzo di contrasto (Lisowska et al., 2017; Olive- Gadea et al., 2020).

La maggior parte delle applicazioni è stata utilizzata soprattutto per il rilevamento delle LVO nelle arterie intracraniche della circolazione anteriore (Adhya et al., 2021; Amukotuwa et al., 2019; Dehkharghani et al., 2021; Rava, Peterson, et al., 2021), perché la trombectomia meccanica viene eseguita molto meno spesso nelle occlusioni dei vasi della circolazione posteriore (Adusumilli et al., 2022).

In una revisione delle evidenze relative alle applicazioni basate sull’IA per il rilevamento delle LVO, la sensibilità variava dall’80% al 96% e le specificità dal 90% al 98% (Wardlaw et al., 2022). I falsi positivi negli studi inclusi nella revisione delle evidenze erano per lo più dovuti a stenosi arteriosa, emorragia intracranica, tumori ipervascolari o occlusioni di vasi distali che non soddisfacevano i criteri di una LVO (Amukotuwa et al., 2019; Yahav-Dovrat et al., 2021). Purtroppo non sono disponibili dati pubblicati relativi alle prestazioni per una serie di soluzioni basate sull’IA con marchio CE, comprese alcune soluzioni specificamente studiate per il rilevamento delle LVO (van Leeuwen et al., 2021).

Al momento della stesura di questa pubblicazione, esiste un solo studio che ha analizzato il rapporto costo-efficacia degli strumenti basati sull’IA per il rilevamento delle LVO, la cui analisi ha dimostrato che, presupponendo che il 6% delle LVO non venga rilevato dai medici e che l'IA possa contribuire a dimezzare tale tasso, nel Regno Unito si potrebbe ottenere un risparmio di 11 milioni di dollari all’anno sui costi (van Leeuwen, Meijer, et al., 2021).

Poiché le LVO di solito vengono rilevate da parte dei radiologi e degli specializzandi in radiologia con le tecniche angiografiche (Duvekot et al., 2021), il principale vantaggio del rilevamento delle LVO con l’IA potrebbe essere la riduzione del tempo che trascorre fino all’inizio del trattamento grazie ad una valutazione più rapida. In alcune delle applicazioni attualmente disponibili ci vogliono da 1 a 3,5 minuti circa per elaborare i dati e prendere una decisione in merito alla presenza di una LVO (Amukotuwa et al., 2019; Dehkharghani et al., 2021; Olive-Gadea et al., 2020). Alcuni strumenti sono stati associati ad una riduzione di circa 22,5 minuti del tempo che trascorre dall’esecuzione dell’esame di imaging al trasferimento del paziente in un ospedale in grado di eseguire la trombectomia meccanica (Hassan et al., 2020), di circa 15 minuti del tempo che trascorre dall'arrivo del paziente in ospedale all’avviso dell’équipe neuro-endovascolare (Morey et al., 2021) e di circa 25 minuti del tempo che trascorre dall’imaging alla puntura inguinale per la trombectomia meccanica (Adhya et al., 2021).

Alterazioni ischemiche precoci del tessuto cerebrale

Alla TC, le prime alterazioni del tessuto cerebrale che si associano all'ischemia sono il gonfiore dei tessuti e la minore attenuazione dei tessuti dovuta all'edema ionico (Marks et al., 1999). Queste alterazioni sono incluse negli strumenti di valutazione visiva utilizzati dai radiologi, il più comune dei quali è l'Alberta Stroke Program Early CT Score (ASPECTS). Il punteggio ASPECTS può aiutare a prevedere sia gli esiti funzionali che lo sviluppo di emorragia intracranica sintomatica dopo la trombolisi endovenosa (Schröder & Thomalla, 2016). La maggior parte delle applicazioni basate sull'IA che mirano a rilevare le alterazioni ischemiche precoci alla TC senza mezzo di contrasto lo fanno fornendo una valutazione automatica del punteggio ASPECTS (Wardlaw et al., 2022). Altre applicazioni mirano a identificare le alterazioni ischemiche precoci mediante angio-TAC (Abdelkhaleq et al., 2021; Öman et al., 2019) o TC perfusionale (Hakim et al., 2021).

La maggior parte degli algoritmi basati sull’IA creati per identificare le alterazioni ischemiche precoci alla TC hanno utilizzato la valutazione visiva della TC senza mezzo di contrasto da parte di radiologi, neuroradiologi o altri medici come standard di riferimento (Goebel et al., 2018; Hoelter et al., 2020; Kniep et al., 2020; Maegerlein et al., 2019; Seker et al., 2019), mentre altri hanno utilizzato la RM pesata in diffusione (Abdelkhaleq et al., 2021; Herweh et al., 2016; H. Kuang et al., 2019; Qiu et al., 2020) o il core infartuale definito mediante TC perfusionale (Olive -Gadea et al., 2019). La maggior parte di queste applicazioni utilizzano algoritmi Random Forest (Guberina et al., 2018; Herweh et al., 2016; Kniep et al., 2020; H. Kuang et al., 2019; Maegerlein et al., 2019; Nagel et al., 2017; Olive-Gadea et al., 2019; Qiu et al., 2020) o reti neurali convoluzionali (Öman et al., 2019). Inoltre, molti studi si sono concentrati sull’identificazione automatica delle alterazioni ischemiche precoci alla RM pesata in diffusione (Boldsen et al., 2018; Mohd Saad et al., 2019; Nazari-Farsani et al., 2020; Siddique et al., 2022; Song, 2019; Wong et al., 2022), che è un metodo ad alta sensibilità ma non ampiamente disponibile in contesti acuti.

Come nelle applicazioni per le LVO, non sono disponibili dati pubblici sulle prestazioni di alcune soluzioni basate sull'IA con marchio CE per il rilevamento delle alterazioni ischemiche precoci (van Leeuwen et al., 2021). L'algoritmo per il quale sono disponibili più dati pubblicati è un approccio Random Forest alla valutazione del punteggio ASPECTS, che è risultato non inferiore ai neuroradiologi con una sensibilità del 44% e una specificità del 93% utilizzando la TC di follow-up come verità di base (Nagel et al. al., 2017). Un altro studio con lo stesso algoritmo e metodo di determinazione della verità di base ha mostrato come l'algoritmo avesse una sensibilità più elevata (83% vs. 73%), ma una specificità inferiore (57% vs. 84%) per il punteggio ASPECTS rispetto ai neuroradiologi (Guberina et al., 2018). In un terzo studio, questo algoritmo ha ottenuto risultati migliori nel punteggio ASPECTS rispetto ai neurologi e agli specializzandi in neurologia e ha ottenuto risultati simili rispetto ai neuroradiologi (Ferreti et al., 2020).

Nel complesso, pochi studi hanno confrontato direttamente diverse applicazioni basate sull’IA per il rilevamento delle alterazioni ischemiche precoci alla TC senza mezzo di contrasto (Goebel et al., 2018; Hoelter et al., 2020). Uno studio ha confrontato tre applicazioni disponibili in commercio (due basate sull’apprendimento automatico e una sulla densitometria) in 131 pazienti (Hoelter et al., 2020) e ha rilevato che le applicazioni basate sull’IA avevano un’area sotto la curva (Area Under Curve, AUC) compresa tra 0,73 e 0,76 rispetto al consenso di tre neuroradiologi.

La valutazione visiva delle alterazioni ischemiche precoci alla TC senza mezzo di contrasto è particolarmente difficile nella fossa cranica posteriore, dove si trovano spesso artefatti che ostacolano l'interpretabilità (Hwang et al., 2012). In una coorte di 69 pazienti con occlusioni dell'arteria basilare esaminati mediante TC senza mezzo di contrasto entro 6 ore dall'insorgenza dei sintomi, un algoritmo Random Forest ha identificato le alterazioni ischemiche precoci nella circolazione posteriore con un’AUC compresa tra 0,70 (nel cervelletto) e 0,82 (nel talamo) utilizzando come verità di base la TC senza mezzo di contrasto di follow-up (Kniep et al., 2020). Oltre alla posizione anatomica, ci sono anche vari altri fattori che influiscono sulla rilevabilità delle alterazioni ischemiche precoci alla TC senza mezzo di contrasto. Uno studio ha rilevato che l'accuratezza della valutazione ASPECTS differiva in base al tipo di ricostruzione usata per la TC, ma un algoritmo automatico è risultato più coerente in diverse ricostruzioni delle TC esaminate rispetto agli specializzandi o ai consulenti di radiologia (Seker et al., 2019). Inoltre, l'accuratezza delle valutazioni ASPECTS effettuate sia dalle persone che dall'IA sembra aumentare quando il tempo trascorso tra l’insorgenza dei sintomi e la TC senza mezzo di contrasto è più lungo, poiché le alterazioni ischemiche precoci diventano più pronunciate (Potreck et al., 2022).

Ictus con tempo di insorgenza sconosciuto

Sapere quanto tempo è trascorso dall’inizio dei sintomi dell’ictus è fondamentale per scegliere il trattamento più appropriato, poiché la trombolisi endovenosa è indicata solo se somministrata entro 4,5 ore dall’insorgenza dei sintomi (Powers et al., 2018). L'insorgenza dell'ictus non è sempre nota, come accade se il pazienta presenta ictus al risveglio. L'ictus al risveglio si verifica nel 14% circa dei pazienti, come riportato in uno studio condotto su una popolazione di pazienti che arrivano al pronto soccorso (Mackey et al., 2011). Sono stati proposti diversi approcci basati sulla diagnostica per immagini per identificare i pazienti all’interno della finestra temporale della trombolisi.

Un approccio ad oggi accuratamente studiato è quello della presenza di una lesione acuta da ictus alla RM con sequenze pesate in diffusione (Diffusion-Weighted Imaging, DWI) e la sua assenza alla risonanza magnetica FLAIR (Fluid Attenuated Inversion Recovery) (Ebinger et al., 2010; Thomalla et al., 2011; Thomalla et al., 2018). Anche l’interpretazione automatizzata delle immagini di DWI e RM FLAIR è diventata un obiettivo degli algoritmi basati sull’IA progettati per assistere i radiologi.

I metodi per la classificazione basata sull'IA dei tempi di insorgenza dell'ictus sono stati: le reti neurali convoluzionali (CNN) (Polson et al., 2022) o una combinazione di diversi algoritmi di apprendimento automatico (Jiang et al., 2022; H. Lee et al., 2020; Zhu et al., 2021). Alcuni studi hanno adottato un approccio basato sulla radiomica, che prevede la segmentazione delle lesioni DWI e FLAIR, l'estrazione da esse di diverse caratteristiche di imaging e quindi l'inserimento di queste caratteristiche in diversi algoritmi di classificazione (Jiang et al., 2022; H. Lee et al., 2020; Zhu et al., 2021).

In diversi studi, la classificazione basata sull’IA dei tempi di insorgenza dell’ictus ha prodotto sensibilità più elevate ma specificità inferiori rispetto alla valutazione visiva da parte dei radiologi (H. Lee et al., 2020; Polson et al., 2022). Sono state riportate sensibilità comprese tra il 73 e l'86% e specificità comprese tra il 68 e l'85% (Jiang et al., 2022; H. Lee et al., 2020; Polson et al., 2022; Zhu et al., 2021). Uno studio che ha utilizzato un approccio radiomico basato solo sulle immagini di DWI e pesate in T1 combinate con un algoritmo di deep learning ha rilevato una sensibilità del 95% e una specificità del 50% nell’identificare i pazienti all'interno della finestra temporale per la trombolisi (Y.-Q. Zhang et al., 2022).

Trauma cranico

Il trauma cranico (Traumatic Brain Injury, TBI) acuto è un trauma fisico improvviso che danneggia il cervello. Le sue manifestazioni comprendono l’ICH, il danno assonale diffuso e le fratture del cranio e del viso. Inoltre, con le tecniche di imaging è possibile rilevare le conseguenze di alcune di queste manifestazioni, come lo spostamento della linea mediana e l’ernia cerebrale, che, se gravi, possono richiedere interventi di emergenza (Schweitzer et al., 2019).

Sebbene le fratture composte del cranio senza ICH associata vengano trattate in modo conservativo (Skull Fractures, senza data), pochi studi hanno esaminato il loro rilevamento mediante tecniche basate sull'IA. Recentemente, però, sono stati compiuti alcuni tentativi di classificare le fratture del cranio rilevate mediante TC dell'encefalo senza mezzo di contrasto.

Un algoritmo basato su un approccio di apprendimento multietichetta e addestrato con 174 TC senza mezzo di contrasto (103 con fratture) ha mostrato una precisione del 98% e una specificità del 92% per il rilevamento delle fratture craniche (Emon et al., 2022). La precisione e la specificità più basse riguardavano le fratture depresse, mentre la precisione e la specificità più elevate riguardavano le fratture lineari e le fratture del viso. Un'applicazione basata sul deep learning volta a rilevare reperti critici in TC dell’encefalo senza mezzo di contrasto ha mostrato una sensibilità dell'81,2-87,2% e una specificità del 77,5-86,1% (a seconda del set di dati dell’esame) nel rilevare le fratture craniche (Chilamkurthy et al., 2018). Nello stesso studio, lo spostamento della linea mediana e l'effetto massa, entrambi conseguenze comuni dell'ICH correlata al trauma, sono stati identificati rispettivamente con una sensibilità dell'87,5-90,1% e del 70,9-81,2% e con una specificità dell'83,7-89,4% e del 61,6-73,4% (a seconda del set di dati dell’esame). Un algoritmo che combinava l'estrazione delle caratteristiche morfologiche del cranio con le CNN e che era stato addestrato con 25 TC senza mezzo di contrasto e testato su 10 TC dell'encefalo senza mezzo di contrasto di pazienti con trauma cranico ha mostrato una precisione media del 60% nel rilevare le fratture craniche (Z. Kuang et al., 2020). Un altro algoritmo di deep learning ha mostrato una sensibilità del 91,4% e una specificità dell’87,5% nell’identificare le fratture craniche in una serie di 150 TC dell’encefalo effettuate post-mortem (Heimer et al., 2018).

Patologie neurodegenerative

Molte malattie neurologiche possono essere descritte come neurodegenerative, ma il termine è solitamente usato per riferirsi alle malattie neurologiche croniche associate alla perdita graduale di tessuto cerebrale e che generalmente causano demenza e/o disfunzione motoria (Lamptey et al., 2022). Più di un quinto degli algoritmi basati sull’IA con marchio CE o approvati dalla FDA in neuroradiologia sono rivolti a pazienti affetti da demenza (IA for Radiology, senza data). La maggior parte di questi calcola automaticamente i volumi cerebrali regionali, misura lo spessore corticale e quantifica le lesioni della sostanza bianca causate dalla malattia dei piccoli vasi cerebrali (IA per Radiologia, senza data).

Molti algoritmi basati sull'IA specifici per la malattia sono stati sviluppati per rilevare la malattia di Alzheimer, che è patologicamente caratterizzata da placche extracellulari composte da β-amiloide e grovigli neurofibrillari intracellulari di proteina tau e porta a una progressiva compromissione cognitiva amnesica e non amnesica (Knopman et al., 2021). Alcuni di questi algoritmi sono in grado di distinguere tra soggetti con malattia di Alzheimer e soggetti con stato cognitivo normale mediante RM, con sensibilità comprese tra il 78% e il 99,1% e specificità comprese tra il 70% e il 92,68% (Battineni et al., 2022). Un approccio basato su macchine a vettori di supporto non lineari è stato in grado di differenziare tra malattia di Alzheimer e altre sindromi di demenza, come la degenerazione lobare frontotemporale, con una precisione dell'84% (Davatzikos et al., 2008).

Si è anche cercato di prevedere il passaggio dalla fase prodromica della malattia di Alzheimer all’Alzheimer conclamato, poiché si ritiene che nella prima fase gli interventi terapeutici possano essere particolarmente efficaci (Crous-Bou et al., 2017).

Nella compromissione cognitiva lieve (Mild Cognitive Impairment, MCI) il soggetto presenta deficit cognitivi più gravi di quelli che sarebbero previsti per la sua età, ma che non interferiscono ancora in modo significativo con le attività quotidiane (Petersen, 2016). Sono stati utilizzati diversi approcci basati sull’IA per prevedere il passaggio dalla MCI alla malattia di Alzheimer con una precisione del 66-92% (Amoroso et al., 2018; Bron et al., 2015; Lebedev et al., 2014; G. Lee et al., 2019; Lu et al., 2018; Moradi et al., 2015; Ocasio & Duong, 2021; Salvatore et al., 2015; Spasov et al., 2019).

La diagnosi precoce è considerata importante anche per il trattamento efficace della malattia di Parkinson (Pagan, 2012), un'altra malattia neurodegenerativa comune, caratterizzata patologicamente dalla degenerazione dei neuroni dopaminergici nella sostanza grigia. Quando compaiono i sintomi motori che suggeriscono una diagnosi clinica di malattia di Parkinson, si stima che oltre il 60% dei neuroni dopaminergici del cervello siano già andati perduti (GBD 2016 Parkinson's Disease Collaborators, 2018). Sono stati sviluppati diversi approcci di apprendimento automatico per distinguere tra malattia di Parkinson e controlli sani sulla base delle caratteristiche morfologiche rilevate alla RM strutturale (Adeli et al., 2016; Chakraborty et al., 2020; Peng et al., 2017), alla RM funzionale (Long et al., 2012; Pläschke et al., 2017; Tang et al., 2017), alla tomografia a emissione di positroni (PET) (Piccardo et al., 2021) e alla tomografia computerizzata a emissione di fotoni singoli (SPECT) (Choi et al. , 2017; Hirschauer et al., 2015; Ozsahin et al., 2020), spesso in associazione con punteggi clinici.

Poiché i sintomi motori della malattia di Parkinson sono simili a quelli di altre patologie neurologiche, le caratteristiche cliniche da sole spesso non sono sufficienti per diagnosticare con certezza questa patologia (Rizzo et al., 2016). Distinguere la malattia di Parkinson idiopatica dalle sindromi parkinsoniane atipiche come l'atrofia multisistemica e la paralisi sopranucleare progressiva sulla base delle caratteristiche cliniche è dunque particolarmente difficile (Rizzo et al., 2016). Sfruttando il potenziale del neuroimaging per contribuire a fare questa distinzione, uno studio iniziale ha utilizzato l'apprendimento automatico basato sul metodo dei vettori di supporto per classificare la malattia di Parkinson idiopatica e altre cause di parkinsonismo utilizzando l'imaging con tensore di diffusione, con una sensibilità del 94% e una specificità del 100% (Haller et al., 2012). Vari altri studi hanno mostrato un’elevata accuratezza nel distinguere tra malattia di Parkinson idiopatica e parkinsonismo atipico mediante RM strutturale (Duchesne et al., 2009; Focke et al., 2011; Huppertz et al., 2016; Marquand et al., 2013; Salvatore et al., 2014), imaging pesato in suscettività magnetica (Susceptibility Weighted Imaging, SWI) (Haller et al., 2013) e una combinazione di imaging con tensore di diffusione e RM strutturale (Cherubini et al., 2014).

Sono stati condotti studi anche utilizzando modelli di apprendimento automatico per orientare i medici nella scelta del trattamento della malattia di Parkinson. Uno studio condotto su 67 pazienti con malattia di Parkinson ha scoperto che le caratteristiche estratte dalla RM funzionale possono classificare parametri ottimali e subottimali per la stimolazione cerebrale profonda, con una precisione dell'88% (Boutet et al., 2021). Questo potrebbe aiutare a ottimizzare il processo, attualmente lungo, costoso e ingombrante, degli esami clinici approfonditi necessari per ottimizzare i parametri per la stimolazione cerebrale profonda nei pazienti con malattia di Parkinson.

