Artificial intelligence in the diagnosis of multiple sclerosis: A systematic review

In recent years Artificial intelligence (AI) techniques are rapidly evolving into clinical practices such as diagnosis and prognosis processes, assess treatment effectiveness, and monitoring of diseases. The previous studies showed interesting results regarding the diagnostic efficiency of AI methods in differentiating Multiple sclerosis (MS) patients from healthy controls or other demyelinating diseases. There is a great lack of a comprehensive systematic review study on the role of AI in the diagnosis of MS. We aimed to perform a systematic review to document the performance of AI in MS diagnosis.

A systematic search was performed using four databases including PubMed, Scopus, Web of Science, and IEEE on August 2021. All original studies which focused on deep learning or AI to analyze any modalities with the purpose of diagnosing MS were included in our study.

Finally, 38 studies were included in our systematic review after the abstract and full-text screening. A total of 5433 individuals were included, including 2924 cases of MS and 2509 healthy controls. Sensitivity and specificity were reported in 29 studies which ranged from 76.92 to 100 for sensitivity and 74 to 100 for specificity. Furthermore, 34 studies reported accuracy ranged 81 to 100. Among included studies, Magnetic Resonance Imaging (MRI) (20 studies), OCT (six studies), serum and cerebrospinal fluid markers (six studies), movement function (three studies), and other modalities such as breathing and evoked potential was used for detecting MS via AI.

In conclusion, diagnosis of MS based on new markers and AI is a growing field of research with MRI images, followed by images obtained from OCT, serum and CSF biomarkers, and motor associated markers. All of these results show that with advances made in AI, the way we monitor and diagnose our MS patients can change drastically.