Deep Learning Modeling to Differentiate Multiple Sclerosis From MOG Antibody-Associated Disease

Abstract

Background and objectives: Multiple sclerosis (MS) is common in adults while myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD) is rare. Our previous machine-learning algorithm, using clinical variables, ≤6 brain lesions, and no Dawson fingers, achieved 79% accuracy, 78% sensitivity, and 80% specificity in distinguishing MOGAD from MS but lacked validation. The aim of this study was to (1) evaluate the clinical/MRI algorithm for distinguishing MS from MOGAD, (2) develop a deep learning (DL) model, (3) assess the benefit of combining both, and (4) identify key differentiators using probability attention maps (PAMs).

Methods: This multicenter, retrospective, cross-sectional MAGNIMS study included scans from 19 centers. Inclusion criteria were as follows: adults with non-acute MS and MOGAD, with high-quality T2-fluid-attenuated inversion recovery and T1-weighted scans. Brain scans were scored by 2 readers to assess the performance of the clinical/MRI algorithm on the validation data set. A DL-based classifier using a ResNet-10 convolutional neural network was developed and tested on an independent validation data set. PAMs were generated by averaging correctly classified attention maps from both groups, identifying key differentiating regions.

Results: We included 406 MRI scans (218 with relapsing remitting MS [RRMS], mean age: 39 years ±11, 69% F; 188 with MOGAD, age: 41 years ±14, 61% F), split into 2 data sets: a training/testing set (n = 265: 150 with RRMS, age: 39 years ±10, 72% F; 115 with MOGAD, age: 42 years ±13, 61% F) and an independent validation set (n = 141: 68 with RRMS, age: 40 years ±14, 65% F; 73 with MOGAD, age: 40 years ±15, 63% F). The clinical/MRI algorithm predicted RRMS over MOGAD with 75% accuracy (95% CI 67-82), 96% sensitivity (95% CI 88-99), and specificity 56% (95% CI 44-68) in the validation cohort. The DL model achieved 77% accuracy (95% CI 64-89), 73% sensitivity (95% CI 57-89), and 83% specificity (95% CI 65-96) in the training/testing cohort, and 70% accuracy (95% CI 63-77), 67% sensitivity (95% CI 55-79), and 73% specificity (95% CI 61-83) in the validation cohort without retraining. When combined, the classifiers reached 86% accuracy (95% CI 81-92), 84% sensitivity (95% CI 75-92), and 89% specificity (95% CI 81-96). PAMs identified key region volumes: corpus callosum (1872 mm³), left precentral gyrus (341 mm³), right thalamus (193 mm³), and right cingulate cortex (186 mm³) for identifying RRMS and brainstem (629 mm³), hippocampus (234 mm³), and parahippocampal gyrus (147 mm³) for identifying MOGAD.

Discussion: Both classifiers effectively distinguished RRMS from MOGAD. The clinical/MRI model showed higher sensitivity while the DL model offered higher specificity, suggesting complementary roles. Their combination improved diagnostic accuracy, and PAMs revealed distinct damage patterns. Future prospective studies should validate these models in diverse, real-world settings.

Classification of evidence: This study provides Class III evidence that both a clinical/MRI algorithm and an MRI-based DL model accurately distinguish RRMS from MOGAD.

Figure 1. Schematic Representation of the Study Design

The work was structured in 3 steps: (A) FLAIR/T2 sequences of the scans from the validation test set were assessed and scored considering the number of lesions and the presence of Dawson fingers (blue). (B) ResNet-10 architecture was trained using cross-entropy loss function and then accuracy was assessed on the validation test set (green). Then, individual attention maps were created using a guided backpropagation approach and then normalized using z-scores. Finally, 2 PAMs (1 for each disease group) were created by nonlinearly registering to MNI and then averaging all the correctly classified attention maps of RRMS and MOGAD, respectively (yellow). (C) The DL-based and clinical/MRI classifiers were combined to assess their performances in differentiating the 2 disorders (red). In brief, both the number of lesions and the presence of Dawson fingers were concatenated to the features extracted by the DL network and used to discriminate between the 2 classes. CNN = convolutional neural network; DL = deep learning; FLAIR = fluid-attenuated inversion recovery; MNI = Montreal Neurological Institute; MOGAD = myelin oligodendrocyte glycoprotein antibody–associated disease; MRI = magnetic resonance imaging; PAMs = probability attention maps; RRMS = relapsing remitting multiple sclerosis.

Figure 2. Topographical Distributions of the Areas Significantly Contributing to the Differentiation Between Patients With RRMS and With MOGAD

PAMs revealed the involvement of corpus callosum, right cingulate cortex, right thalamus, and left precentral gyrus for the identification of RRMS (A) while the relevant role of the brainstem, bilateral hippocampi, and parahippocampal gyri in the identification of MOGAD (B). Significant voxels are shown in a color scale from light blue to dark blue for RRMS and from yellow to red for MOGAD, from the most to the less significant, respectively (all p < 0.001). MOGAD = myelin oligodendrocyte glycoprotein antibody–associated disease; PAMs = probability attention maps; RRMS = relapsing remitting multiple sclerosis.

Figure 3. Voxel-Wise Comparison of RRMS and MOGAD Lesion Maps

The voxel-wise comparison between RRMS and MOGAD showed a highest probability of RRMS lesions in the periventricular area. Significant voxels are shown in a color scale from yellow to red for RRMS, from the most to the least significant, respectively (all p < 0.001). MOGAD = myelin oligodendrocyte glycoprotein antibody–associated disease; RRMS = relapsing remitting multiple sclerosis.

Figure 4. Performance of DL-Based Classifiers in Differentiating RRMS and MOGAD

The figure shows examples of MRI, as classified by the DL classifier. (A) For MS, the classifier correctly identified oval-shaped periventricular and deep white matter lesions on axial FLAIR, with Dawson fingers and involvement of the temporal lobe. (B) For MOGAD, the classifier accurately detected large, ill-defined lesions in the juxtacortical/cortical and periventricular regions. (C) An example of MS misclassified as MOGAD revealed a poorly defined hyperintensity in the right cerebellar peduncle, along with more than 6 supratentorial lesions, some of which were round-shaped, and 1 Dawson finger. (D) By contrast, MOGAD misclassified as MS exhibited a well-defined peripheral lesion in the pons, alongside 4 supratentorial confluent lesions, including 2 round-shaped lesions involving the corpus callosum. The integration of clinical and imaging classifiers enabled accurate diagnosis in (C) and (D). DL = deep learning; FLAIR = fluid-attenuated inversion recovery; MNI = Montreal Neurological Institute; MOGAD = myelin oligodendrocyte glycoprotein antibody–associated disease; MS = multiple sclerosis; RRMS = relapsing remitting MS.