Sclerosi multipla

La sclerosi multipla (SM) è una malattia autoimmune comune del sistema nervoso centrale, caratterizzata patologicamente da demielinizzazione infiammatoria e che porta a un’ampia gamma di manifestazioni neurologiche (McGinley et al., 2021). La RM svolge un ruolo importante nella diagnosi e nella gestione della SM ed è la tecnica di imaging preferita per quantificare e classificare le lesioni da SM nel cervello e nel midollo spinale (Matthews et al., 2016). Le caratteristiche dell'imaging sono una parte essenziale dei criteri diagnostici per la SM (Thompson et al., 2018), e le linee guida raccomandano l’uso della RM per monitorare i pazienti e orientare il trattamento (Wattjes et al., 2015). Vari algoritmi basati sull’IA hanno ricevuto l’autorizzazione della FDA e la certificazione CE per la quantificazione dell’atrofia cerebrale e la segmentazione automatizzata delle lesioni nella SM (Cavedo et al., 2022; Qubiotech Neurocloud Vol, 2021; Zaki et al., 2022).

Nella SM, molti algoritmi basati sull’IA si concentrano sull’estrazione automatizzata delle caratteristiche dell’imaging (Afzal et al., 2022; Bonacchi et al., 2022; Eichinger et al., 2020; Moazami et al., 2021). La valutazione visiva della presenza di lesioni da SM e della loro progressione nel tempo è una parte importante della diagnosi e del monitoraggio di questa malattia, ma richiede tempo ed è difficile (Danelakis et al., 2018). Sono stati quindi sviluppati diversi approcci tradizionali di apprendimento automatico (Brosch et al., 2016; Goldberg-Zimring et al., 1998; Karimian & Jafari, 2015; Samarasekera et al., 1997; Schmidt et al., 2012; S. Zhang et al., 2018) e deep learning (Birenbaum & Greenspan, 2017; Deshpande et al., 2015; Roy et al., 2018; Valverde et al., 2017, 2019) per la segmentazione automatica delle lesioni da SM. Circa il 30% di questi studi ha utilizzato le CNN e il 40% ha utilizzato approcci di apprendimento automatico basato sul metodo dei vettori di supporto (Afzal et al., 2022).

Gli approcci di deep learning hanno prodotto indici di similarità (una misura di sovrapposizione spaziale che va da 0 a 1) compresi tra 0,52 e 0,67 rispetto alle segmentazioni manuali delle lesioni (Afzal et al., 2022). Sono stati studiati (Dolz et al., 2018; Kushibar et al., 2018; Wachinger et al., 2018) anche diversi approcci basati sull’IA per la quantificazione automatica dell’atrofia cerebrale, che costituisce un altro predittore dell’evoluzione della SM mediante imaging (Andravizou et al., 2019).

Sono stati sfruttati algoritmi basati sull’IA anche per identificare anomalie alla RM che non erano chiaramente visibili a occhio nudo e non erano incluse negli attuali criteri diagnostici per la SM, come le anomalie delle vene cerebrali e i depositi di ferro rilevati mediante SWI (Lopatina et al., 2020) e le anomalie in aree apparentemente normali della sostanza bianca e grigia in sequenze di RM sia convenzionali (Eitel et al., 2019) sia avanzate (Neeb & Schenk, 2019; Saccà et al., 2019; Yoo et al., 2018; Zurita et al., 2018).

Prima di diagnosticare una sclerosi multipla occorre escludere le malattie che hanno una presentazione clinica simile, ma a volte questo compito è difficile (Wildner et al., 2020). Utilizzando le caratteristiche estratte dalla RM, gli algoritmi Random Forest e le CNN si sono dimostrati accurati nel distinguere tra SM e disturbi dello spettro della neuromielite ottica (Eshaghi et al., 2016; Rocca et al., 2021), disturbi non infiammatori della sostanza bianca (Mangeat et al., 2020; Theocharakis et al., 2009), emicrania (Rocca et al., 2021), vasculite del sistema nervoso centrale (Rocca et al., 2021) e tumori cerebrali (Ekşi et al., 2021).

La SM si suddivide in diversi fenotipi clinici che hanno prognosi e strategie di trattamento ottimali diverse (Lublin et al., 2014). Con la RM con tensore di diffusione (Kocevar et al., 2016; Marzullo et al., 2019), la spettroscopia a risonanza magnetica (EkŞİ et al., 2020; Ion-Mărgineanu et al., 2017) e le misure dell’atrofia basate su RM (Bonacchi et al. al., 2020), vari studi hanno studiato le potenzialità degli approcci basati sull’IA progettati per distinguere tra i diversi fenotipi clinici della SM.

Il trattamento della SM viene personalizzato sulla base di marcatori prognostici clinici, demografici, di laboratorio e di imaging (Rotstein & Montalban, 2019). Sono stati valutati vari algoritmi basati sull’IA per vedere se fossero capaci di prevedere il passaggio dal primo episodio clinico indicativo di malattia infiammatoria cronica del sistema nervoso centrale, la cosiddetta "sindrome clinicamente isolata", alla SM conclamata utilizzando le caratteristiche della RM, con una sensibilità del 64-77% e una specificità del 66-78% (Bendfeldt et al., 2019; Wottschel et al., 2015, 2019). Sono stati inoltre progettati algoritmi basati sull’IA che associano i dati clinici e della RM per prevedere il decorso della malattia e la disabilità clinica (Filippi et al., 2013; Roca et al., 2020; Tommasin et al., 2021; Zhao et al., 2017, 2020). Utilizzando macchine a vettori di supporto e alberi estremamente randomizzati, uno studio ha scoperto che una “impronta digitale” di imaging ad alta dimensionalità derivata da immagini pesate in T1 e FLAIR ha previsto meglio la risposta al trattamento nella SM rispetto alle misure della risposta al trattamento derivate dalla RM convenzionale, come il volume cerebrale e il numero e il volume delle lesioni (AUC 0,89 vs. 0,69) (Kanber et al., 2019).

Inoltre, gli algoritmi basati sull’IA hanno dimostrato di poter molto probabilmente essere d’aiuto nei protocolli di RM utilizzati nella SM, ad esempio nell’estrazione di informazioni da sequenze di RM convenzionali e nella generazione di sequenze sintetiche da immagini acquisite, ad esempio immagini contrastografiche da RM senza mezzo di contrasto (Bonacchi et al., 2022).

Neuro-oncologia

I tumori cerebrali comprendono sia i tumori primari del cervello (che possono essere benigni o maligni) sia i tumori metastatici presenti in altre parti del corpo. Distinguere tra tumori cerebrali e altre patologie mediante esami di imaging è importante per evitare biopsie non necessarie ed orientare la gestione (Abd-Ellah et al., 2019). Estraendo le caratteristiche da risonanze magnetiche pesate in T1 e in T2, un classificatore basato sulla regressione ha raggiunto AUC fino a 0,99 per distinguere tra gliomi senza aumento di segnale di contrasto e lesioni cerebrali infiammatorie (Y. Han et al., 2021). Sono stati progettati anche approcci basati sull’IA volti a prevedere i gradi istopatologici del glioma mediante RM, ottenendo un'accuratezza media dell'89% ± 0,09%, come è stato riportato in una revisione sistematica della letteratura (Bahar et al., 2022). Inoltre, un classificatore Random Forest che ha utilizzato le caratteristiche della RM multiparametrica ha ottenuto risultati migliori rispetto a due radiologi esperti nell’identificare il tumore primario in caso di metastasi cerebrale (Kniep et al., 2019).

Uno degli utilizzi più promettenti dell’IA nell’imaging neuro-oncologico sembra essere l’impiego nella RM per identificare le mutazioni genetiche dei tumori. Questo campo, chiamato radiogenomica, è molto importante perché tumori istopatologicamente simili ma con mutazioni diverse rispondono in modo diverso a specifiche strategie di trattamento (Singh et al., 2021). Nella radiogenomica si estraggono caratteristiche da più sequenze di RM (anatomiche e pesate in diffusione e perfusione) per poi utilizzarle per prevedere le alterazioni genomiche nel tumore. Sono stati studiati diversi approcci basati su alberi decisionali e algoritmi Random Forest per prevedere sia singole mutazioni genetiche sia insiemi più complessi di alterazioni genomiche nei tumori cerebrali (Akkus et al., 2017; P. Chang et al., 2018; L. Han & Kamdar, 2018; Hu et al., 2017; Kickingereder et al., 2016; Park et al., 2020). L’imaging svolge un ruolo importante anche nella valutazione della risposta del tumore cerebrale a trattamenti come la radioterapia, l’immunoterapia, la chemioterapia e l’intervento chirurgico, ma ciò richiede notevoli competenze, soprattutto perché le caratteristiche all’imaging della risposta e della recidiva sono simili a quelle di altre modifiche correlate al trattamento come la pseudoprogressione, che consiste in un aumento transitorio nell’impregnazione del mezzo di contrasto e/o dell’edema peritumorale che si verifica in seguito a radio e chemioterapia (Raimbault et al., 2014; Thint et al., 2018).

Nel glioblastoma, il tumore cerebrale primario maligno più comune negli adulti, la valutazione della risposta al trattamento prevede la misurazione manuale del volume di tessuto tumorale che assorbe il mezzo di contrasto (Leao et al., 2020). Utilizzando l’apprendimento automatico basato sul metodo dei vettori di supporto, algoritmi Random Forest e CNN (Havaei et al., 2017; Kickingereder et al., 2019; Menze et al., 2015) si è ottenuta una segmentazione automatizzata accurata del tessuto tumorale cerebrale con alcune applicazioni con certificazione CE disponibili (BioMind, senza data). Una meta-analisi ha rilevato che il deep learning supera gli approcci tradizionali di machine learning nella segmentazione del tumore (Kouli et al., 2022). In un ampio studio multicentrico su dati clinici, la volumetria automatizzata del tumore effettuata utilizzando CNN ha mostrato una probabile superiorità rispetto alla volumetria manuale nel calcolo del tempo alla progressione della malattia (Kickingereder et al., 2019).

Anche alcuni approcci di classificazione basata sulla regressione e con CNN hanno avuto dei discreti risultati nel distinguere tra pseudoprogressione e progressione effettiva dei tumori cerebrali alla RM (Jang et al., 2018, 2020; J. Y. Kim et al., 2019).

Le CNN si sono mostrate promettenti anche per la differenziazione tra necrosi da radiazioni e progressione del tumore, raggiungendo una sensibilità del 99,4% e una specificità del 97,5% (Q. Zhang et al., 2019). I margini dei tumori cerebrali e l’infiltrazione del tessuto circostante possono essere molto difficili da distinguere dall’edema peritumorale. I classificatori basati sulla regressione, i classificatori di apprendimento automatico basato sul metodo dei vettori di supporto e le CNN hanno prodotto mappe accurate dell'infiltrazione peritumorale che potrebbero rivelarsi utili per la programmazione dell’intervento chirurgico (Akbari et al., 2016; P. D. Chang, Chow, et al., 2017; P. D. Chang, Malone, et al., 2017).

Si è visto un certo interesse per gli approcci basati sull’IA volti a migliorare la RM e assistere i radiologi nella diagnosi dei tumori cerebrali. Alcuni obiettivi di questi approcci basati sull’IA sono stati gli algoritmi di ricostruzione delle immagini basati sulle CNN, che migliorano la risoluzione spaziale e riducono il rumore, consentendo di vedere strutture anatomiche più piccole e i componenti tumorali. Tali approcci hanno mostrato risultati nel rilevamento di microadenomi ipofisari, nell’identificazione di tumori residui o recidivanti dopo il trattamento e nella caratterizzazione dell’invasione tumorale (M. Kim et al., 2021; D. H. Lee et al., 2021). Inoltre, le immagini contrastografiche sintetizzate da immagini senza contrasto utilizzando reti generative avversarie (Generative Adversarial Networks, GAN) si sono rivelate utili per valutare la risposta al trattamento del glioblastoma e possono aiutare a ridurre l’uso di mezzi di contrasto nella RM (Jayachandran Preetha et al., 2021).

Conclusione

Nell’arco di circa un decennio, la ricerca sulle applicazioni di IA in neuroradiologia ha fatto notevoli progressi. L’IA è stata particolarmente utile nel supportare la diagnosi di patologie come l’ictus e l’emorragia intracranica, in cui il rilevamento tempestivo è fondamentale. Vi sono inoltre evidenze crescenti del fatto che l’IA possa essere utilizzata per monitorare la progressione delle malattie neurologiche, prevederne gli esiti e, in definitiva, consentire strategie di trattamento più personalizzate ed efficaci. In futuro si auspica dunque che la ricerca sugli algoritmi basati sull’IA integri anche il rapporto costo-efficacia di queste applicazioni e misuri l’effetto della loro implementazione sugli esiti complessivi dei pazienti. Inoltre, per incoraggiare l’uso di queste applicazioni occorrerebbe che fossero supportate da un numero maggiore di dati pubblicati sulle loro prestazioni. Nel complesso, l’uso dell’IA in neuroradiologia è molto promettente e potrebbe migliorare la qualità dell’assistenza sanitaria dei pazienti.

Bibliografia

Abdelkhaleq, R., Kim, Y., Khose, S., Kan, P., Salazar-Marioni, S., Giancardo, L., & Sheth, S. A. (2021). Automated prediction of final infarct volume in patients with large-vessel occlusion acute ischemic stroke. Neurosurgical Focus, 51(1), E13. https://doi.org/10.3171/2021.4.FOCUS21134

Abd-Ellah, M. K., Awad, A. I., Khalaf, A. A. M., & Hamed, H. F. A. (2019). A review on brain tumor diagnosis from MRI images: Practical implications, key achievements, and lessons learned. Magnetic Resonance Imaging, 61, 300–318. https://doi.org/10.1016/j.mri.2019.05.028

Adeli, E., Shi, F., An, L., Wee, C.-Y., Wu, G., Wang, T., & Shen, D. (2016). Joint feature-sample selection and robust diagnosis of Parkinson’s disease from MRI data. NeuroImage, 141, 206–219. https://doi.org/10.1016/j.neuroimage.2016.05.054

Adhya, J., Li, C., Eisenmenger, L., Cerejo, R., Tayal, A., Goldberg, M., & Chang, W. (2021). Positive predictive value and stroke workflow outcomes using automated vessel density (RAPID-CTA) in stroke patients: One year experience. The Neuroradiology Journal, 34(5), 476–481. https://doi.org/10.1177/19714009211012353

Adusumilli, G., Pederson, J. M., Hardy, N., Kallmes, K. M., Hutchison, K., Kobeissi, H., Heiferman, D. M., & Heit, J. J. (2022). Mechanical thrombectomy in anterior vs. posterior circulation stroke: A systematic review and meta-analysis. Interventional Neuroradiology: Journal of Peritherapeutic Neuroradiology, Surgical Procedures and Related Neurosciences, 15910199221100796. https://doi. org/10.1177/15910199221100796

Afzal, H. M. R., Luo, S., Ramadan, S., & Lechner-Scott, J. (2022). The emerging role of artificial intelligence in multiple sclerosis imaging. Multiple Sclerosis, 28(6), 849–858. https://doi.org/10.1177/1352458520966298

AI Central. (n.d.). Retrieved July 2, 2022, from https://aicentral.com/

AI for radiology(n.d.). Retrieved July 2, 2022, from https://grand-challenge.org/aiforradiology/

Akbari, H., Macyszyn, L., Da, X., Bilello, M., Wolf, R. L., Martinez-Lage, M., Biros, G., Alonso-Basanta, M., O’Rourke, D. M., & Davatzikos, C. (2016). Imaging Surrogates of Infiltration Obtained Via Multiparametric Imaging Pattern Analysis Predict Subsequent Location of Recurrence of Glioblastoma. Neurosurgery, 78(4), 572–580. https://doi.org/10.1227/NEU.0000000000001202

Akkus, Z., Ali, I., Sedlář, J., Agrawal, J. P., Parney, I. F., Giannini, C., & Erickson, B. J. (2017). Predicting Deletion of Chromosomal Arms 1p/19q in Low-Grade Gliomas from MR Images Using Machine Intelligence. Journal of Digital Imaging, 30(4), 469–476. https://doi.org/10.1007/s10278-017-9984-3

Amoroso, N., Diacono, D., Fanizzi, A., La Rocca, M., Monaco, A., Lombardi, A., Guaragnella, C., Bellotti, R., Tangaro, S., & Alzheimer’s Disease Neuroimaging Initiative. (2018). Deep learning reveals Alzheimer’s disease onset in MCI subjects: Results from an international challenge. Journal of Neuroscience Methods, 302, 3–9. https://doi.org/10.1016/j. jneumeth.2017.12.011

Amukotuwa, S. A., Straka, M., Smith, H., Chandra, R. V., Dehkharghani, S., Fischbein, N. J., & Bammer, R. (2019). Automated Detection of Intracranial Large Vessel Occlusions on Computed Tomography Angiography: A Single Center Experience. Stroke; a Journal of Cerebral Circulation, 50(10), 2790–2798. https://doi.org/10.1161/STROKEAHA.119.026259

Andravizou, A., Dardiotis, E., Artemiadis, A., Sokratous, M., Siokas, V., Tsouris, Z., Aloizou, A.-M., Nikolaidis, I., Bakirtzis, C., Tsivgoulis, G., Deretzi, G., Grigoriadis, N., Bogdanos, D. P., & Hadjigeorgiou, G. M. (2019). Brain atrophy in multiple sclerosis: mechanisms, clinical relevance and treatment options. Auto- Immunity Highlights, 10(1), 7. https://doi.org/10.1186/ s13317-019-0117-5

Bahar, R. C., Merkaj, S., Cassinelli Petersen, G. I., Tillmanns, N., Subramanian, H., Brim, W. R., Zeevi, T., Staib, L., Kazarian, E., Lin, M., Bousabarah, K., Huttner, A. J., Pala, A., Payabvash, S., Ivanidze, J., Cui, J., Malhotra, A., & Aboian, M. S. (2022). Machine Learning Models for Classifying Highand Low-Grade Gliomas: A Systematic Review and Quality of Reporting Analysis. Frontiers in Oncology, 12, 856231. https://doi.org/10.3389/fonc.2022.856231

Bar, A., Mauda, M., Turner, Y., Safadi, M., & Elnekave, E. (2019). Improved ICH classification using task-dependent learning. In arXiv [cs.CV]. arXiv. https://uploads-ssl.webflow. com/602a32732226833dce680ffe/61733c14eabf6d96471c5 2b4_7-Bar_bloodNet__Improved_ICH_classification_using_task_ dependent_learning__isbi2018.pdf

Battineni, G., Chintalapudi, N., Hossain, M. A., Losco, G., Ruocco, C., Sagaro, G. G., Traini, E., Nittari, G., & Amenta, F. (2022). Artificial Intelligence Models in the Diagnosis of Adult-Onset Dementia Disorders: A Review. Bioengineering (Basel, Switzerland), 9(8). https://doi.org/10.3390/ bioengineering9080370

Bendfeldt, K., Taschler, B., Gaetano, L., Madoerin, P., Kuster, P., Mueller-Lenke, N., Amann, M., Vrenken, H., Wottschel, V., Barkhof, F., Borgwardt, S., Klöppel, S., Wicklein, E.- M., Kappos, L., Edan, G., Freedman, M. S., Montalbán, X., Hartung, H.-P., Pohl, C., … Nichols, T. E. (2019). MRI-based prediction of conversion from clinically isolated syndrome to clinically definite multiple sclerosis using SVM and lesion geometry. Brain Imaging and Behavior, 13(5), 1361–1374. https://doi.org/10.1007/s11682-018-9942-9

BioMind. (n.d.). Retrieved December 20, 2022, from https://biomind.ai/paper

Birenbaum, A., & Greenspan, H. (2017). Multi-view longitudinal CNN for multiple sclerosis lesion segmentation. Engineering Applications of Artificial Intelligence, 65, 111–118. https://doi.org/10.1016/j.engappai.2017.06.006

Boldsen, J. K., Engedal, T. S., Pedraza, S., Cho, T.-H., Thomalla, G., Nighoghossian, N., Baron, J.-C., Fiehler, J., Østergaard, L., & Mouridsen, K. (2018). Better Diffusion Segmentation in Acute Ischemic Stroke Through Automatic Tree Learning Anomaly Segmentation. Frontiers in Neuroinformatics, 12, 21. https://doi.org/10.3389/fninf.2018.00021

