Abstract Details

The diagnosis of MS relies on exclusion of alternate neurological disorders. Misdiagnosis of MS is mainly due to improper interpretation of MRI and distinguishing MS from radiological mimics can be challenging.

To evaluate radiological features to optimally distinguish multiple sclerosis (MS) from other disorders with white matter lesions.

Patients with standardized MRI acquired on 3T Siemens scanners across the health care system were included and classified, based on clinical chart review as MS (final diagnosis of MS) and non-MS cohorts (final diagnosis non-specific changes, migraine related changes, small vessel disease ischemic changes, and other diagnoses). The earliest available MRIs within the first 5 years of disease onset were included for MS patients. Variables included lesion volumes, whole brain, and regional volumes. Variables with greater than 80% correlation with other variables were removed. A logistic regression model with Ridge regularization and grid-search was applied to the cohort to select the top 6 features distinguishing MS from other disease states.

The MS cohort included 665 patients: 210 (32%) male, 532 (80%) Caucasian, and 89 (13%) African American. The non-MS cohort included 501 patients: 102 (20%) male, 395 (79%) Caucasian, and 44 (9%) African American. In the MS cohort, the average disease duration was 2.9 ± 1.4 years, the average age at time of MRI scan was 40.5 ± 11.7, and the average age at MS diagnosis was 36.6 ± 11.1 years. The top 6 variables included T2 lesion volume and thalamic sub-regional volumes and the logistic regression model was able to predict MS with an accuracy of 63% and AUC of 71%.