Alessandro Cagol,Rosa Cortese,Muhamed Barakovic,Sabine Schaedelin,Esther Ruberte,Martina Absinta,Frederik Barkhof,Massimiliano Calabrese,Marco Castellaro,Olga Ciccarelli,Sirio Cocozza,Nicola De Stefano,Christian Enzinger,Massimo Filippi,Maciej Jurynczyk,Pietro Maggi,Nima Mahmoudi,Silvia Messina,Xavier Montalban,Jacqueline Palace,Giuseppe Pontillo,Anne-Katrin Pröbstel,Maria A Rocca,Stefan Ropele,Àlex Rovira,Menno M Schoonheim,Piotr Sowa,Eva Strijbis,Mike P Wattjes,Maria Pia Sormani,Ludwig Kappos,Cristina Granziera,MAGNIMS Study Group,Jaume Sastre-Garriga,Claudio Gasperini,Hugo Vrenken,Tarek Yousry

Abstract:Importance: Multiple sclerosis (MS) misdiagnosis remains an important issue in clinical practice. Objective: To quantify the performance of cortical lesions (CLs) and central vein sign (CVS) in distinguishing MS from other conditions showing brain lesions on magnetic resonance imaging (MRI). Design, setting, and participants: This was a retrospective, cross-sectional multicenter study, with clinical and MRI data acquired between January 2010 and May 2020. Centralized MRI analysis was conducted between July 2020 and December 2022 by 2 raters blinded to participants' diagnosis. Participants were recruited from 14 European centers and from a multicenter pan-European cohort. Eligible participants had a diagnosis of MS, clinically isolated syndrome (CIS), or non-MS conditions; availability of a brain 3-T MRI scan with at least 1 sequence suitable for CL and CVS assessment; presence of T2-hyperintense white matter lesions (WMLs). A total of 1051 individuals were included with either MS/CIS (n = 599; 386 [64.4%] female; mean [SD] age, 41.5 [12.3] years) or non-MS conditions (including other neuroinflammatory disorders, cerebrovascular disease, migraine, and incidental WMLs in healthy control individuals; n = 452; 302 [66.8%] female; mean [SD] age, 49.2 [14.5] years). Five individuals were excluded due to missing clinical or demographic information (n = 3) or unclear diagnosis (n = 2). Exposures: MS/CIS vs non-MS conditions. Main outcomes and measures: Area under the receiver operating characteristic curves (AUCs) were used to explore the diagnostic performance of CLs and the CVS in isolation and in combination; sensitivity, specificity, and accuracy were calculated for various cutoffs. The diagnostic importance of CLs and CVS compared to conventional MRI features (ie, presence of infratentorial, periventricular, and juxtacortical WMLs) was ranked with a random forest model. Results: The presence of CLs and the previously proposed 40% CVS rule had a sensitivity, specificity, and accuracy for MS of 59.0% (95% CI, 55.1-62.8), 93.6% (95% CI, 91.4-95.6), and 73.9% (95% CI, 71.6-76.3) and 78.7% (95% CI, 75.5-82.0), 86.0% (95% CI, 82.1-89.5), and 81.5% (95% CI, 78.9-83.7), respectively. The diagnostic performance of the CVS (AUC, 0.89 [95% CI, 0.86-0.91]) was superior to that of CLs (AUC, 0.77 [95% CI, 0.75-0.80]; P < .001), and was increased when combining the 2 imaging markers (AUC, 0.92 [95% CI, 0.90-0.94]; P = .04); in the random forest model, both CVS and CLs outperformed the presence of infratentorial, periventricular, and juxtacortical WMLs in supporting MS differential diagnosis. Conclusions and relevance: The findings in this study suggest that CVS and CLs may be valuable tools to increase the accuracy of MS diagnosis.

Classification of Radiologically Isolated Syndrome and Clinically Isolated Syndrome with Machine-Learning Techniques

Joint Radiomics and Spatial Distribution Model for MRI-based Discrimination of Multiple Sclerosis, Neuromyelitis Optica Spectrum Disorder, and Myelin-Oligodendrocyte-glycoprotein-igg-associated Disorder.

Interpretable Machine Learning for Predicting Multiple Sclerosis Conversion from Clinically Isolated Syndrome

Emerging imaging markers in radiologically isolated syndrome: implications for earlier treatment initiation

Prediction of Conversion from CIS to Clinically Definite Multiple Sclerosis Using Convolutional Neural Networks

Cortical lesions, central vein sign, and paramagnetic rim lesions in multiple sclerosis: emerging machine learning techniques and future avenues

Radiologically isolated syndromes: to treat or not to treat?

Early Disease-Modifying Treatments for Presymptomatic Multiple Sclerosis

Multi-lesion radiomics model for discrimination of relapsing-remitting multiple sclerosis and neuropsychiatric systemic lupus erythematosus

Quantitative Susceptibility Mapping-Derived Radiomic Features in Discriminating Multiple Sclerosis From Neuromyelitis Optica Spectrum Disorder

Differentiating multiple sclerosis from non-specific white matter changes using a convolutional neural network image classification model

Predicting MCI to AD Conversation Using Integrated sMRI and rs-fMRI: Machine Learning and Graph Theory Approach

Predicting multiple sclerosis disease progression and outcomes with machine learning and MRI-based biomarkers: a review

Diagnostic Performance of Cortical Lesions and the Central Vein Sign in Multiple Sclerosis

Comparison of the diagnostic accuracy of resting-state fMRI driven machine learning algorithms in the detection of mild cognitive impairment

Multi-parametric MRI Phenotype with Trustworthy Machine Learning for Differentiating CNS Demyelinating Diseases

An Algorithm to Predict the Need for Magnetic Resonance Imaging to Support Disease Monitoring in Multiple Sclerosis

Independent component analysis-based classification of Alzheimer's disease MRI data.

Multiple Sclerosis Lesion Analysis in Brain Magnetic Resonance Images: Techniques and Clinical Applications

Machine learning-driven diagnosis of multiple sclerosis from whole blood transcriptomics

Diagnosis of Schizophrenia and its Subtypes Using MRI and Machine Learning