Bonacchi, R., Filippi, M., & Rocca, M. A. (2022). Role of artificial intelligence in MS clinical practice. NeuroImage. Clinical, 35, 103065. https://doi.org/10.1016/j.nicl.2022.103065

Bonacchi, R., Pagani, E., Meani, A., Cacciaguerra, L., Preziosa, P., De Meo, E., Filippi, M., & Rocca, M. A. (2020). Clinical Relevance of Multiparametric MRI Assessment of Cervical Cord Damage in Multiple Sclerosis. Radiology, 296(3), 605–615. https://doi.org/10.1148/radiol.2020200430

Boutet, A., Madhavan, R., Elias, G. J. B., Joel, S. E., Gramer, R., Ranjan, M., Paramanandam, V., Xu, D., Germann, J., Loh, A., Kalia, S. K., Hodaie, M., Li, B., Prasad, S., Coblentz, A., Munhoz, R. P., Ashe, J., Kucharczyk, W., Fasano, A., & Lozano, A. M. (2021). Predicting optimal deep brain stimulation parameters for Parkinson’s disease using functional MRI and machine learning. Nature Communications, 12(1), 3043. https://doi.org/10.1038/s41467-021-23311-9

Broderick, J., Connolly, S., Feldmann, E., Hanley, D., Kase, C., Krieger, D., Mayberg, M., Morgenstern, L., Ogilvy, C. S., Vespa, P., Zuccarello, M., American Heart Association, American Stroke Association Stroke Council, High Blood Pressure Research Council, & Quality of Care and Outcomes in Research Interdisciplinary Working Group. (2007). Guidelines for the management of spontaneous intracerebral hemorrhage in adults: 2007 update: a guideline from the American Heart Association/American Stroke Association Stroke Council, High Blood Pressure Research Council, and the Quality of Care and Outcomes in Research Interdisciplinary Working Group. Stroke; a Journal of Cerebral Circulation, 38(6), 2001–2023. https://doi.org/10.1161/STROKEAHA.107.183689

Bron, E. E., Smits, M., Niessen, W. J., & Klein, S. (2015). Feature Selection Based on the SVM Weight Vector for Classification of Dementia. IEEE Journal of Biomedical and Health Informatics, 19(5), 1617–1626. https://doi.org/10.1109/ JBHI.2015.2432832

Brosch, T., Tang, L. Y. W., Youngjin Yoo, Li, D. K. B., Traboulsee, A., & Tam, R. (2016). Deep 3D Convolutional Encoder Networks With Shortcuts for Multiscale Feature Integration Applied to Multiple Sclerosis Lesion Segmentation. IEEE Transactions on Medical Imaging, 35(5), 1229–1239. https://doi.org/10.1109/TMI.2016.2528821

Cacciaguerra, L., Meani, A., Mesaros, S., Radaelli, M., Palace, J., Dujmovic-Basuroski, I., Pagani, E., Martinelli, V., Matthews, L., Drulovic, J., Leite, M. I., Comi, G., Filippi, M., & Rocca, M. A. (2019). Brain and cord imaging features in neuromyelitis optica spectrum disorders. Annals of Neurology, 85(3), 371–384. https://doi.org/10.1002/ana.25411

Cavedo, E., Tran, P., Thoprakarn, U., Martini, J.-B., Movschin, A., Delmaire, C., Gariel, F., Heidelberg, D., Pyatigorskaya, N., Ströer, S., Krolak-Salmon, P., Cotton, F., Dos Santos, C. L., & Dormont, D. (2022). Validation of an automatic tool for the rapid measurement of brain atrophy and white matter hyperintensity: QyScore^®. European Radiology, 32(5), 2949–2961. https://doi.org/10.1007/s00330-021-08385-9

Chakraborty, S., Aich, S., & Kim, H.-C. (2020). Detection of Parkinson’s Disease from 3T T1 Weighted MRI Scans Using 3D Convolutional Neural Network. Diagnostics, 10(6), 402. https://doi.org/10.3390/diagnostics10060402

Chalela, J. A., Kidwell, C. S., Nentwich, L. M., Luby, M., Butman, J. A., Demchuk, A. M., Hill, M. D., Patronas, N., Latour, L., & Warach, S. (2007). Magnetic resonance imaging and computed tomography in emergency assessment of patients with suspected acute stroke: a prospective comparison. The Lancet, 369(9558), 293–298. https://doi.org/10.1016/S0140- 6736(07)60151-2

Chang, P. D., Chow, D. S., Yang, P. H., Filippi, C. G., & Lignelli, A. (2017). Predicting Glioblastoma Recurrence by Early Changes in the Apparent Diffusion Coefficient Value and Signal Intensity on FLAIR Images. AJR. American Journal of Roentgenology, 208(1), 57–65. https://doi.org/10.2214/AJR.16.16234

Chang, P. D., Malone, H. R., Bowden, S. G., Chow, D. S., Gill, B. J. A., Ung, T. H., Samanamud, J., Englander, Z. K., Sonabend, A. M., Sheth, S. A., McKhann, G. M., 2nd, Sisti, M. B., Schwartz, L. H., Lignelli, A., Grinband, J., Bruce, J. N., & Canoll, P. (2017). A Multiparametric Model for Mapping Cellularity in Glioblastoma Using Radiographically Localized Biopsies. AJNR. American Journal of Neuroradiology, 38(5), 890–898. https://doi.org/10.3174/ajnr.A5112

Chang, P., Grinband, J., Weinberg, B. D., Bardis, M., Khy, M., Cadena, G., Su, M.-Y., Cha, S., Filippi, C. G., Bota, D., Baldi, P., Poisson, L. M., Jain, R., & Chow, D. (2018). Deep-Learning Convolutional Neural Networks Accurately Classify Genetic Mutations in Gliomas. AJNR. American Journal of Neuroradiology, 39(7), 1201–1207. https://doi.org/10.3174/ajnr.A5667

Cherubini, A., Morelli, M., Nisticó, R., Salsone, M., Arabia, G., Vasta, R., Augimeri, A., Caligiuri, M. E., & Quattrone, A. (2014). Magnetic resonance support vector machine discriminates between Parkinson disease and progressive supranuclear palsy. Movement Disorders: Official Journal of the Movement Disorder Society, 29(2), 266–269. https://doi.org/10.1002/mds.25737

Chien, H.-W. C., Yang, T.-L., Juang, W.-C., Chen, Y.-Y. A., Li, Y.-C. J., & Chen, C.-Y. (2022). Pilot Report for Intracranial Hemorrhage Detection with Deep Learning Implanted Head Computed Tomography Images at Emergency Department. Journal of Medical Systems, 46(7), 49. https://doi.org/10.1007/ s10916-022-01833-z

Chilamkurthy, S., Ghosh, R., Tanamala, S., Biviji, M., Campeau, N. G., Venugopal, V. K., Mahajan, V., Rao, P., & Warier, P. (2018). Deep learning algorithms for detection of critical findings in head CT scans: a retrospective study. The Lancet, 392(10162), 2388–2396. https://doi.org/10.1016/S0140- 6736(18)31645-3

Choi, H., Ha, S., Im, H. J., Paek, S. H., & Lee, D. S. (2017). Refining diagnosis of Parkinson’s disease with deep learningbased interpretation of dopamine transporter imaging. NeuroImage. Clinical, 16, 586–594. https://doi.org/10.1016/j. nicl.2017.09.010

Connor, S. E. J., Tan, G., Fernando, R., & Chaudhury, N. (2005). Computed tomography pseudofractures of the mid face and skull base. Clinical Radiology, 60(12), 1268–1279. https://doi. org/10.1016/j.crad.2005.05.016

Crous-Bou, M., Minguillón, C., Gramunt, N., & Molinuevo, J. L. (2017). Alzheimer’s disease prevention: from risk factors to early intervention. Alzheimer’s Research & Therapy, 9(1), 1–9. https://doi.org/10.1186/s13195-017-0297-z

Danelakis, A., Theoharis, T., & Verganelakis, D. A. (2018). Survey of automated multiple sclerosis lesion segmentation techniques on magnetic resonance imaging. Computerized Medical Imaging and Graphics: The Official Journal of the Computerized Medical Imaging Society, 70, 83–100. https://doi.org/10.1016/j.compmedimag.2018.10.002

Davatzikos, C., Resnick, S. M., Wu, X., Parmpi, P., & Clark, C. M. (2008). Individual patient diagnosis of AD and FTD via highdimensional pattern classification of MRI. NeuroImage, 41(4), 1220–1227. https://doi.org/10.1016/j.neuroimage.2008.03.050

Dehkharghani, S., Lansberg, M., Venkatsubramanian, C., Cereda, C., Lima, F., Coelho, H., Rocha, F., Qureshi, A., Haerian, H., Mont’Alverne, F., Copeland, K., & Heit, J. (2021). High-Performance Automated Anterior Circulation CT Angiographic Clot Detection in Acute Stroke: A Multireader Comparison. Radiology, 298(3), 665–670. https://doi.org/10.1148/radiol.2021202734

Deshpande, H., Maurel, P., & Barillot, C. (2015). Classification of multiple sclerosis lesions using adaptive dictionary learning. Computerized Medical Imaging and Graphics: The Official Journal of the Computerized Medical Imaging Society, 46 Pt 1 , 2–10. https://doi.org/10.1016/j.compmedimag.2015.05.003

Dolz, J., Desrosiers, C., & Ben Ayed, I. (2018). 3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study. NeuroImage, 170 , 456–470. https://doi.org/10.1016/j.neuroimage.2017.04.039

Duchesne, S., Rolland, Y., & Vérin, M. (2009). Automated computer differential classification in Parkinsonian Syndromes via pattern analysis on MRI. Academic Radiology, 16(1), 61–70. https://doi.org/10.1016/j.acra.2008.05.024

Duvekot, M. H. C., van Es, A. C. G. M., Venema, E., Wolff, L., Rozeman, A. D., Moudrous, W., Vermeij, F. H., Lingsma, H. F., Bakker, J., Plaisier, A. S., Hensen, J.-H. J., Lycklama à Nijeholt, G. J., Jan van Doormaal, P., Dippel, D. W. J., Kerkhoff, H., Roozenbeek, B., & van der Lugt, A. (2021). Accuracy of CTA evaluations in daily clinical practice for large and medium vessel occlusion detection in suspected stroke patients. European Stroke Journal, 23969873211058576. https:// doi.org/10.1177/23969873211058576

Ebinger, M., Galinovic, I., Rozanski, M., Brunecker, P., Endres, M., & Fiebach, J. B. (2010). Fluid-attenuated inversion recovery evolution within 12 hours from stroke onset: a reliable tissue clock? Stroke; a Journal of Cerebral Circulation, 41(2), 250–255. https://doi.org/10.1161/STROKEAHA.109.568410

Eichinger, P., Zimmer, C., & Wiestler, B. (2020). AI in Radiology: Where are we today in Multiple Sclerosis Imaging? RoFo: Fortschritte Auf Dem Gebiete Der Rontgenstrahlen Und Der Nuklearmedizin, 192(9), 847–853. https://doi.org/10.1055/a-1167-8402

Eitel, F., Soehler, E., Bellmann-Strobl, J., Brandt, A. U., Ruprecht, K., Giess, R. M., Kuchling, J., Asseyer, S., Weygandt, M., Haynes, J.-D., Scheel, M., Paul, F., & Ritter, K. (2019). Uncovering convolutional neural network decisions for diagnosing multiple sclerosis on conventional MRI using layerwise relevance propagation. NeuroImage. Clinical, 24, 102003. https://doi.org/10.1016/j.nicl.2019.102003

EkŞİ, Z., ÇakiroĞlu, M., Öz, C., AralaŞmak, A., Karadelİ, H. H., & Özcan, M. E. (2020). Differentiation of relapsing-remitting and secondary progressive multiple sclerosis: a magnetic resonance spectroscopy study based on machine learning. Arquivos de Neuro-Psiquiatria, 78(12), 789–796. https://doi. org/10.1590/0004-282X20200094

Ekşi, Z., Özcan, M. E., Çakıroğlu, M., Öz, C., & Aralaşmak, A. (2021). Differentiation of multiple sclerosis lesions and lowgrade brain tumors on MRS data: machine learning approaches. Neurological Sciences: Official Journal of the Italian Neurological Society and of the Italian Society of Clinical Neurophysiology, 42(8), 3389–3395. https://doi.org/10.1007/s10072-020-04950-0

Eldaya, R. W., Kansagra, A. P., Zei, M., Mason, E., Holder, D., Heitsch, L., Vo, K. D., & Goyal, M. S. (2022). Performance of Automated RAPID Intracranial Hemorrhage Detection in Real-World Practice: A Single-Institution Experience. Journal of Computer Assisted Tomography, 46(5), 770–774. https://doi.org/10.1097/RCT.0000000000001335

Emon, M. M., Ornob, T. R., & Rahman, M. (2022). Classifications of Skull Fractures using CT Scan Images via CNN with Lazy Learning Approach. In arXiv [eess.IV]. arXiv. http://arxiv.org/abs/2203.10786

Erly, W. K., Berger, W. G., Krupinski, E., Seeger, J. F., & Guisto, J. A. (2002). Radiology resident evaluation of head CT scan orders in the emergency department. AJNR. American Journal of Neuroradiology, 23 (1), 103–107. https://www.ncbi.nlm.nih.gov/ pubmed/11827881

Eshaghi, A., Wottschel, V., Cortese, R., Calabrese, M., Sahraian, M. A., Thompson, A. J., Alexander, D. C., & Ciccarelli, O. (2016). Gray matter MRI differentiates neuromyelitis optica from multiple sclerosis using random forest. Neurology, 87(23), 2463–2470. https://doi.org/10.1212/ WNL.0000000000003395

Feigin, V. L., Vos, T., Nichols, E., Owolabi, M. O., Carroll, W. M., Dichgans, M., Deuschl, G., Parmar, P., Brainin, M., & Murray, C. (2020). The global burden of neurological disorders: translating evidence into policy. Lancet Neurology, 19(3), 255–265. https://doi.org/10.1016/S1474-4422(19)30411-9

Ferreti, L. A., Leitao, C. A., Teixeira, B. C. de A., Lopes Neto, F. D. N., ZÉtola, V. F., & Lange, M. C. (2020). The use of e-ASPECTS in acute stroke care: validation of method performance compared to the performance of specialists. Arquivos de Neuro- Psiquiatria, 78(12), 757–761. https://doi.org/10.1590/0004- 282X20200072

Filippi, M., Preziosa, P., Copetti, M., Riccitelli, G., Horsfield, M. A., Martinelli, V., Comi, G., & Rocca, M. A. (2013). Gray matter damage predicts the accumulation of disability 13 years later in MS. Neurology, 81(20), 1759–1767. https://doi.org/10.1212/01.wnl.0000435551.90824.d0

Flacke, S., Urbach, H., Keller, E., Träber, F., Hartmann, A., Textor, J., Gieseke, J., Block, W., Folkers, P. J., & Schild, H. H. (2000). Middle cerebral artery (MCA) susceptibility sign at susceptibility-based perfusion MR imaging: clinical importance and comparison with hyperdense MCA sign at CT. Radiology, 215(2), 476–482. https://doi.org/10.1148/ radiology.215.2.r00ma09476

Focke, N. K., Helms, G., Scheewe, S., Pantel, P. M., Bachmann, C. G., Dechent, P., Ebentheuer, J., Mohr, A., Paulus, W., & Trenkwalder, C. (2011). Individual voxel-based subtype prediction can differentiate progressive supranuclear palsy from idiopathic Parkinson syndrome and healthy controls. Human Brain Mapping, 32(11), 1905–1915. https://doi.org/10.1002/hbm.21161

Fugate, J. E., & Rabinstein, A. A. (2015). Absolute and Relative Contraindications to IV rt-PA for Acute Ischemic Stroke. The Neurohospitalist, 5(3), 110–121. https://doi. org/10.1177/1941874415578532

Gács, G., Fox, A. J., Barnett, H. J., & Vinuela, F. (1983). CT visualization of intracranial arterial thromboembolism. Stroke; a Journal of Cerebral Circulation, 14(5), 756–762. https://doi. org/10.1161/01.str.14.5.756

GBD 2016 Parkinson’s Disease Collaborators. (2018). Global, regional, and national burden of Parkinson’s disease, 1990- 2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Neurology, 17(11), 939–953. https://doi. org/10.1016/S1474-4422(18)30295-3

Gibson, E., Georgescu, B., Ceccaldi, P., Trigan, P.-H., Yoo, Y., Das, J., Re, T. J., Rs, V., Balachandran, A., Eibenberger, E., Chekkoury, A., Brehm, B., Bodanapally, U. K., Nicolaou, S., Sanelli, P. C., Schroeppel, T. J., Flohr, T., Comaniciu, D., & Lui, Y. W. (2022). Artificial Intelligence with Statistical Confidence Scores for Detection of Acute or Subacute Hemorrhage on Noncontrast CT Head Scans. Radiology. Artificial Intelligence, 4(3), e210115. https://doi.org/10.1148/ryai.210115

Ginat, D. (2021). Implementation of Machine Learning Software on the Radiology Worklist Decreases Scan View Delay for the Detection of Intracranial Hemorrhage on CT. Brain Sciences, 11(7). https://doi.org/10.3390/brainsci11070832

Goebel, J., Stenzel, E., Guberina, N., Wanke, I., Koehrmann, M., Kleinschnitz, C., Umutlu, L., Forsting, M., Moenninghoff, C., & Radbruch, A. (2018). Automated ASPECT rating: comparison between the Frontier ASPECT Score software and the Brainomix software. Neuroradiology, 60(12), 1267–1272. https://doi.org/10.1007/s00234-018-2098-x

Goldberg-Zimring, D., Achiron, A., Miron, S., Faibel, M., & Azhari, H. (1998). Automated detection and characterization of multiple sclerosis lesions in brain MR images. Magnetic Resonance Imaging, 16(3), 311–318. https://doi.org/10.1016/ s0730-725x(97)00300-7

Guberina, N., Dietrich, U., Radbruch, A., Goebel, J., Deuschl, C., Ringelstein, A., Köhrmann, M., Kleinschnitz, C., Forsting, M., & Mönninghoff, C. (2018). Detection of early infarction signs with machine learning-based diagnosis by means of the Alberta Stroke Program Early CT score (ASPECTS) in the clinical routine. Neuroradiology, 60(9), 889–901. https://doi.org/10.1007/ s00234-018-2066-5

Hakim, A., Christensen, S., Winzeck, S., Lansberg, M. G., Parsons, M. W., Lucas, C., Robben, D., Wiest, R., Reyes, M., & Zaharchuk, G. (2021). Predicting Infarct Core From Computed Tomography Perfusion in Acute Ischemia With Machine Learning: Lessons From the ISLES Challenge. Stroke; a Journal of Cerebral Circulation, 52(7), 2328–2337. https://doi.org/10.1161/STROKEAHA.120.030696

Haller, S., Badoud, S., Nguyen, D., Barnaure, I., Montandon, M.-L., Lovblad, K.-O., & Burkhard, P. R. (2013). Differentiation between Parkinson disease and other forms of Parkinsonism using support vector machine analysis of susceptibilityweighted imaging (SWI): initial results. European Radiology, 23(1), 12–19. https://doi.org/10.1007/s00330-012-2579-y

Haller, S., Badoud, S., Nguyen, D., Garibotto, V., Lovblad, K. O., & Burkhard, P. R. (2012). Individual detection of patients with Parkinson disease using support vector machine analysis of diffusion tensor imaging data: initial results. AJNR. American Journal of Neuroradiology, 33(11), 2123–2128. https://doi.org/10.3174/ajnr.A3126

Han, L., & Kamdar, M. R. (2018). MRI to MGMT: predicting methylation status in glioblastoma patients using convolutional recurrent neural networks. Pacific Symposium on Biocomputing. Pacific Symposium on Biocomputing, 23, 331–342. https://www.ncbi.nlm.nih.gov/pubmed/29218894

Han, Y., Yang, Y., Shi, Z.-S., Zhang, A.-D., Yan, L.-F., Hu, Y.-C., Feng, L.-L., Ma, J., Wang, W., & Cui, G.-B. (2021). Distinguishing brain inflammation from grade II glioma in population without contrast enhancement: a radiomics analysis based on conventional MRI. European Journal of Radiology, 134, 109467. https://doi.org/10.1016/j.ejrad.2020.109467

Hassan, A. E., Ringheanu, V. M., Rabah, R. R., Preston, L., Tekle, W. G., & Qureshi, A. I. (2020). Early experience utilizing artificial intelligence shows significant reduction in transfer times and length of stay in a hub and spoke model. Interventional Neuroradiology: Journal of Peritherapeutic Neuroradiology, Surgical Procedures and Related Neurosciences, 26(5), 615–622. https://doi.org/10.1177/1591019920953055

Havaei, M., Davy, A., Warde-Farley, D., Biard, A., Courville, A., Bengio, Y., Pal, C., Jodoin, P.-M., & Larochelle, H. (2017). Brain tumor segmentation with Deep Neural Networks. Medical Image Analysis, 35, 18–31. https://doi.org/10.1016/j.media.2016.05.004

Heimer, J., Thali, M. J., & Ebert, L. (2018). Classification based on the presence of skull fractures on curved maximum intensity skull projections by means of deep learning. Journal of Forensic Radiology and Imaging, 14, 16–20. https://doi.org/10.1016/j. jofri.2018.08.001

Heit, J. J., Coelho, H., Lima, F. O., Granja, M., Aghaebrahim, A., Hanel, R., Kwok, K., Haerian, H., Cereda, C. W., Venkatasubramanian, C., Dehkharghani, S., Carbonera, L. A., Wiener, J., Copeland, K., & Mont’Alverne, F. (2021). Automated Cerebral Hemorrhage Detection Using RAPID. AJNR. American Journal of Neuroradiology, 42(2), 273–278. https://doi.org/10.3174/ajnr.A6926

Herweh, C., Ringleb, P. A., Rauch, G., Gerry, S., Behrens, L., Möhlenbruch, M., Gottorf, R., Richter, D., Schieber, S., & Nagel, S. (2016). Performance of e-ASPECTS software in comparison to that of stroke physicians on assessing CT scans of acute ischemic stroke patients. International Journal of Stroke: Official Journal of the International Stroke Society, 11(4), 438–445. https://doi.org/10.1177/1747493016632244

Hirschauer, T. J., Adeli, H., & Buford, J. A. (2015). Computer- Aided Diagnosis of Parkinson’s Disease Using Enhanced Probabilistic Neural Network. Journal of Medical Systems, 39(11), 179. https://doi.org/10.1007/s10916-015-0353-9

Hitziger, S., Ling, W. X., Fritz, T., D’Albis, T., Lemke, A., & Grilo, J. (2022). Triplanar U-Net with lesion-wise voting for the segmentation of new lesions on longitudinal MRI studies. Frontiers in Neuroscience, 16, 964250. https://doi.org/10.3389/ fnins.2022.964250

Hoelter, P., Muehlen, I., Goelitz, P., Beuscher, V., Schwab, S., & Doerfler, A. (2020). Automated ASPECT scoring in acute ischemic stroke: comparison of three software tools. Neuroradiology, 62(10), 1231–1238. https://doi.org/10.1007/ s00234-020-02439-3

Hu, L. S., Ning, S., Eschbacher, J. M., Baxter, L. C., Gaw, N., Ranjbar, S., Plasencia, J., Dueck, A. C., Peng, S., Smith, K. A., Nakaji, P., Karis, J. P., Quarles, C. C., Wu, T., Loftus, J. C., Jenkins, R. B., Sicotte, H., Kollmeyer, T. M., O’Neill, B. P., … Mitchell, J. R. (2017). Radiogenomics to characterize regional genetic heterogeneity in glioblastoma. Neuro-Oncology, 19(1), 128–137. https://doi.org/10.1093/neuonc/now135

Huppertz, H.-J., Möller, L., Südmeyer, M., Hilker, R., Hattingen, E., Egger, K., Amtage, F., Respondek, G., Stamelou, M., Schnitzler, A., Pinkhardt, E. H., Oertel, W. H., Knake, S., Kassubek, J., & Höglinger, G. U. (2016). Differentiation of neurodegenerative parkinsonian syndromes by volumetric magnetic resonance imaging analysis and support vector machine classification. Movement Disorders: Official Journal of the Movement Disorder Society, 31(10), 1506– 1517. https://doi.org/10.1002/mds.26715

Hwang, D. Y., Silva, G. S., Furie, K. L., & Greer, D. M. (2012). Comparative sensitivity of computed tomography vs. magnetic resonance imaging for detecting acute posterior fossa infarct. The Journal of Emergency Medicine, 42(5), 559–565. https://doi. org/10.1016/j.jemermed.2011.05.101

Ion-Mărgineanu, A., Kocevar, G., Stamile, C., Sima, D. M., Durand-Dubief, F., Van Huffel, S., & Sappey-Marinier, D. (2017). Machine Learning Approach for Classifying Multiple Sclerosis Courses by Combining Clinical Data with Lesion Loads and Magnetic Resonance Metabolic Features. Frontiers in Neuroscience, 11, 398. https://doi.org/10.3389/fnins.2017.00398

Jain, A., Malhotra, A., & Payabvash, S. (2021). Imaging of Spontaneous Intracerebral Hemorrhage. Neuroimaging Clinics of North America, 31(2), 193–203. https://doi.org/10.1016/j. nic.2021.02.003

Jain, S., Ribbens, A., Sima, D. M., Cambron, M., De Keyser, J., Wang, C., Barnett, M. H., Van Huffel, S., Maes, F., & Smeets, D. (2016). Two Time Point MS Lesion Segmentation in Brain MRI: An Expectation-Maximization Framework. Frontiers in Neuroscience, 10, 576. https://doi.org/10.3389/fnins.2016.00576

Jang, B.-S., Jeon, S. H., Kim, I. H., & Kim, I. A. (2018). Prediction of Pseudoprogression versus Progression using Machine Learning Algorithm in Glioblastoma. Scientific Reports, 8(1), 12516. https://doi.org/10.1038/s41598-018-31007-2

Jang, B.-S., Park, A. J., Jeon, S. H., Kim, I. H., Lim, D. H., Park, S.-H., Lee, J. H., Chang, J. H., Cho, K. H., Kim, J. H., Sunwoo, L., Choi, S. H., & Kim, I. A. (2020). Machine Learning Model to Predict Pseudoprogression Versus Progression in Glioblastoma Using MRI: A Multi-Institutional Study (KROG 18-07). Cancers, 12(9). https://doi.org/10.3390/cancers12092706

Jayachandran Preetha, C., Meredig, H., Brugnara, G., Mahmutoglu, M. A., Foltyn, M., Isensee, F., Kessler, T., Pflüger, I., Schell, M., Neuberger, U., Petersen, J., Wick, A., Heiland, S., Debus, J., Platten, M., Idbaih, A., Brandes, A. A., Winkler, F., van den Bent, M. J., … Vollmuth, P. (2021). Deeplearning- based synthesis of post-contrast T1-weighted MRI for tumour response assessment in neuro-oncology: a multicentre, retrospective cohort study. The Lancet. Digital Health, 3(12), e784–e794. https://doi.org/10.1016/S2589-7500(21)00205-3

Jiang, L., Wang, S., Ai, Z., Shen, T., Zhang, H., Duan, S., Chen, Y.-C., Yin, X., & Sun, J. (2022). Development and external validation of a stability machine learning model to identify wake-up stroke onset time from MRI. European Radiology, 32(6), 3661–3669. https://doi.org/10.1007/s00330-021-08493-6

Kanber, B., Nachev, P., Barkhof, F., Calvi, A., Cardoso, J., Cortese, R., Prados, F., Sudre, C. H., Tur, C., Ourselin, S., & Ciccarelli, O. (2019). High-dimensional detection of imaging response to treatment in multiple sclerosis. NPJ Digital Medicine, 2, 49. https://doi.org/10.1038/s41746-019-0127-8

Karimian, A., & Jafari, S. (2015). A New Method to Segment the Multiple Sclerosis Lesions on Brain Magnetic Resonance Images. Journal of Medical Signals and Sensors, 5(4), 238–244. https:// www.ncbi.nlm.nih.gov/pubmed/26955567

Kickingereder, P., Bonekamp, D., Nowosielski, M., Kratz, A., Sill, M., Burth, S., Wick, A., Eidel, O., Schlemmer, H.-P., Radbruch, A., Debus, J., Herold-Mende, C., Unterberg, A., Jones, D., Pfister, S., Wick, W., von Deimling, A., Bendszus, M., & Capper, D. (2016). Radiogenomics of Glioblastoma: Machine Learning-based Classification of Molecular Characteristics by Using Multiparametric and Multiregional MR Imaging Features. Radiology, 281(3), 907–918. https://doi. org/10.1148/radiol.2016161382

Kickingereder, P., Isensee, F., Tursunova, I., Petersen, J., Neuberger, U., Bonekamp, D., Brugnara, G., Schell, M., Kessler, T., Foltyn, M., Harting, I., Sahm, F., Prager, M., Nowosielski, M., Wick, A., Nolden, M., Radbruch, A., Debus, J., Schlemmer, H.-P., … Maier-Hein, K. H. (2019). Automated quantitative tumour response assessment of MRI in neurooncology with artificial neural networks: a multicentre, retrospective study. The Lancet Oncology, 20(5), 728–740. https://doi.org/10.1016/S1470-2045(19)30098-1

Kim, B. J., Kim, Y.-H., Kim, Y.-J., Ahn, S. H., Lee, D. H., Kwon, S. U., Kim, S. J., Kim, J. S., & Kang, D.-W. (2014). Color-coded fluid-attenuated inversion recovery images improve inter-rater reliability of fluid-attenuated inversion recovery signal changes within acute diffusion-weighted image lesions. Stroke; a Journal of Cerebral Circulation, 45(9), 2801–2804. https://doi.org/10.1161/STROKEAHA.114.006515

Kim, J. Y., Park, J. E., Jo, Y., Shim, W. H., Nam, S. J., Kim, J. H., Yoo, R.-E., Choi, S. H., & Kim, H. S. (2019). Incorporating diffusion- and perfusion-weighted MRI into a radiomics model improves diagnostic performance for pseudoprogression in glioblastoma patients. Neuro-Oncology, 21(3), 404–414. https://doi.org/10.1093/neuonc/noy133

Kim, M., Kim, H. S., Kim, H. J., Park, J. E., Park, S. Y., Kim, Y.- H., Kim, S. J., Lee, J., & Lebel, M. R. (2021). Thin-Slice Pituitary MRI with Deep Learning-based Reconstruction: Diagnostic Performance in a Postoperative Setting. Radiology, 298(1), 114–122. https://doi.org/10.1148/radiol.2020200723

Kniep, H. C., Madesta, F., Schneider, T., Hanning, U., Schönfeld, M. H., Schön, G., Fiehler, J., Gauer, T., Werner, R., & Gellissen, S. (2019). Radiomics of Brain MRI: Utility in Prediction of Metastatic Tumor Type. Radiology, 290(2), 479–487. https://doi.org/10.1148/radiol.2018180946

Kniep, H. C., Sporns, P. B., Broocks, G., Kemmling, A., Nawabi, J., Rusche, T., Fiehler, J., & Hanning, U. (2020). Posterior circulation stroke: machine learning-based detection of early ischemic changes in acute non-contrast CT scans. Journal of Neurology, 267(9), 2632–2641. https://doi. org/10.1007/s00415-020-09859-4

Knopman, D. S., Amieva, H., Petersen, R. C., Chételat, G., Holtzman, D. M., Hyman, B. T., Nixon, R. A., & Jones, D. T. (2021). Alzheimer disease. Nature Reviews. Disease Primers, 7(1), 33. https://doi.org/10.1038/s41572-021-00269-y

Kocevar, G., Stamile, C., Hannoun, S., Cotton, F., Vukusic, S., Durand-Dubief, F., & Sappey-Marinier, D. (2016). Graph Theory-Based Brain Connectivity for Automatic Classification of Multiple Sclerosis Clinical Courses. Frontiers in Neuroscience, 10, 478. https://doi.org/10.3389/fnins.2016.00478

Kouli, O., Hassane, A., Badran, D., Kouli, T., Hossain-Ibrahim, K., & Steele, J. D. (2022). Automated brain tumor identification using magnetic resonance imaging: A systematic review and meta-analysis. Neuro-Oncology Advances, 4(1), vdac081. https:// doi.org/10.1093/noajnl/vdac081

Kuang, H., Najm, M., Chakraborty, D., Maraj, N., Sohn, S. I., Goyal, M., Hill, M. D., Demchuk, A. M., Menon, B. K., & Qiu, W. (2019). Automated ASPECTS on Noncontrast CT Scans in Patients with Acute Ischemic Stroke Using Machine Learning. AJNR. American Journal of Neuroradiology, 40(1), 33–38. https://doi.org/10.3174/ajnr.A5889

Kuang, Z., Deng, X., Yu, L., Zhang, H., Lin, X., & Ma, H. (2020). Skull R-CNN: A CNN-based network for the skull fracture detection. In T. Arbel, I. Ben Ayed, M. de Bruijne, M. Descoteaux, H. Lombaert, & C. Pal (Eds.), Proceedings of the Third Conference on Medical Imaging with Deep Learning (Vol. 121, pp. 382–392). PMLR. https://proceedings.mlr.press/v121/kuang20a.html

Kushibar, K., Valverde, S., González-Villà, S., Bernal, J., Cabezas, M., Oliver, A., & Lladó, X. (2018). Automated sub-cortical brain structure segmentation combining spatial and deep convolutional features. Medical Image Analysis, 48, 177–186. https://doi.org/10.1016/j.media.2018.06.006

Lamptey, R. N. L., Chaulagain, B., Trivedi, R., Gothwal, A., Layek, B., & Singh, J. (2022). A Review of the Common Neurodegenerative Disorders: Current Therapeutic Approaches and the Potential Role of Nanotherapeutics. International Journal of Molecular Sciences, 23(3). https://doi.org/10.3390/ ijms23031851

Leao, D. J., Craig, P. G., Godoy, L. F., Leite, C. C., & Policeni, B. (2020). Response Assessment in Neuro-Oncology Criteria for Gliomas: Practical Approach Using Conventional and Advanced Techniques. AJNR. American Journal of Neuroradiology, 41(1), 10–20. https://doi.org/10.3174/ajnr.A6358

Lebedev, A. V., Westman, E., Van Westen, G. J. P., Kramberger, M. G., Lundervold, A., Aarsland, D., Soininen, H., Kłoszewska, I., Mecocci, P., Tsolaki, M., Vellas, B., Lovestone, S., Simmons, A., & Alzheimer’s Disease Neuroimaging Initiative and the AddNeuroMed consortium. (2014). Random Forest ensembles for detection and prediction of Alzheimer’s disease with a good between-cohort robustness. NeuroImage. Clinical, 6, 115–125. https://doi.org/10.1016/j. nicl.2014.08.023

Lee, D. H., Park, J. E., Nam, Y. K., Lee, J., Kim, S., Kim, Y.-H., & Kim, H. S. (2021). Deep learning-based thin-section MRI reconstruction improves tumour detection and delineation in pre- and post-treatment pituitary adenoma. Scientific Reports, 11(1), 21302. https://doi.org/10.1038/s41598-021-00558-2

Lee, G., Nho, K., Kang, B., Sohn, K.-A., Kim, D., & for Alzheimer’s Disease Neuroimaging Initiative. (2019). Predicting Alzheimer’s disease progression using multi-modal deep learning approach. Scientific Reports, 9(1), 1952. https://doi.org/10.1038/s41598-018-37769-z

Lee, H., Lee, E.-J., Ham, S., Lee, H.-B., Lee, J. S., Kwon, S. U., Kim, J. S., Kim, N., & Kang, D.-W. (2020). Machine Learning Approach to Identify Stroke Within 4.5 Hours. Stroke; a Journal of Cerebral Circulation, 51(3), 860–866. https://doi.org/10.1161/ STROKEAHA.119.027611

Lisowska, A., O’Neil, A., Dilys, V., Daykin, M., Beveridge, E., Muir, K., Mclaughlin, S., & Poole, I. (2017). Context-Aware Convolutional Neural Networks for Stroke Sign Detection in Noncontrast CT Scans. Medical Image Understanding and Analysis, 494–505. https://doi.org/10.1007/978-3-319-60964-5_43

Long, D., Wang, J., Xuan, M., Gu, Q., Xu, X., Kong, D., & Zhang, M. (2012). Automatic classification of early Parkinson’s disease with multi-modal MR imaging. PloS One, 7(11), e47714. https://doi.org/10.1371/journal.pone.0047714

Lopatina, A., Ropele, S., Sibgatulin, R., Reichenbach, J. R., & Güllmar, D. (2020). Investigation of Deep-Learning- Driven Identification of Multiple Sclerosis Patients Based on Susceptibility-Weighted Images Using Relevance Analysis. Frontiers in Neuroscience, 14, 609468. https://doi.org/10.3389/ fnins.2020.609468

Lublin, F. D., Reingold, S. C., Cohen, J. A., Cutter, G. R., Sørensen, P. S., Thompson, A. J., Wolinsky, J. S., Balcer, L. J., Banwell, B., Barkhof, F., Bebo, B., Jr, Calabresi, P. A., Clanet, M., Comi, G., Fox, R. J., Freedman, M. S., Goodman, A. D., Inglese, M., Kappos, L., … Polman, C. H. (2014). Defining the clinical course of multiple sclerosis: the 2013 revisions. Neurology, 83(3), 278–286. https://doi.org/10.1212/ WNL.0000000000000560

Lu, D., Popuri, K., Ding, G. W., Balachandar, R., Beg, M. F., & Alzheimer’s Disease Neuroimaging Initiative. (2018). Multimodal and Multiscale Deep Neural Networks for the Early Diagnosis of Alzheimer’s Disease using structural MR and FDG-PET images. Scientific Reports, 8(1), 5697. https://doi. org/10.1038/s41598-018-22871-z

Mackey J, Kleindorfer D, Sucharew H, et al. Population-based study of wake-up strokes. Neurology. 2011;76(19):1662-1667. doi:10.1212/WNL.0b013e318219fb30

Maegerlein, C., Fischer, J., Mönch, S., Berndt, M., Wunderlich, S., Seifert, C. L., Lehm, M., Boeckh-Behrens, T., Zimmer, C., & Friedrich, B. (2019). Automated Calculation of the Alberta Stroke Program Early CT Score: Feasibility and Reliability. Radiology, 291(1), 141–148. https://doi.org/10.1148/ radiol.2019181228

Mangeat, G., Ouellette, R., Wabartha, M., De Leener, B., Plattén, M., Danylaité Karrenbauer, V., Warntjes, M., Stikov, N., Mainero, C., Cohen-Adad, J., & Granberg, T. (2020). Machine Learning and Multiparametric Brain MRI to Differentiate Hereditary Diffuse Leukodystrophy with Spheroids from Multiple Sclerosis. Journal of Neuroimaging: Official Journal of the American Society of Neuroimaging, 30(5), 674–682. https://doi.org/10.1111/jon.12725

Marks, M. P., Holmgren, E. B., Fox, A. J., Patel, S., von Kummer, R., & Froehlich, J. (1999). Evaluation of early computed tomographic findings in acute ischemic stroke. Stroke; a Journal of Cerebral Circulation, 30(2), 389–392. https://doi.org/10.1161/01.str.30.2.389

Marquand, A. F., Filippone, M., Ashburner, J., Girolami, M., Mourao-Miranda, J., Barker, G. J., Williams, S. C. R., Leigh, P. N., & Blain, C. R. V. (2013). Automated, high accuracy classification of Parkinsonian disorders: a pattern recognition approach. PloS One, 8(7), e69237. https://doi.org/10.1371/ journal.pone.0069237

Marzullo, A., Kocevar, G., Stamile, C., Durand-Dubief, F., Terracina, G., Calimeri, F., & Sappey-Marinier, D. (2019). Classification of Multiple Sclerosis Clinical Profiles via Graph Convolutional Neural Networks. Frontiers in Neuroscience, 13, 594. https://doi.org/10.3389/fnins.2019.00594

Matthews, P. M., Roncaroli, F., Waldman, A., Sormani, M. P., De Stefano, N., Giovannoni, G., & Reynolds, R. (2016). A practical review of the neuropathology and neuroimaging of multiple sclerosis. Practical Neurology, 16(4), 279–287. https://doi.org/10.1136/practneurol-2016-001381

McGinley, M. P., Goldschmidt, C. H., & Rae-Grant, A. D. (2021). Diagnosis and Treatment of Multiple Sclerosis: A Review.JAMA: The Journal of the American Medical Association, 325(8), 765–779. https://doi.org/10.1001/jama.2020.26858

McLouth, J., Elstrott, S., Chaibi, Y., Quenet, S., Chang, P. D., Chow, D. S., & Soun, J. E. (2021). Validation of a Deep Learning Tool in the Detection of Intracranial Hemorrhage and Large Vessel Occlusion. Frontiers in Neurology, 12, 656112. https://doi.org/10.3389/fneur.2021.656112

Menze, B. H., Jakab, A., Bauer, S., Kalpathy-Cramer, J., Farahani, K., Kirby, J., Burren, Y., Porz, N., Slotboom, J., Wiest, R., Lanczi, L., Gerstner, E., Weber, M.-A., Arbel, T., Avants, B. B., Ayache, N., Buendia, P., Collins, D. L., Cordier, N., … Van Leemput, K. (2015). The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS). IEEE Transactions on Medical Imaging, 34(10), 1993–2024. https://doi.org/10.1109/ TMI.2014.2377694

Merkaj, S., Bahar, R. C., Zeevi, T., Lin, M., Ikuta, I., Bousabarah, K., Cassinelli Petersen, G. I., Staib, L., Payabvash, S., Mongan, J. T., Cha, S., & Aboian, M. S. (2022). Machine Learning Tools for Image-Based Glioma Grading and the Quality of Their Reporting: Challenges and Opportunities. Cancers, 14(11). https://doi.org/10.3390/cancers14112623

Moazami, F., Lefevre-Utile, A., Papaloukas, C., & Soumelis, V. (2021). Machine Learning Approaches in Study of Multiple Sclerosis Disease Through Magnetic Resonance Images. Frontiers in Immunology, 12, 700582. https://doi.org/10.3389/ fimmu.2021.700582

Mohd Saad, N., Abdullah, A. R., Mohd Noor, N. S., & Mohd Ali, N. (2019). Automated Segmentation And Classification Technique For Brain Stroke. International Journal of Electrical, Computer, and Systems Engineering, 9(3), 1832–1841. https://doi.org/10.11591/ijece.v9i3.pp1832-1841

Mokin, M., Ansari, S. A., McTaggart, R. A., Bulsara, K. R., Goyal, M., Chen, M., Fraser, J. F., & Society of NeuroInterventional Surgery. (2019). Indications for thrombectomy in acute ischemic stroke from emergent large vessel occlusion (ELVO): report of the SNIS Standards and Guidelines Committee. Journal of Neurointerventional Surgery, 11(3), 215–220. https://doi.org/10.1136/ neurintsurg-2018-014640

Moradi, E., Pepe, A., Gaser, C., Huttunen, H., Tohka, J., & Alzheimer’s Disease Neuroimaging Initiative. (2015). Machine learning framework for early MRI-based Alzheimer’s conversion prediction in MCI subjects. NeuroImage, 104, 398–412. https://doi.org/10.1016/j.neuroimage.2014.10.002

Morey, J. R., Zhang, X., Yaeger, K. A., Fiano, E., Marayati, N. F., Kellner, C. P., De Leacy, R. A., Doshi, A., Tuhrim, S., & Fifi, J. T. (2021). Real-World Experience with Artificial Intelligence- Based Triage in Transferred Large Vessel Occlusion Stroke Patients. Cerebrovascular Diseases, 50(4), 450–455. https://doi.org/10.1159/000515320

Murray, N. M., Unberath, M., Hager, G. D., & Hui, F. K. (2020). Artificial intelligence to diagnose ischemic stroke and identify large vessel occlusions: a systematic review. Journal of Neurointerventional Surgery, 12(2), 156–164. https://doi.org/10.1136/neurintsurg-2019-015135

Nagel, S., Sinha, D., Day, D., Reith, W., Chapot, R., Papanagiotou, P., Warburton, E. A., Guyler, P., Tysoe, S., Fassbender, K., Walter, S., Essig, M., Heidenrich, J., Konstas, A. A., Harrison, M., Papadakis, M., Greveson, E., Joly, O., Gerry, S., … Grunwald, I. Q. (2017). e-ASPECTS software is non-inferior to neuroradiologists in applying the ASPECT score to computed tomography scans of acute ischemic stroke patients. International Journal of Stroke: Official Journal of the International Stroke Society, 12(6), 615–622. https://doi. org/10.1177/1747493016681020

Nazari-Farsani, S., Nyman, M., Karjalainen, T., Bucci, M., Isojärvi, J., & Nummenmaa, L. (2020). Automated segmentation of acute stroke lesions using a data-driven anomaly detection on diffusion weighted MRI. Journal of Neuroscience Methods, 333, 108575. https://doi.org/10.1016/j. jneumeth.2019.108575

Neeb, H., & Schenk, J. (2019). Multivariate prediction of multiple sclerosis using robust quantitative MR-based image metrics. Zeitschrift Fur Medizinische Physik, 29(3), 262–271. https://doi.org/10.1016/j.zemedi.2018.10.004

Ocasio, E., & Duong, T. Q. (2021). Deep learning prediction of mild cognitive impairment conversion to Alzheimer’s disease at 3 years after diagnosis using longitudinal and whole-brain 3D MRI. PeerJ. Computer Science, 7, e560. https://doi.org/10.7717/ peerj-cs.560

Olive-Gadea, M., Crespo, C., Granes, C., Hernandez-Perez, M., Pérez de la Ossa, N., Laredo, C., Urra, X., Carlos Soler, J., Soler, A., Puyalto, P., Cuadras, P., Marti, C., & Ribo, M. (2020). Deep Learning Based Software to Identify Large Vessel Occlusion on Noncontrast Computed Tomography. Stroke; a Journal of Cerebral Circulation, 51(10), 3133–3137. https://doi.org/10.1161/ STROKEAHA.120.030326

Olive-Gadea, M., Martins, N., Boned, S., Carvajal, J., Moreno, M. J., Muchada, M., Molina, C. A., Tomasello, A., Ribo, M., & Rubiera, M. (2019). Baseline ASPECTS and e-ASPECTS Correlation with Infarct Volume and Functional Outcome in Patients Undergoing Mechanical Thrombectomy. Journal of Neuroimaging: Official Journal of the American Society of Neuroimaging, 29(2), 198–202. https://doi.org/10.1111/ jon.12564

Olthof, A. W., van Ooijen, P. M. A., & Rezazade Mehrizi, M. H. (2020). Promises of artificial intelligence in neuroradiology: a systematic technographic review. Neuroradiology, 62(10), 1265–1278. https://doi.org/10.1007/s00234-020-02424-w

Öman, O., Mäkelä, T., Salli, E., Savolainen, S., & Kangasniemi, M. (2019). 3D convolutional neural networks applied to CT angiography in the detection of acute ischemic stroke. European Radiology Experimental, 3(1), 8. https://doi.org/10.1186/s41747-019-0085-6

Ozsahin, I., Sekeroglu, B., Pwavodi, P. C., & Mok, G. S. P. (2020). High-accuracy Automated Diagnosis of Parkinson’s Disease. Current Medical Imaging Reviews, 16(6), 688–694. https://doi.org/10.2174/1573405615666190620113607

Pagan, F. L. (2012). Improving outcomes through early diagnosis of Parkinson’s disease. The American Journal of Managed Care, 18(7 Suppl), S176–S182. https://www.ncbi.nlm. nih.gov/pubmed/23039866

Park, J. E., Kim, H. S., Park, S. Y., Nam, S. J., Chun, S.-M., Jo, Y., & Kim, J. H. (2020). Prediction of Core Signaling Pathway by Using Diffusion- and Perfusion-based MRI Radiomics and Next-generation Sequencing in Isocitrate Dehydrogenase Wildtype Glioblastoma. Radiology, 294(2), 388–397. https://doi. org/10.1148/radiol.2019190913

Peng, B., Wang, S., Zhou, Z., Liu, Y., Tong, B., Zhang, T., & Dai, Y. (2017). A multilevel-ROI-features-based machine learning method for detection of morphometric biomarkers in Parkinson’s disease. Neuroscience Letters, 651, 88–94. https://doi. org/10.1016/j.neulet.2017.04.034

Petersen, R. C. (2016). Mild Cognitive Impairment. Continuum, 22(2 Dementia), 404–418. https://doi.org/10.1212/CON.0000000000000313

Piccardo, A., Cappuccio, R., Bottoni, G., Cecchin, D., Mazzella, L., Cirone, A., Righi, S., Ugolini, M., Bianchi, P., Bertolaccini, P., Lorenzini, E., Massollo, M., Castaldi, A., Fiz, F., Strada, L., Cistaro, A., & Del Sette, M. (2021). The role of the deep convolutional neural network as an aid to interpreting brain [18F]DOPA PET/CT in the diagnosis of Parkinson’s disease. European Radiology, 31(9), 7003–7011. https://doi.org/10.1007/ s00330-021-07779-z

Pirson, F. A. V., Boodt, N., Brouwer, J., Bruggeman, A. A. E., den Hartog, S. J., Goldhoorn, R.-J. B., Langezaal, L. C. M., Staals, J., van Zwam, W. H., van der Leij, C., Brans, R. J. B., Majoie, C. B. L. M., Coutinho, J. M., Emmer, B. J., Dippel, D. W. J., van der Lugt, A., Vos, J.-A., van Oostenbrugge, R. J., Schonewille, W. J., & MR CLEAN Registry Investigators†. (2022). Endovascular Treatment for Posterior Circulation Stroke in Routine Clinical Practice: Results of the Multicenter Randomized Clinical Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands Registry. Stroke; a Journal of Cerebral Circulation, 53(3), 758–768. https://doi.org/10.1161/ STROKEAHA.121.034786

Pläschke, R. N., Cieslik, E. C., Müller, V. I., Hoffstaedter, F., Plachti, A., Varikuti, D. P., Goosses, M., Latz, A., Caspers, S., Jockwitz, C., Moebus, S., Gruber, O., Eickhoff, C. R., Reetz, K., Heller, J., Südmeyer, M., Mathys, C., Caspers, J., Grefkes, C., … Eickhoff, S. B. (2017). On the integrity of functional brain networks in schizophrenia, Parkinson’s disease, and advanced age: Evidence from connectivity-based single-subject classification. Human Brain Mapping, 38(12), 5845–5858. https://doi.org/10.1002/hbm.23763

Polson, J. S., Zhang, H., Nael, K., Salamon, N., Yoo, B. Y., El-Saden, S., Starkman, S., Kim, N., Kang, D.-W., Speier, W. F., 4th, & Arnold, C. W. (2022). Identifying acute ischemic stroke patients within the thrombolytic treatment window using deep learning. Journal of Neuroimaging: Official Journal of the American Society of Neuroimaging. https://doi.org/10.1111/jon.13043

Potreck, A., Weyland, C. S., Seker, F., Neuberger, U., Herweh, C., Hoffmann, A., Nagel, S., Bendszus, M., & Mutke, M. A. (2022). Accuracy and prognostic role of NCCT-ASPECTS depend on time from acute stroke symptom-onset for both human and machine-learning based evaluation. Clinical Neuroradiology, 32(1), 133–140. https://doi.org/10.1007/s00062-021-01110-5

Powers, W. J., Rabinstein, A. A., Ackerson, T., Adeoye, O. M., Bambakidis, N. C., Becker, K., Biller, J., Brown, M., Demaerschalk, B. M., Hoh, B., Jauch, E. C., Kidwell, C. S., Leslie-Mazwi, T. M., Ovbiagele, B., Scott, P. A., Sheth, K. N., Southerland, A. M., Summers, D. V., & Tirschwell, D. L. (2018). 2018 Guidelines for the Early Management of Patients With Acute Ischemic Stroke: A Guideline for Healthcare Professionals From the American Heart Association/American Stroke Association. Stroke; a Journal of Cerebral Circulation, 49(3), e46–e110. https://doi.org/10.1161/STR.0000000000000158

Qiu, W., Kuang, H., Teleg, E., Ospel, J. M., Sohn, S. I., Almekhlafi, M., Goyal, M., Hill, M. D., Demchuk, A. M., & Menon, B. K. (2020). Machine Learning for Detecting Early Infarction in Acute Stroke with Non-Contrast-enhanced CT. Radiology, 294(3), 638–644. https://doi.org/10.1148/ radiol.2020191193

Qubiotech. (2021, November 9). Qubiotech. https://qubiotech.com/en/resources/

Qureshi, A. I., Mendelow, A. D., & Hanley, D. F. (2009). Intracerebral haemorrhage. The Lancet, 373(9675), 1632–1644. https://doi.org/10.1016/S0140-6736(09)60371-8

Raimbault, A., Cazals, X., Lauvin, M.-A., Destrieux, C., Chapet, S., & Cottier, J.-P. (2014). Radionecrosis of malignant glioma and cerebral metastasis: a diagnostic challenge in MRI. Diagnostic and Interventional Imaging, 95(10), 985–1000. https:// doi.org/10.1016/j.diii.2014.06.013

Rao, B., Zohrabian, V., Cedeno, P., Saha, A., Pahade, J., & Davis, M. A. (2021). Utility of Artificial Intelligence Tool as a Prospective Radiology Peer Reviewer - Detection of Unreported Intracranial Hemorrhage. Academic Radiology, 28(1), 85–93. https://doi.org/10.1016/j.acra.2020.01.035

Rava, R. A., Peterson, B. A., Seymour, S. E., Snyder, K. V., Mokin, M., Waqas, M., Hoi, Y., Davies, J. M., Levy, E. I., Siddiqui, A. H., & Ionita, C. N. (2021). Validation of an artificial intelligence-driven large vessel occlusion detection algorithm for acute ischemic stroke patients. The Neuroradiology Journal, 34(5), 408–417. https://doi.org/10.1177/1971400921998952

Rava, R. A., Seymour, S. E., LaQue, M. E., Peterson, B. A., Snyder, K. V., Mokin, M., Waqas, M., Hoi, Y., Davies, J. M., Levy, E. I., Siddiqui, A. H., & Ionita, C. N. (2021). Assessment of an Artificial Intelligence Algorithm for Detection of Intracranial Hemorrhage. World Neurosurgery, 150, e209–e217. https://doi. org/10.1016/j.wneu.2021.02.134

Rezazade Mehrizi, M. H., van Ooijen, P., & Homan, M. (2021). Applications of artificial intelligence (AI) in diagnostic radiology: a technography study. European Radiology, 31(4), 1805–1811. https://doi.org/10.1007/s00330-020-07230-9

Rizzo, G., Copetti, M., Arcuti, S., Martino, D., Fontana, A., & Logroscino, G. (2016). Accuracy of clinical diagnosis of Parkinson disease: A systematic review and metaanalysis. Neurology, 86(6), 566–576. https://doi.org/10.1212/ WNL.0000000000002350

Roca, P., Attye, A., Colas, L., Tucholka, A., Rubini, P., Cackowski, S., Ding, J., Budzik, J.-F., Renard, F., Doyle, S., Barbier, E. L., Bousaid, I., Casey, R., Vukusic, S., Lassau, N., Verclytte, S., Cotton, F., OFSEP Investigators, Steering Committee, … Imaging group. (2020). Artificial intelligence to predict clinical disability in patients with multiple sclerosis using FLAIR MRI. Diagnostic and Interventional Imaging, 101(12), 795–802. https://doi.org/10.1016/j.diii.2020.05.009

Rocca, M. A., Anzalone, N., Storelli, L., Del Poggio, A., Cacciaguerra, L., Manfredi, A. A., Meani, A., & Filippi, M. (2021). Deep Learning on Conventional Magnetic Resonance Imaging Improves the Diagnosis of Multiple Sclerosis Mimics. Investigative Radiology, 56(4), 252–260. https://doi.org/10.1097/ RLI.0000000000000735

Rotstein, D., & Montalban, X. (2019). Reaching an evidencebased prognosis for personalized treatment of multiple sclerosis. Nature Reviews. Neurology, 15(5), 287–300. https://doi.org/10.1038/s41582-019-0170-8

Roy, S., Butman, J. A., Reich, D. S., Calabresi, P. A., & Pham, D. L. (2018). Multiple Sclerosis Lesion Segmentation from Brain MRI via Fully Convolutional Neural Networks. In arXiv [cs.CV]. arXiv. http://arxiv.org/abs/1803.09172

Saccà, V., Sarica, A., Novellino, F., Barone, S., Tallarico, T., Filippelli, E., Granata, A., Chiriaco, C., Bruno Bossio, R., Valentino, P., & Quattrone, A. (2019). Evaluation of machine learning algorithms performance for the prediction of early multiple sclerosis from resting-state FMRI connectivity data. Brain Imaging and Behavior, 13(4), 1103–1114. https://doi.org/10.1007/s11682-018-9926-9

Salvatore, C., Cerasa, A., Battista, P., Gilardi, M. C., Quattrone, A., Castiglioni, I., & Alzheimer’s Disease Neuroimaging Initiative. (2015). Magnetic resonance imaging biomarkers for the early diagnosis of Alzheimer’s disease: a machine learning approach. Frontiers in Neuroscience, 9, 307. https://doi.org/10.3389/fnins.2015.00307

Salvatore, C., Cerasa, A., Castiglioni, I., Gallivanone, F., Augimeri, A., Lopez, M., Arabia, G., Morelli, M., Gilardi, M. C., & Quattrone, A. (2014). Machine learning on brain MRI data for differential diagnosis of Parkinson’s disease and Progressive Supranuclear Palsy. Journal of Neuroscience Methods, 222, 230–237. https://doi.org/10.1016/j.jneumeth.2013.11.016

Samarasekera, S., Udupa, J. K., Miki, Y., Wei, L., & Grossman, R. I. (1997). A new computer-assisted method for the quantification of enhancing lesions in multiple sclerosis. Journal of Computer Assisted Tomography, 21(1), 145–151. https://doi.org/10.1097/00004728-199701000-00028

Schmidt, P., Gaser, C., Arsic, M., Buck, D., Förschler, A., Berthele, A., Hoshi, M., Ilg, R., Schmid, V. J., Zimmer, C., Hemmer, B., & Mühlau, M. (2012). An automated tool for detection of FLAIR-hyperintense white-matter lesions in Multiple Sclerosis. NeuroImage, 59(4), 3774–3783. https://doi.org/10.1016/j.neuroimage.2011.11.032

Schmitt, N., Mokli, Y., Weyland, C. S., Gerry, S., Herweh, C., Ringleb, P. A., & Nagel, S. (2022). Automated detection and segmentation of intracranial hemorrhage suspect hyperdensities in non-contrast-enhanced CT scans of acute stroke patients. European Radiology, 32(4), 2246–2254. https://doi.org/10.1007/s00330-021-08352-4

Schröder, J., & Thomalla, G. (2016). A Critical Review of Alberta Stroke Program Early CT Score for Evaluation of Acute Stroke Imaging. Frontiers in Neurology, 7, 245. https://doi.org/10.3389/ fneur.2016.00245

Schweitzer, A. D., Niogi, S. N., Whitlow, C. T., & Tsiouris, A. J. (2019). Traumatic Brain Injury: Imaging Patterns and Complications. Radiographics: A Review Publication of the Radiological Society of North America, Inc, 39(6), 1571–1595. https://doi.org/10.1148/rg.2019190076

Seker, F., Pfaff, J., Nagel, S., Vollherbst, D., Gerry, S., Möhlenbruch, M. A., Bendszus, M., & Herweh, C. (2019). CT Reconstruction Levels Affect Automated and Reader- Based ASPECTS Ratings in Acute Ischemic Stroke. Journal of Neuroimaging: Official Journal of the American Society of Neuroimaging, 29(1), 62–64. https://doi.org/10.1111/jon.12562

Siddique, M. M. R., Yang, D., He, Y., Xu, D., & Myronenko, A. (2022). Automated ischemic stroke lesion segmentation from 3D MRI. In arXiv [eess.IV]. arXiv. http://arxiv.org/abs/2209.09546

Singh, G., Manjila, S., Sakla, N., True, A., Wardeh, A. H., Beig, N., Vaysberg, A., Matthews, J., Prasanna, P., & Spektor, V. (2021). Radiomics and radiogenomics in gliomas: a contemporary update. British Journal of Cancer, 125(5), 641–657. https://doi.org/10.1038/s41416-021-01387-w

Skull fractures. (n.d.). Retrieved January 7, 2023, from https:// bestpractice.bmj.com/topics/en-gb/3000207

Song, T. (2019). Generative Model-Based Ischemic Stroke Lesion Segmentation. In arXiv [eess.IV]. arXiv. http://arxiv.org/ abs/1906.02392

Spasov, S., Passamonti, L., Duggento, A., Liò, P., Toschi, N., & Alzheimer’s Disease Neuroimaging Initiative. (2019). A parameter-efficient deep learning approach to predict conversion from mild cognitive impairment to Alzheimer’s disease. NeuroImage, 189, 276–287. https://doi.org/10.1016/j. neuroimage.2019.01.031

Strub, W. M., Leach, J. L., Tomsick, T., & Vagal, A. (2007). Overnight preliminary head CT interpretations provided by residents: locations of misidentified intracranial hemorrhage. AJNR. American Journal of Neuroradiology, 28(9), 1679–1682. https://doi.org/10.3174/ajnr.A0653

Tang, Y., Xiao, X., Xie, H., Wan, C.-M., Meng, L., Liu, Z.-H., Liao, W.-H., Tang, B.-S., & Guo, J.-F. (2017). Altered Functional Brain Connectomes between Sporadic and Familial Parkinson’s Patients. Frontiers in Neuroanatomy, 11, 99. https://doi.org/10.3389/fnana.2017.00099

Theocharakis, P., Glotsos, D., Kalatzis, I., Kostopoulos, S., Georgiadis, P., Sifaki, K., Tsakouridou, K., Malamas, M., Delibasis, G., Cavouras, D., & Nikiforidis, G. (2009). Pattern recognition system for the discrimination of multiple sclerosis from cerebral microangiopathy lesions based on texture analysis of magnetic resonance images. Magnetic Resonance Imaging, 27(3), 417–422. https://doi.org/10.1016/j. mri.2008.07.014

Thomalla, G., Cheng, B., Ebinger, M., Hao, Q., Tourdias, T., Wu, O., Kim, J. S., Breuer, L., Singer, O. C., Warach, S., Christensen, S., Treszl, A., Forkert, N. D., Galinovic, I., Rosenkranz, M., Engelhorn, T., Köhrmann, M., Endres, M., Kang, D. W., … Gerloff, C. (2011). DWI-FLAIR mismatch for the identification of patients with acute ischaemic stroke within 4·5 h of symptom onset (PRE-FLAIR): A multicentre observational study. Lancet Neurology, 10(11), 978–986. https:// doi.org/10.1016/S1474-4422(11)70192-2

Thomalla, G., Simonsen, C. Z., Boutitie, F., Andersen, G., Berthezene, Y., Cheng, B., Cheripelli, B., Cho, T.-H., Fazekas, F., Fiehler, J., Ford, I., Galinovic, I., Gellissen, S., Golsari, A., Gregori, J., Günther, M., Guibernau, J., Häusler, K. G., Hennerici, M., … Gerloff, C. (2018). MRI-Guided Thrombolysis for Stroke with Unknown Time of Onset. The New England Journal of Medicine, NEJMoa1804355. https://doi.org/10.1056/ NEJMoa1804355

Thompson, A. J., Banwell, B. L., Barkhof, F., Carroll, W. M., Coetzee, T., Comi, G., Correale, J., Fazekas, F., Filippi, M., Freedman, M. S., Fujihara, K., Galetta, S. L., Hartung, H. P., Kappos, L., Lublin, F. D., Marrie, R. A., Miller, A. E., Miller, D. H., Montalban, X., … Cohen, J. A. (2018). Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. Lancet Neurology, 17(2), 162–173. https://doi.org/10.1016/S1474- 4422(17)30470-2

Thust, S. C., van den Bent, M. J., & Smits, M. (2018). Pseudoprogression of brain tumors. Journal of Magnetic Resonance Imaging: JMRI, 48(3), 571–589. https://doi. org/10.1002/jmri.26171

Tommasin, S., Cocozza, S., Taloni, A., Giannì, C., Petsas, N., Pontillo, G., Petracca, M., Ruggieri, S., De Giglio, L., Pozzilli, C., Brunetti, A., & Pantano, P. (2021). Machine learning classifier to identify clinical and radiological features relevant to disability progression in multiple sclerosis. Journal of Neurology, 268(12), 4834–4845. https://doi.org/10.1007/s00415- 021-10605-7

Tsai, J. P., & Albers, G. W. (2017). Wake-Up Stroke. Topics in Magnetic Resonance Imaging: TMRI, 1. https://doi.org/10.1097/ RMR.0000000000000126

Valverde, S., Cabezas, M., Roura, E., González-Villà, S., Pareto, D., Vilanova, J. C., Ramió-Torrentà, L., Rovira, À., Oliver, A., & Lladó, X. (2017). Improving automated multiple sclerosis lesion segmentation with a cascaded 3D convolutional neural network approach. NeuroImage, 155, 159–168. https://doi.org/10.1016/j.neuroimage.2017.04.034

Valverde, S., Salem, M., Cabezas, M., Pareto, D., Vilanova, J. C., Ramió-Torrentà, L., Rovira, À., Salvi, J., Oliver, A., & Lladó, X. (2019). One-shot domain adaptation in multiple sclerosis lesion segmentation using convolutional neural networks. NeuroImage. Clinical, 21, 101638. https://doi.org/10.1016/j. nicl.2018.101638

van Asch, C. J., Luitse, M. J., Rinkel, G. J., van der Tweel, I., Algra, A., & Klijn, C. J. (2010). Incidence, case fatality, and functional outcome of intracerebral haemorrhage over time, according to age, sex, and ethnic origin: a systematic review and meta-analysis. Lancet Neurology, 9(2), 167–176. https://doi. org/10.1016/S1474-4422(09)70340-0

van Leeuwen, K. G., Meijer, F. J. A., Schalekamp, S., Rutten, M. J. C. M., van Dijk, E. J., van Ginneken, B., Govers, T. M., & de Rooij, M. (2021). Cost-effectiveness of artificial intelligence aided vessel occlusion detection in acute stroke: an early health technology assessment. Insights into Imaging, 12(1), 133. https://doi.org/10.1186/s13244-021-01077-4

van Leeuwen, K. G., Schalekamp, S., Rutten, M. J. C. M., van Ginneken, B., & de Rooij, M. (2021). Artificial intelligence in radiology: 100 commercially available products and their scientific evidence. European Radiology, 31(6), 3797–3804. https://doi.org/10.1007/s00330-021-07892-z

Wachinger, C., Reuter, M., & Klein, T. (2018). DeepNAT: Deep convolutional neural network for segmenting neuroanatomy. NeuroImage, 170, 434–445. https://doi.org/10.1016/j. neuroimage.2017.02.035

Wang, X., Shen, T., Yang, S., Lan, J., Xu, Y., Wang, M., Zhang, J., & Han, X. (2021). A deep learning algorithm for automatic detection and classification of acute intracranial hemorrhages in head CT scans. NeuroImage. Clinical, 32, 102785. https://doi. org/10.1016/j.nicl.2021.102785

Wardlaw, J. M., Mair, G., von Kummer, R., Williams, M. C., Li, W., Storkey, A. J., Trucco, E., Liebeskind, D. S., Farrall, A., Bath, P. M., & White, P. (2022). Accuracy of Automated Computer-Aided Diagnosis for Stroke Imaging: A Critical Evaluation of Current Evidence. Stroke; a Journal of Cerebral Circulation, 53(7), 2393–2403. https://doi.org/10.1161/ STROKEAHA.121.036204

Wattjes, M. P., Rovira, À., Miller, D., Yousry, T. A., Sormani, M. P., de Stefano, M. P., Tintoré, M., Auger, C., Tur, C., Filippi, M., Rocca, M. A., Fazekas, F., Kappos, L., Polman, C., Frederik Barkhof, Xavier Montalban, & MAGNIMS study group. (2015). Evidence-based guidelines: MAGNIMS consensus guidelines on the use of MRI in multiple sclerosis--establishing disease prognosis and monitoring patients. Nature Reviews. Neurology, 11(10), 597–606. https://doi.org/10.1038/nrneurol.2015.157

Wildner, P., Stasiołek, M., & Matysiak, M. (2020). Differential diagnosis of multiple sclerosis and other inflammatory CNS diseases. Multiple Sclerosis and Related Disorders, 37, 101452. https://doi.org/10.1016/j.msard.2019.101452

Wismüller, A., & Stockmaster, L. (2020). A prospective randomized clinical trial for measuring radiology study reporting time on Artificial Intelligence-based detection of intracranial hemorrhage in emergent care head CT. Medical Imaging 2020: Biomedical Applications in Molecular, Structural, and Functional Imaging, 11317, 144–150. https://doi. org/10.1117/12.2552400

Wong, K. K., Cummock, J. S., Li, G., Ghosh, R., Xu, P., Volpi, J. J., & Wong, S. T. C. (2022). Automatic Segmentation in Acute Ischemic Stroke: Prognostic Significance of Topological Stroke Volumes on Stroke Outcome. Stroke; a Journal of Cerebral Circulation, 101161STROKEAHA121037982. https://doi. org/10.1161/STROKEAHA.121.037982

World Stroke Organisation (2022) Global Stroke Fact Sheet. Retrieved December 2022 from https://www.world-stroke.org/ assets/downloads/WSO_Global_Stroke_Fact_Sheet.pdf

Wottschel, V., Alexander, D. C., Kwok, P. P., Chard, D. T., Stromillo, M. L., De Stefano, N., Thompson, A. J., Miller, D. H., & Ciccarelli, O. (2015). Predicting outcome in clinically isolated syndrome using machine learning. NeuroImage. Clinical, 7, 281–287. https://doi.org/10.1016/j.nicl.2014.11.021

Wottschel, V., Chard, D. T., Enzinger, C., Filippi, M., Frederiksen, J. L., Gasperini, C., Giorgio, A., Rocca, M. A., Rovira, A., De Stefano, N., Tintoré, M., Alexander, D. C., Barkhof, F., Ciccarelli, O., & MAGNIMS study group and the EuroPOND consortium. (2019). SVM recursive feature elimination analyses of structural brain MRI predicts nearterm relapses in patients with clinically isolated syndromes suggestive of multiple sclerosis. NeuroImage. Clinical, 24, 102011. https://doi.org/10.1016/j.nicl.2019.102011

Yahav-Dovrat, A., Saban, M., Merhav, G., Lankri, I., Abergel, E., Eran, A., Tanne, D., Nogueira, R. G., & Sivan-Hoffmann, R. (2021). Evaluation of Artificial Intelligence-Powered Identification of Large-Vessel Occlusions in a Comprehensive Stroke Center. AJNR. American Journal of Neuroradiology, 42(2), 247–254. https://doi.org/10.3174/ajnr.A6923

Ye, H., Gao, F., Yin, Y., Guo, D., Zhao, P., Lu, Y., Wang, X., Bai, J., Cao, K., Song, Q., Zhang, H., Chen, W., Guo, X., & Xia, J. (2019). Precise diagnosis of intracranial hemorrhage and subtypes using a three-dimensional joint convolutional and recurrent neural network. European Radiology, 29(11), 6191–6201. https:// doi.org/10.1007/s00330-019-06163-2

Yoo, Y., Tang, L. Y. W., Brosch, T., Li, D. K. B., Kolind, S., Vavasour, I., Rauscher, A., MacKay, A. L., Traboulsee, A., & Tam, R. C. (2018). Deep learning of joint myelin and T1w MRI features in normal-appearing brain tissue to distinguish between multiple sclerosis patients and healthy controls. NeuroImage. Clinical, 17, 169–178. https://doi.org/10.1016/j. nicl.2017.10.015

Zaki, L. A. M., Vernooij, M. W., Smits, M., Tolman, C., Papma, J. M., Visser, J. J., & Steketee, R. M. E. (2022). Comparing two artificial intelligence software packages for normative brain volumetry in memory clinic imaging. Neuroradiology, 64(7), 1359–1366. https://doi.org/10.1007/s00234-022-02898-w

Zhang, Q., Cao, J., Zhang, J., Bu, J., Yu, Y., Tan, Y., Feng, Q., & Huang, M. (2019). Differentiation of Recurrence from Radiation Necrosis in Gliomas Based on the Radiomics of Combinational Features and Multimodality MRI Images. Computational and Mathematical Methods in Medicine, 2019, 2893043. https://doi. org/10.1155/2019/2893043

Zhang, S., Nguyen, T. D., Zhao, Y., Gauthier, S. A., Wang, Y., & Gupta, A. (2018). Diagnostic accuracy of semiautomatic lesion detection plus quantitative susceptibility mapping in the identification of new and enhancing multiple sclerosis lesions. NeuroImage. Clinical, 18, 143–148. https://doi.org/10.1016/j. nicl.2018.01.013

Zhang, Y.-Q., Liu, A.-F., Man, F.-Y., Zhang, Y.-Y., Li, C., Liu, Y.-E., Zhou, J., Zhang, A.-P., Zhang, Y.-D., Lv, J., & Jiang, W.-J. (2022). MRI radiomic features-based machine learning approach to classify ischemic stroke onset time. Journal of Neurology, 269(1), 350–360. https://doi.org/10.1007/s00415-021-10638-y

Zhao, Y., Healy, B. C., Rotstein, D., Guttmann, C. R. G., Bakshi, R., Weiner, H. L., Brodley, C. E., & Chitnis, T. (2017). Exploration of machine learning techniques in predicting multiple sclerosis disease course. PloS One, 12(4), e0174866. https://doi.org/10.1371/journal.pone.0174866

Zhao, Y., Wang, T., Bove, R., Cree, B., Henry, R., Lokhande, H., Polgar-Turcsanyi, M., Anderson, M., Bakshi, R., Weiner, H. L., Chitnis, T., & SUMMIT Investigators. (2020). Ensemble learning predicts multiple sclerosis disease course in the SUMMIT study. NPJ Digital Medicine, 3, 135. https://doi. org/10.1038/s41746-020-00338-8

Zhu, H., Jiang, L., Zhang, H., Luo, L., Chen, Y., & Chen, Y. (2021). An automatic machine learning approach for ischemic stroke onset time identification based on DWI and FLAIR imaging. NeuroImage. Clinical, 31, 102744. https://doi.org/10.1016/j. nicl.2021.102744

Zurita, M., Montalba, C., Labbé, T., Cruz, J. P., Dalboni da Rocha, J., Tejos, C., Ciampi, E., Cárcamo, C., Sitaram, R., & Uribe, S. (2018). Characterization of relapsing-remitting multiple sclerosis patients using support vector machine classifications of functional and diffusion MRI data. NeuroImage. Clinical, 20, 724–730. https://doi.org/10.1016/j. nicl.2018.09.002

Guide to Artificial Intelligence in Radiology

Introduction

Artificial intelligence (AI) is playing a growing role in all our lives and has shown promise in addressing some of the greatest current and upcoming societal challenges we face. The healthcare industry, though notoriously complex and resistant to disruption, potentially has a lot to gain from the use of AI. With an established history of leading digital transformation in healthcare and an urgent need for improved efficiency, radiology has been at the forefront of harnessing AI’s potential.

This book covers how and why AI can address challenges faced by radiology departments, provides an overview of the fundamental concepts related to AI, and describes some of the most promising use cases for AI in radiology. In addition, the major challenges associated with the adoption of AI into routine radiological practice are discussed. The book also covers some crucial points radiology departments should keep in mind when deciding on which AI-based solutions to purchase. Finally, it provides an outlook on what new and evolving aspects of AI in radiology to expect in the near future.

The case for artificial intelligence in radiology

The healthcare industry has experienced a number of trends over the past few decades that demand a change in the way certain things are done. These trends are particularly salient in radiology, where the diagnostic quality of imaging scans has improved dramatically while scan times have decreased. As a result, the amount and complexity of medical imaging data acquired have increased substantially over the past few decades (Smith-Bindman et al., 2019; Winder et al., 2021) and are expected to continue to increase (Tsao, 2020). This issue is complicated by a widespread global shortage of radiologists (AAMC Report Reinforces Mounting Physician Shortage, 2021, Clinical Radiology UK Workforce Census 2019 Report, 2019). Healthcare workers, including radiologists, have an increasing workload (Bruls & Kwee, 2020; Levin et al., 2017) that contributes to burnout and medical errors (Harry et al., 2021). Being an essential service provider to virtually all other hospital departments, staff shortages within radiology have significant effects that spread throughout the hospital and to society as a whole (England & Improvement, 2019; Sutherland et al., n.d.).

With an ageing global population and a rising burden of chronic illnesses, these issues are expected to pose even more of a challenge to the healthcare industry in the future.

AI-based medical imaging solutions have the potential to ameliorate these challenges for several reasons. They are particularly suited to handling large, complex datasets (Alzubaidi et al., 2021). Moreover, they are well suited to automate some of the tasks traditionally performed by radiologists and radiographers, potentially freeing up time and making workflows within radiology departments more efficient (Allen et al., 2021; Baltruschat et al., 2021; Kalra et al., 2020; O’Neill et al., 2021; van Leeuwen et al., 2021; Wong et al., 2019). AI is also capable of detecting complex patterns in data that humans cannot necessarily find or quantify (Dance, 2021; Korteling et al., 2021; Kühl et al., 2020).

Fundamentals of artificial intelligence

The term “artificial intelligence” refers to the use of computer systems to solve specific problems in a way that simulates human reasoning. One fundamental characteristic of AI is that, like humans, these systems can tailor their solutions to changing circumstances. Note that, while these systems are meant to mimic on a fundamental level how humans think, their capacity to do so (e.g. in terms of the amount of data they can handle at one time, the nature and amount of patterns they can find in the data, and the speed at which they do so) often exceeds that of humans.

AI solutions come in the form of computer algorithms, which are pieces of computer code representing instructions to be followed to solve a specific problem. In its most fundamental form, the algorithm takes data as an input, performs some computation on that data, and returns an output.

An AI algorithm can be explicitly programmed to solve a specific task, analogous to a step-by-step recipe for baking a cake. On the other hand, the algorithm can be programmed to look for patterns within the data in order to solve the problem. These types of algorithms are known as machine learning algorithms. Thus, all machine learning algorithms are AI, but not all AI is machine learning. The patterns in the data that the algorithm can be explicitly programmed to look for or that it can “discover” by itself are known as features. An important characteristic of machine learning is that such algorithms learn from the data itself, and their performance improves the more data they are given.

One of the most common uses of machine learning is in classification - assigning a piece of data a particular label. For example, a machine learning algorithm might be used to tell if a photo (the input) shows a dog or a cat (the label). The algorithm can learn to do so in a supervised or unsupervised way.

Supervised learning

In supervised learning, the machine learning algorithm is given data that has been labelled with the ground truth, in this example, photos of dogs and cats that have been labelled as such. The process then goes through the following phases:

1.Training phase: The algorithm learns the features associated with dogs and cats using the aforementioned data (training data).
2.Test phase: The algorithm is then given a new set of photos (the test data), it labels them and the performance of the algorithm on that data is assessed.

In some cases, there is a phase in between training and test, known as the validation phase. In this phase, the algorithm is given a new set of photos (not included in either the training or test data), its performance is assessed on this data, and the model is tweaked and retrained on the training data. This is repeated until some predefined performance-based criterion is reached, and the algorithm then enters the test phase.

Unsupervised learning

In unsupervised learning, the algorithm identifies features within the input data that allow it to assign classes to the individual data points without being told explicitly what those classes are or should be. Such algorithms can identify patterns or group data points together without human intervention and include clustering and dimensionality reduction algorithms. Not all machine learning algorithms perform classification. Some are used to predict a continuous metric (e.g. the temperature in four weeks’ time) instead of a discrete label (e.g. cats vs dogs). These are known as regression algorithms.

Neural networks and deep learning

A neural network is made up of an input layer and an output layer, which are themselves composed of nodes. In simple neural networks, features that are manually derived from a dataset are fed into the input layer, which performs some computations, the results of which are relayed to the output layer. In deep learning, multiple “hidden” layers exist between the input and output layers. Each node of the hidden layers performs calculations using certain weights and relays the output to the next hidden layer until the output layer is reached.

In the beginning, random values are assigned to the weights and the accuracy of the algorithm is calculated. The values of the weights are then iteratively adjusted until a set of weight values that maximize accuracy is found. This iterative adjustment of the weight values is usually done by moving backwards from the output layer to the input layer, a technique called backpropagation. This entire process is done on the training data.

Performance evaluation

Understanding how the performance of AI algorithms is assessed is key to interpreting the AI literature. Several performance metrics exist for assessing how well a model performs certain tasks. No single metric is perfect, so a combination of several metrics provides a fuller picture of model performance.

In regression, the most commonly used metrics include:

Mean absolute error (MAE): the average difference between the predicted values and the ground truth.
Root mean square error (RMSE): the differences between the predicted values and the ground truth are squared and then averaged over the sample. Then the square root of the average is taken. Unlike the MAE, the RMSE thus gives higher weight to larger differences.
R²: the proportion of the total variance in the ground truth explained by the variance in the predicted values. It ranges from 0 to 1.

The following metrics are commonly used in classification tasks:

Accuracy: this is the proportion of all predictions that were predicted correctly. It ranges from 0 to 1.
Sensitivity: also known as the true positive rate (TPR) or recall, this is the proportion of true positives that were predicted correctly. It ranges from 0 to 1.
Specificity: Also known as the true negative rate (TNR), this is the proportion of true negatives that were predicted correctly. It ranges from 0 to 1.
Precision: also known as positive predictive value (PPV), this is the proportion of positive classifications that were predicted correctly. It ranges from 0 to 1.

An inherent trade-off exists between sensitivity and specificity. The relevant importance of each, as well as their interpretation, highly depends on the specific research question and classification task.

Importantly, although classification models are meant to reach a binary conclusion, they are inherently probability-based. This means that these models will output a probability that a data point belongs to one class or another. In order to reach a conclusion on the most likely class, a threshold is used. Metrics such as accuracy, sensitivity, specificity and precision refer to the performance of the algorithm based on a certain threshold. The area under the receiver operating characteristic curve (AUC) is a threshold-independent performance metric. The AUC can be interpreted as the probability that a random positive example is ranked higher by the algorithm than a random negative example.

In image segmentation tasks, which are a type of classification task, the following metrics are commonly used:

Dice similarity coefficient: a measure of overlap between two sets (e.g. two images) that is calculated as two times the number of elements common to the sets divided by the sum of the number of elements in each set. It ranges from 0 (no overlap) to 1 (perfect overlap).
Hausdorff distance: a measure of how far two sets (e.g. two images) within a space are far from each other. It is basically the largest distance from one point in one set to the closest point in the other set.

Internal and external validity

Internally valid models perform well in their task on the data being used to train and validate them. The degree to which they are internally valid is assessed using the performance metrics outlined above and depends on the characteristics of the model itself and the quality of the data that the model was trained and validated on.

Externally valid models perform well in their tasks on new data (Ramspek et al., 2021). The better the model performs on data that differs from the data the models were trained and validated on, the higher the external validity. In practice, this often requires the performance of the models to be tested on data from hospitals or geographical areas that were not part of the model’s training and validation datasets.

Guidelines for evaluating AI research

Several guidelines have been developed to assess the evidence behind AI-based interventions in healthcare (X. Liu et al., 2020; Mongan et al., 2020; Shelmerdine et al., 2021; Weikert et al., 2021). These provide a template for those doing AI research in healthcare and ensure that relevant information is reported transparently and comprehensively, but can also be used by other stakeholders to assess the quality of published research. This helps ensure that AI-based solutions with substantial potential or actual limitations, particularly those caused by poor reporting (Bozkurt et al., 2020; D. W. Kim et al., 2019; X. Liu et al., 2019; Nagendran et al., 2020; Yusuf et al., 2020), are not prematurely adopted (CONSORT-AI and SPIRIT-AI Steering Group, 2019). Guidelines have also been proposed for evaluating the trustworthiness of AI-based solutions in terms of transparency, confidentiality, security, and accountability (Buruk et al., 2020; Lekadir et al., 2021; Zicari et al., 2021).

Clinical uses

Over the past few years, AI has shown great potential in addressing a broad range of tasks within a medical imaging department, including many that happen before the patient is scanned. Implementations of AI to improve the efficiency of radiology workflows prior to patient scanning are sometimes referred to as “upstream AI” (Kapoor et al., 2020; M. L. Richardson et al., 2021).

Scheduling

One promising upstream AI application is predicting whichpatients arelikelytomisstheirscanappointments. Missed appointments are associated with significantly increased workload and costs (Dantas et al., 2018). Using a Gradient Boosting approach, Nelson et al. predicted missed hospital magnetic resonance imaging (MRI) appointments in the United Kingdom’s National Health Service (NHS) with high accuracy (Nelson et al., 2019). Their simulations also suggested that acting on the predictions of this model by targeting patients who are likely to miss their appointments would potentially yield a net benefit of several pounds per appointment across a range of model thresholds and missed appointment rates (Nelson et al., 2019). Similar results were recently found in a study of a single hospital in Singapore. For the 6-month period following the deployment of the predictive tool they were able to significantly reduce the no show rate from 19.3 % tp 15.9 % which translated into a potential economic benefit of $180,000 (Chong et. al., 2020).

Scheduling scans in a radiology department is a challenging endeavour because, although it is largely an administrative task, it depends heavily on medical information. The task of assigning patients to specific appointments thus often requires the input of someone with domain knowledge, which stipulates that either the person making the appointments must be a radiologist or radiology technician, or these people will have to provide input regularly. In either scenario, the process is somewhat inefficient and can potentially be streamlined using AI-based algorithms that check scan indications and contraindications and provide the people scheduling the scans with information about scan urgency (Letourneau-Guillon et al., 2020).

Protocolling

Depending on hospital or clinic policy, the decision on what exact scan protocol a patient receives is usually made based on the information on the referring physician’s scan request and the judgement of the radiologist. This is often supplemented by direct communication between the referring physician and radiologist and the radiologist’s review of the patient’s medical information. This process improves patient care (Boland et al., 2014) but can be time-consuming and inefficient, particularly with modalities like MRI, where a large number of protocol permutations exist. In one study, protocolling alone accounted for about 6 % of the radiologist’s working time (Schemmel et al., 2016). Radiologists are also often interrupted by tasks such as protocolling when interpreting images, despite the fact that the latter is considered a radiologist’s primary responsibility (Balint et al., 2014; J.-P. J. Yu et al., 2014).

Interpretation of the narrative text of the referring physician’s scan request has been attempted using natural language classifiers, the same technology used in chatbots and virtual assistants. Natural language classifiers based on deep learning have shown promise in assigning patients to either a contrast-enhanced or non-enhanced MRI protocol for musculoskeletal MRI, with an accuracy of 83 % (Trivedi et al., 2018) and 94 % (Y. H. Lee, 2018). Similar algorithms have shown an accuracy of 95 % for predicting the appropriate brain MRI protocol using a combination of up to 41 different MRI sequences (Brown & Marotta, 2018). Across a wide range of body regions, a deep-learning-based natural language classifier decided based on the narrative text of the scan requests whether to automatically assign a specific computed tomography (CT) or MRI protocol (which it did with 95 % accuracy) or, in more difficult cases, recommend a list of three most appropriate protocols to the radiologist (which it did with 92 % accuracy) (Kalra et al., 2020).

AI has also been used to decide whether already protocolled scans need to be extended, a decision which has to be made in real-time while the patient is inside the scanner. One such example is in prostate MRI, where a decision on whether to administer a contrast agent is often made after the non-contrast sequences. Hötker et al. found that a convolutional neural network (CNN) assigned 78 % of patients to the appropriate prostate MRI protocol (Hötker et al., 2021). The sensitivity of the CNN for the need for contrast was 94.4 % with a specificity of 68.8 % and only 2 % of patients in their study would have had to be called back for a contrast- enhanced scan (Hötker et al., 2021).

Image quality improvement and monitoring

Many AI-based solutions that work in the background of radiology workflows to improve image quality have recently been established. These include solutions for monitoring image quality, reducing image artefacts, improving spatial resolution, and speeding up scans.

Such solutions are entering the radiology mainstream, particularly for computed tomography, which for decades used established but artefact-prone methods for reconstructing interpretable images from the raw sensor data (Deák et al., 2013; Singh et al., 2010).

These are gradually being replaced by deep-learning- based reconstruction methods, which improve image quality while maintaining low radiation doses (Akagi et al., 2019; H. Chen et al., 2017; Choe et al., 2019; Shan et al., 2019). This reconstruction is performed on supercomputers on the CT scanner itself or on the cloud. The balance between radiation dose and image quality can be adjusted on a protocol-specific basis to tailor scans to individual patients and clinical scenarios (McLeavy et al., 2021; Willemink & Noël, 2019). Such approaches have found particular use when scanning children, pregnant women, and obese patients as well as CT scans of the urinary tract and heart (McLeavy et al., 2021).

AI-based solutions have also been used to speed up scans while maintaining diagnostic quality. Scan time reduction not only improves overall efficiency but also contributes to an overall better patient experience and compliance with imaging examination. A multi- centre study of spine MRI showed that a deep-learning- based image reconstruction algorithm that enhanced images using filtering and detail-preserving noise reduction reduced scan times by 40 % (Bash, Johnson, et al., 2021). For T1-weighted MRI scans of the brain, a similar algorithm that improves image sharpness and reduces image noise reduced scan times by 60 % while maintaining the accuracy of brain region volumetry compared to standard scans (Bash, Wang, et al., 2021).

In routine radiological practice, images often contain artefacts that reduce their interpretability. These artefacts are the result of characteristics of the specific imaging modality or protocol used or factors intrinsic to the patient being scanned, such as the presence of foreign bodies or the patient moving during the scan. Particularly with MRI, imaging protocols that demand fast scanning often introduce certain artefacts to the reconstructed image. In one study, a deep-learning- based algorithm reduced banding artefacts associated with balanced steady-state free precession MRI sequences of the brain and knee (K. H. Kim & Park, 2017). For real-time imaging of the heart using MRI, another study found that the aliasing artefacts introduced by the data undersampling were reduced by using a deep-learning-based approach (Hauptmann et al., 2019). The presence of metallic foreign bodies such as dental, orthopaedic or vascular implants is a common patient-related factor causing image artefacts in both CT and MRI (Boas & Fleischmann, 2012; Hargreaves et al., 2011). Although not yet well established, several deep-learning-based approaches for reducing these artefacts have been investigated (Ghani & Clem Karl, 2019; Puvanasunthararajah et al., 2021; Zhang & Yu, 2018). Similar approaches are being tested for reducing motion-related artefacts in MRI (Tamada et al., 2020; B. Zhao et al., 2022).

AI-based solutions for monitoring image quality potentially reduce the need to call patients back to repeat imaging examinations, which is a common problem (Schreiber-Zinaman & Rosenkrantz, 2017). A deep-learning-based algorithm that identifies the radiographic view acquired and extracts quality-related metrics from ankle radiographs was able to predict image quality with about 94 % accuracy (Mairhöfer et al., 2021). Another deep-learning-based approach was capable of predicting nondiagnostic liver MRI scans with a negative predictive value of between 86 % and 94 % (Esses et al., 2018). This real-time automated quality control potentially allows radiology technicians to rerun scans or run additional scans with greater diagnostic value.

Scan reading prioritization

With staff shortages and increasing scan numbers, radiologists face long reading lists. To optimize efficiency and patient care, AI-based solutions have been suggested as a way to prioritize which scans radiologists read and report first, usually by screening acquired images for findings that require urgent intervention (O’Connor & Bhalla, 2021). This has been most extensively studied in neuroradiology, where moving CT scans that were found to have intracranial haemorrhage by an AI-based tool to the top of the reading list reduced the time it took radiologists to view the scans by several minutes (O’Neill et al., 2021). Another study found that the time-to diagnosis (which includes the time from image acquisition to viewing by the radiologist and the time to read and report the scans) was reduced from 512 to 19 minutes in an outpatient setting when such a worklist prioritization was used (Arbabshirani et al., 2018). A simulation study using AI-based worklist prioritization based on identifying urgent findings on chest radiographs (such as pneumothorax, pleural effusions, and foreign bodies) also found a substantial reduction in the time it took to view and report the scans compared to standard workflow prioritization (Baltruschat et al., 2021).

Image interpretation

Currently, the majority of commercially available AI- based solutions in medical imaging focus on some aspect of analyzing and interpreting images (Rezazade Mehrizi et al., 2021; van Leeuwen et al., 2021). This includes segmenting parts of the image (for surgical or radiation therapy targeting, for example), bringing suspicious areas to radiologists’ attention, extracting imaging biomarkers (radiomics), comparing images across time, and reaching specific imaging diagnoses.

Neurology

¡ 29–38 % of commercially available AI-based applications in radiology (Rezazade Mehrizi et al., 2021; van Leeuwen et al., 2021).

Most commercially available AI-based solutions targeted at neuroimaging data aim to detect and characterize ischemic stroke, intracranial haemorrhage, dementia, and multiple sclerosis (Olthof et al., 2020). Several studies have shown excellent accuracy of AI- based methods for the detection and classification of intraparenchymal, subarachnoid, and subdural haemorrhage on head CT (Flanders et al., 2020; Ker et al., 2019; Kuo et al., 2019). Subsequent studies showed that, compared to radiologists, some AI-based solutions have substantially lower false positive and negative rates (Ginat, 2020; Rao et al., 2021). In ischemic stroke, AI-based solutions have largely focused on the quantification of the infarct core (Goebel et al., 2018; Maegerlein et al., 2019), the detection of large vessel occlusion (Matsoukas et al., 2022; Morey et al., 2021; Murray et al., 2020; Shlobin et al., 2022), and the prediction of stroke outcomes (Bacchi et al., 2020; Nielsen et al., 2018; Y. Yu et al., 2020, 2021).

In multiple sclerosis, AI has been used to identify and segment lesions (Nair et al., 2020; S.-H. Wang et al., 2018), which can be particularly helpful for the longitudinal follow-up of patients. It has also been used to extract imaging features associated with progressive disease and conversion from clinically isolated syndrome to definite multiple sclerosis (Narayana et al., 2020; Yoo et al., 2019). Other applications of AI in neuroradiology include the detection of intracranial aneurysms (Faron et al., 2020; Nakao et al., 2018; Ueda et al., 2019) and the segmentation of brain tumours (Kao et al., 2019; Mlynarski et al., 2019; Zhou et al., 2020) as well as the prediction of brain tumour genetic markers from imaging data (Choi et al., 2019; J. Zhao et al., 2020)

Chest

¡ 24 %–31 % of commercially available AI-based applications in radiology (Rezazade Mehrizi et al., 2021; van Leeuwen et al., 2021).

When interpreting chest radiographs, radiologists detected substantially more critical and urgent findings when aided by a deep-learning-based algorithm, and did so much faster than without the algorithm (Nam et al., 2021). Deep-learning-based image interpretation algorithms have also been found to improve radiology residents’ sensitivity for detecting urgent findings on chest radiographs from 66 % to 73 % (E. J. Hwang, Nam, et al., 2019). Another study which focused on a broader range of findings on chest radiographs also found that radiologists aided by a deep-learning-based algorithm had higher diagnostic accuracy than radiologists who read the radiographs without assistance (Seah et al., 2021). The uses of AI in chest radiology also extend to cross-sectional imaging like CT. A deep learning algorithm was found to detect pulmonary embolism on CT scans with high accuracy (AUC = 0.85) (Huang, Kothari, et al., 2020). Moreover, a deep learning algorithm was 90 % accurate in detecting aortic dissection on non-contrast-enhanced CT scans, similar to the performance of radiologists (Hata et al., 2021).

Outside the emergency setting, AI-based solutions have been widely tested and implemented for tuberculosis screening on chest radiographs (E. J. Hwang, Park, et al., 2019; S. Hwang et al., 2016; Khan et al., 2020; Qin et al., 2019; WHO Operational Handbook on Tuberculosis Module 2: Screening – Systematic Screening for Tuberculosis Disease, n.d.). In addition, they have been useful for lung cancer screening both in terms of detecting lung nodules on CT (Setio et al., 2017) and chest radiographs (Li et al., 2020) and by classifying whether nodules are likely to be malignant or benign (Ardila et al., 2019; Bonavita et al., 2020; Ciompi et al., 2017; B. Wu et al., 2018). AI-based solutions also show great promise for the diagnosis of pneumonia, chronic obstructive pulmonary disease, and interstitial lung disease (F. Liu et al., 2021).

Breast

¡ 11 % of commercially available AI-based applications in radiology (Rezazade Mehrizi et al., 2021; van Leeuwen et al., 2021).

So far, many of the AI-based algorithms targeting breast imaging aim to reduce the workload of radiologists reading mammograms. Ways to do this have included using AI-based algorithms to triage out negative mammograms, which in one study was associated with a reduction in radiologists’ workload by almost one-fifth (Yala et al., 2019). Other studies that have replaced second readers of mammograms with AI- based algorithms have shown that this leads to fewer false positives and false negatives as well as reduces the workload of the second reader by 88 % (McKinney et al., 2020).

AI-based solutions for mammography have also been found to increase the diagnostic accuracy of radiologists (McKinney et al., 2020; Rodríguez-Ruiz et al., 2019; Watanabe et al., 2019) and some have been found to be highly accurate in independently detecting and classifying breast lesions (Agnes et al., 2019; Al- Antari et al., 2020; Rodriguez-Ruiz et al., 2019).
Despite this, a recent systematic review of 36 AI- based algorithms found that these studies were of poor methodological quality and that all algorithms were less accurate than the consensus of two or more radiologists (Freeman et al., 2021). AI-based algorithms have nonetheless shown potential for extracting cancer-predictive features from mammograms beyond mammographic breast density (Arefan et al., 2020; Dembrower et al., 2020; Hinton et al., 2019). Beyond mammography, AI-based solutions have been developed for detecting and classifying breast lesions on ultrasound (Akkus et al., 2019; Park et al., 2019; G.- G. Wu et al., 2019) and MRI (Herent et al., 2019).

Cardiac

¡ 11 % of commercially available AI-based applications in radiology (Rezazade Mehrizi et al., 2021; van Leeuwen et al., 2021).

Cardiac radiology has always been particularly challenging because of the difficulties inherent in acquiring images of a constantly moving organ. Because of this, it has benefited immensely from advances in imaging technology and seems set to benefit greatly from AI as well (Sermesant et al., 2021). Most of the AI-based applications of the cardiovascular system use MRI, CT or ultrasound data (Weikert et al., 2021). Prominent examples include the automated calculation of ejection fraction on echocardiography, quantification of coronary artery calcification on cardiac CT, determination of right ventricular volume on CT pulmonary angiography, and determination of heart chamber size and thickness on cardiac MRI (Medical AI Evaluation, n.d., The Medical Futurist, n.d.). AI-based solutions for the prediction of patients likely to respond favourably to cardiac interventions, such as cardiac resynchronization therapy, based on imaging and clinical parameters have also shown great promise (Cikes et al., 2019; Hu et al., 2019). Changes in cardiac MRI not readily visible to human readers but potentially useful for differentiating different types of cardiomyopathies can also be detected using AI through texture analysis (Neisius et al., 2019; J. Wang et al., 2020) and other radiomic approaches (Mancio et al., 2022).

Musculoskeletal

¡ 7–11 % of commercially available AI-based applications in radiology (Rezazade Mehrizi et al., 2021; van Leeuwen et al., 2021).

Promising applications of AI in the assessment of muscles, bones and joints include applications where human readers generally show poor between- and within-rater reliability, such as the determination of skeletal age based on bone radiographs (Halabi et al., 2019; Thodberg et al., 2009) and screening for osteoporosis on radiographs (Kathirvelu et al., 2019; J.-S. Lee et al., 2019) and CT (Pan et al., 2020). AI- based solutions have also shown promise for detecting fractures on radiographs and CT (Lindsey et al., 2018; Olczak et al., 2017; Urakawa et al., 2019). One systematic review of AI-based solutions for fracture detection in several different body parts showed AUCs ranging from 0.94 to 1.00 and accuracies of 77 % to 98 % (Langerhuizen et al., 2019). AI-based solutions have also achieved accuracies similar to radiologists for classification of the severity of degenerative changes of the spine (Jamaludin et al., 2017) and extremity joints (F. Liu et al., 2018; Thomas et al., 2020). AI-based solutions have also been developed to determine the origin of skeletal metastases (Lang et al., 2019) and the classification of primary bone tumours (Do et al., 2017).

Abdomen and pelvis

¡ 4 % of commercially available AI-based applications in radiology (Rezazade Mehrizi et al., 2021; van Leeuwen et al., 2021).

Much of the efforts in using AI in abdominal imaging have thus far concentrated on the automated segmentation of organs such as the liver (Dou et al., 2017), spleen (Moon et al., 2019), pancreas (Oktay et al., 2018), and kidneys (Sharma et al., 2017). In addition, a systematic review of 11 studies using deep learning for the detection of malignant liver masses showed accuracies of up to 97 % and AUCs of up to 0.92 (Azer, 2019).

Other applications of AI in abdominal radiology include the detection of liver fibrosis (He et al., 2019; Yasaka et al., 2018), fatty liver disease, hepatic iron content, the detection of free abdominal gas on CT, and automated volumetry and segmentation of the prostate (AI for Radiology, n.d.).

Obstacles to implementation

Despite the great potential of AI in medical imaging, it has yet to find widespread implementation and impact in routine clinical practice. This research-to- clinic translation is being hindered by several complex and interrelated issues that directly or indirectly lower the likelihood of AI-based solutions being adopted. One major way they do so is by creating a lack of trust in AI- based solutions by key stakeholders such as regulators, healthcare professionals and patients (Cadario et al., 2021; Esmaeilzadeh, 2020; J. P. Richardson et al., 2021; Tucci et al., 2022).

Generalizability

One major challenge is to develop AI-based solutions that continue to perform well in new, real-world scenarios. In a large systematic review, almost half of the studied AI-based medical imaging algorithms reported a greater than 0.05 decrease in the AUC when tested on new data (A. C. Yu et al., 2022). This lack of generalizability can lead to adverse effects on how well the model performs in a real-world scenario.

If a solution performs poorly when tested on a dataset with a similar or identical distribution to the training dataset, it is said to lack narrow generalizability and is often a consequence of overfitting (Eche et al., 2021). Potential solutions for overfitting are using larger training datasets and reducing the model’s complexity. If a solution performs poorly when tested on a dataset with a different distribution to the training dataset (e.g. a different distribution of patient ethnicities), it is said to lack broad generalizability (Eche et al., 2021). Solutions to poor broad generalizability include stress-testing the model on datasets with different distributions from the training dataset (Eche et al., 2021).

AI solutions are often developed in a high-resource environment such as large technology companies and academic medical centres in wealthy countries. It is likely that findings and performance in these high-resource contexts will fail to generalize to lower- resource contexts such as smaller hospitals, rural areas or poorer countries (Price & Nicholson, 2019), which complicates the issue further.

Risk of bias

Biases can arise in AI-based solutions due to data or human factors. The former occurs when the data used to train the AI solution does not adequately represent the target population. Datasets can be unrepresentative when they are too small or have been collected in a way that misrepresents a certain population category. AI solutions trained on unrepresentative data perpetuate biases and perform poorly in the population categories underrepresented or misrepresented in the training data. The presence of such biases has been empirically shown in many AI-based medical imaging studies (Larrazabal et al., 2020; Seyyed-Kalantari et al., 2021).

AI-based solutions are prone to several subjective and sometimes implicitly or explicitly prejudiced decisions during their development by humans. These human factors include how the training data is selected, how it is labelled, and how the decision is made to focus on the specific problem the AI-based solution intends to solve (Norori et al., 2021). Some recommendations and tools are available to help minimize the risk of bias in AI research (AIF360: A Comprehensive Set of Fairness Metrics for Datasets and Machine Learning Models, Explanations for These Metrics, and Algorithms to Mitigate Bias in Datasets and Models, n.d., IBM Watson Studio - Model Risk Management, n.d.; Silberg & Manyika, 2019).

Data quantity, quality and variety

Problems such as bias and lack of generalizability can be mitigated by ensuring that training data is of sufficient quantity, quality and variety. However, this is difficult to do because patients are often reluctant to share their data for commercial purposes (Aggarwal, Farag, et al., 2021; Ghafur et al., 2020; Trinidad et al., 2020), hospitals and clinics are usually not equipped to make this data available in a useable and secure manner, and organizing and labelling the data is time- consuming and expensive.

Many datasets can be used for a number of different purposes, and sharing data between companies can help make the process of data collection and organization more efficient, as well as increase the amount of data available for each application. However, developers are often reluctant to share data with each other, or even reveal the exact source of their data, to stay competitive.

Data protection and privacy

The development and implementation of AI-based solutions require that patients are explicitly informed about, and give their consent to, the use of their data for a particular purpose and by certain people. This data also has to be adequately protected from data breaches and misuse. Failure to ensure this greatly undermines the public’s trust in AI-based solutions and hinders their adoption. While regulations governing health data privacy state that the collection of fully anonymized data does not require explicit patient consent (General Data Protection Regulation (GDPR) – Official Legal Text, 2016; Office for Civil Rights (OCR), 2012) and in theory protects from the data being misused, whether or not imaging data can be fully anonymized is controversial (Lotan et al., 2020; Murdoch, 2021). Whether consent can be truly informed considering the complexity of the data being acquired, and the resulting myriad of potential future uses of the data, is also disputed (Vayena & Blasimme, 2017).

IT infrastructure

Among hospital departments, radiology has always been at the forefront ofdigitalization. AI-based solutions that focus on image processing and interpretation are likely to find the prerequisite infrastructure in most radiology departments, for example for linking imaging equipment to computers for analysis and for archiving images and other outputs. However, most radiology departments are likely to require significant infrastructure upgrades for other applications of AI, particularly those requiring the integration of information from multiple sources and having complex outputs. Moreover, it is important to keep in mind that the distribution of necessary infrastructure is highly unequal across and within countries (Health Ethics & Governance, 2021).

In terms of computing power, radiology departments will either have to invest resources into the hardware and personnel necessary to run these AI-based solutions or opt for cloud-based solutions. The former comes with an extra cost but allows data processing within the confines of the hospital or clinic’s local network. Cloud-based solutions for computing (known as “infrastructure as a service” or “IaaS”) are often considered the less secure and less trustworthy option, but this depends on a number of factors and is thus not always true (Baccianella & Gough, n.d.). Guidelines on what to consider when procuring cloud-based solutions in healthcare are available (Cloud Security for Healthcare Services, 2021).

Lack of standardization, interoperability, and integrability

The problem of infrastructure becomes even more complicated when considering how fragmented the AI medical imaging market currently is (Alexander et al., 2020). It is therefore likely that in the near future a single department will have several dozen AI-based solutions from different vendors running simultaneously. Having a separate self-contained infrastructure (e.g. a workstation or server) for each of these would be incredibly complicated and difficult to manage. Suggested solutions for this have included AI solution “marketplaces”, similar to app stores (Advanced AI Solutions for Radiology, n.d., Curated Marketplace, 2018, Imaging AI Marketplace - Overview, n.d., Sectra Amplifier Marketplace, 2021, The Nuance AI Marketplace for Diagnostic Imaging, n.d.), and development of an overarching vendor-neutral infrastructure (Leiner et al., 2021). The successful implementation of such solutions requires close partnerships between AI solution developers, imaging vendors and information technology companies.

Interpretability

It is often impossible to understand exactly how AI- based solutions come to their conclusions, particularly with complex approaches like deep learning. This reduces how transparent the decision-making process for procuring and approving these solutions can be, makes the identification of biases difficult, and makes it harder for clinicians to explain the outputs of these solutions to their patients and to determine whether a solution is working properly or has malfunctioned (Char et al., 2018; Reddy et al., 2020; Vayena et al., 2018; Whittlestone et al., 2019). Some have suggested that techniques that help humans understand how AI- based algorithms made certain decisions or predictions (“interpretable” or “explainable” AI) might help mitigate these challenges. However, others have argued that currently available techniques are unsuitable for understanding individual decisions of an algorithm and have warned against relying on them for ensuring that algorithms work in a safe and reliable way (Ghassemi et al., 2021).

Liability

In healthcare systems, a framework of accountability ensures that healthcare workers and medical institutions can be held responsible for adverse effects resulting from their actions. The question of who should be held accountable for the failures of an AI- based solution is complicated. For pharmaceuticals, for example, the accountability for inherent failures in the product or its use often lies with either the manufacturer or the prescriber. One key difference is that AI-based systems are continuously evolving and learning, and so inherently work in a way that is independent of what their developers could have foreseen (Yeung, 2018). To the end-user such as the healthcare worker, the AI- based solution may be opaque and so they may not be able to tell when the solution is malfunctioning or inaccurate (Habli et al., 2020; Yeung, 2018).

Brittleness

Despite substantial progress in their development over the past few years, deep learning algorithms are still surprising brittle. This means that, when the algorithm faces a scenario that differs substantially from what it faced during training, it cannot contextualize and often produces nonsensical or inaccurate results. This happens because, unlike humans, most algorithms learn to perceive things within the confines of certain assumptions, but fail to generalize outside these assumptions. As an example of how this can be abused with malicious intent, subtle changes to medical images, imperceptible by humans, can render the results of disease-classifying algorithms inaccurate (Finlayson et al., 2018). The lack of interpretability of many AI-based solutions compounds this problem because it makes it difficult to troubleshoot how they reached the wrong conclusion.

Making purchasing decisions

So far, more than 100 AI-based products have gained conformité européenne (CE) marking or Food and Drug Adminstration (FDA) clearance. These products can be found in continuously updated and searchable online databases curated by the FDA (Center for Devices & Radiological Health, n.d.), the American College of Radiology (Assess-AI, n.d.), and others (AI for Radiology, n.d., The Medical Futurist, n.d.; E. Wu et al., 2021). The increasing number of available products, the inherent complexity of many of these solutions, and the fact that many people who usually make purchasing decisions in hospitals are not familiar with evaluating such products make it important to think carefully when deciding on which product to purchase. Such decisions will need to be made after incorporating input from healthcare workers, information technology (IT) professionals, as well as management, finance, legal, and human resources professionals within hospitals.

Deciding on whether to purchase an AI-based solution in radiology, as well as which of the increasing number of commercially available solutions to purchase, includes considerations of quality, safety, and finances. Over the past few years, several guidelines have emerged to help potential buyers make these decisions (A Buyer’s Guide to AI in Health and Care, 2020; Omoumi et al., 2021; Reddy et al., 2021), and these guidelines are likely to evolve in the future with changing expectations from customers, regulatory bodies, and stakeholders involved in reimbursement decisions.

First of all, it has to be clear to the potential buyer what the problem is and whether AI is the appropriate approach to this solution, or whether alternatives exist that are more advantageous on balance. If AI is the appropriate approach, buyers should know exactly what a potential AI-based product’s scope of the solution is - i.e. what specific problem the AI-based solution is designed to solve and in what specific circumstances. This includes whether the solution is intended for screening, diagnosis, monitoring, treatment recommendation or another application. It also includes the intended users of the solution and what kind of specific qualifications or training they are expected to have in order to be able to operate the solution and interpret its outputs. It needs to be clear to buyers whether the solution is intended to replace certain tasks that would normally be performed by the end-user, act as a double-reader, as a triaging mechanism, or for other tasks like quality control. Buyers should also understand whether the solution is intended to provide “new” information (i.e. information that would otherwise be unavailable to the user without the solution), improve the performance of an existing task beyond a human’s or other non-AI-based solution’s performance or if it is intended to save time or other resources.

Buyers should also have access to information that allows them to assess the potential benefits of the AI solution, and this should be backed up by published scientific evidence for the efficacy and cost-efficiency of the solution. How this is done will depend highly on the solution itself and the context in which it is expected to be deployed, but guidelines for this are available (National Institute for Health and Care Excellence (NICE), n.d.). Some questions to ask here would be: How much of an influence will the solution have on patient management? Will it improve diagnostic performance? Will it save time and money? Will it affect patients’ quality of life? It should also be clear to the buyer who exactly is expected to benefit from the use of this solution (Radiologists? Clinicians? Patients? The healthcare system or society as a whole?).

As with any healthcare intervention, all AI-based solutions come with potential risks, and these should be made clear to the buyer. Some of these risks might have legal consequences, such as the potential for misdiagnosis. These risks should be quantified, and potential buyers should have a framework for dealing with them, including identifying a framework for accountability within the organizations implementing these solutions. Buyers should also ensure they clearly understand the potential negative effects on radiologists’ training and the potential disruption to radiologists’ workflows associated with the use of these solutions.

Specifics of the AI solution’s design are also relevant to the decision on whether or not to purchase it. These include how robust the solution is to differences between vendors and scanning parameters, the circumstances under which the algorithm was trained (including potential confounding factors), and the way that performance was assessed. It should also be clear to buyers if and how potential sources of bias were accounted for during development. Because a core characteristic of AI-based solutions is their ability to continuously learn from new data, whether and how exactly this retraining is incorporated into the solution with time should also be clear to the buyer, including whether or not new regulatory approval is needed with each iteration. This also includes whether or not retraining is required, for example, due to changes in imaging equipment at the buyer’s institution.

The main selling points of many AI-based solutions are ease-of-use and improved workflows. Therefore, potential buyers should carefully scrutinize how these solutions are to be integrated into existing workflows, including inter-operability with PACS and electronic medical record systems. Whether or not the solution requires extra hardware (e.g. graphical processing units) or software (e.g. for visualization of the solution’s outputs), or if it can readily be integrated into the existing information technology infrastructure of the buyer’s organization influences the overall cost of the solution for the buyer and is therefore also a critical consideration. In addition, the degree of manual interaction required, both under normal circumstances and for troubleshooting, should be known to the buyer. All potential users of the AI solution should be involved in the purchasing process to ensure that they are familiar with it and that it meets their professional ethical standards and suits their needs.

From a regulatory perspective, it should be clear to the buyer whether the solution complies with medical device and data protection regulations. Has the solution been approved in the buyer’s country? If so, under which risk classification? Buyers should also consider creating data flow maps that display how the data flows in the operation of the AI-based solution, including who has access to the data.

Finally, there are other factors to consider which are not necessarily unique to AI-based solutions and which buyers might be familiar with from purchasing other types of solutions. This includes the licensing model of the solution, how users are to be trained on using the solution, how the solution is maintained, how failures in the solution are dealt with, and whether additional costs are to be expected when scaling up the solution’s implementation (e.g. using the solution for more imaging equipment or more users). This allows the potential buyer to anticipate the current and future costs of purchasing the solution.

Future trends

The past decade of increasing interest and progress in AI-based solutions for medical imaging has set the stage for a number of trends that are likely to appear or intensify in the near future.

Firstly, there is an increasing sentiment that, although AI holds a great deal of promise for interpretive applications (such as the detection of pathology), non-interpretive AI-based solutions might hold the most potential in terms of instilling efficiency into radiology workflows and improving patient experiences. This trend towards involving AI earlier in the patient management process is likely to extend to AI increasingly acting as a clinical decision support system to guide when and which imaging scans are performed.

For this to happen, AI needs to be integrated into existing clinical information systems, and the specific algorithms used need to be able to handle more varied data. This will likely pave the way for the development of algorithms that are capable of integrating demographic, clinical, and laboratory patient data to make recommendations about patient management (Huang, Pareek, et al., 2020; Rockenbach, 2021). The previously mentioned natural language processing algorithms that have been used to interpret scan requests may be useful candidates for this.

In addition, we are likely to see AI algorithms that can interpret multiple different types of imaging data from the same patient. Currently, less than 5 % of commercially available AI-based solutions in medical imaging work with more than one imaging modality (Rezazade Mehrizi et al., 2021; van Leeuwen et al., 2021) despite the fact that the typical patient in a hospital receives multiple imaging scans during their stay (Shinagare et al., 2014). With this, it is also likely that more AI-based solutions will be developed that target hitherto neglected modalities such as nuclear imaging techniques and ultrasound.

The current market for AI-based solutions in radiology is spread across a relatively large number of companies (Alexander et al., 2020). Potential users are likely to expect a streamlined integration of these products in their workflows, which can be challenging in such a fragmented market. Improved integration can be achieved in several different ways, including with vendor-neutral marketplaces or by the gradual consolidation of providers of AI-based solutions.

With the expanding use of AI, the issue of trust between AI developers, healthcare professionals, regulators, and patients will become more relevant. It is therefore likely that efforts will intensify to take steps towards strengthening that trust. This will potentially include raising the expected standards of evidence for AI- based solutions (Aggarwal, Sounderajah, et al., 2021; X. Liu et al., 2019; van Leeuwen et al., 2021; Yusuf et al., 2020), making them more transparent through the use and improvement of interpretable AI techniques (Holzinger et al., 2017; Reyes et al., 2020; “Towards Trustable Machine Learning,” 2018), and enhancing techniques for maintaining patient data privacy (G. Kaissis et al., 2021; G. A. Kaissis et al., 2020).

Furthermore, while most existing regulations stipulate that AI-based algorithms cannot be modified after regulatory approval, this is likely to change in the future. The potential for these algorithms to learn from data acquired after approval and adapt to changing circumstances is a major advantage of AI. Still, frameworks for doing so have thus far been lacking in the healthcare sector. However, promising ideas have recently emerged, including adapting existing hospital quality assurance and improvement frameworks to monitor AI-based algorithms’ performance and the data they are trained on and update the algorithms accordingly (Feng et al., 2022). This will likely require the development of multidisciplinary teams within hospitals consisting of clinicians, IT professionals, and biostatisticians who closely collaborate with model developers and regulators (Feng et al., 2022).

While the obstacles discussed in previous sections might slow down the adoption of AI in radiology somewhat, the fear of AI potentially replacing radiologists is unlikely to be one of them. A recent survey from Europe showed that most radiologists did not perceive a reduction in their clinical workload after adopting AI-based solutions (European Society of Radiology (ESR), 2022), likely because, at the same time, demand for radiologists’ services has been continuously rising. Studies from around the world have shown that radiology professionals, particularly those with AI exposure and experience, are generally optimistic about the role of AI in their practice (Y. Chen et al., 2021; Huisman et al., 2021; Ooi et al., 2021; Santomartino & Yi, 2022; Scott et al., 2021).

Conclusion

AI has shown promise in positively impacting virtually every facet of a radiology department’s work - from scheduling and protocolling patient scans to interpreting images and reaching diagnoses. Promising research on AI-based tools in radiology has not yet been widely translated to adoption in routine practice, however, because of a number of complex, partially intertwined issues. Potential solutions exist for many of these challenges, but many of these solutions require further refinement and testing. In the meantime, guidelines are emerging to help potential users of AI-based solutions in radiology navigate the increasing number of commercial products. This encourages their adoption in real-world scenarios, thus allowing their true potential to be uncovered, as well as their weaknesses to be identified and addressed in a safe and effective way. As these incremental improvements are made, these tools will likely evolve to handle more varied data, become integrated into consolidated workflows, become more transparent, and ultimately more useful for increasing efficiency and improving patient care.

References