Differential diagnosis of migraine and tension-type headache based on brain volumes using machine learning methods.

PURPOSE: Structural brain changes in migraine and tension-type headache (TTH) have mostly been investigated using voxel-based morphometry or regional volume analyses. However, the diagnostic discriminatory power of these volumetric parameters within machine learning (ML) models has not been sufficiently explored. This study evaluated the performance of brain volumes normalized to total intracranial volume (TIV) in distinguishing migraine from TTH. METHODS: This retrospective, cross-sectional study included 220 participants (110 migraine, 110 TTH). High-resolution 3D T1-weighted MRI images were analyzed using the VolBrain automatic segmentation system. The volumes of the cerebrum, cerebellum, thalamus, amygdala, caudate nucleus, hippocampus, and putamen were normalized to TIV. Age, sex, and TIV-normalized brain volumetric features were used as model inputs, and feature standardization was performed within each cross-validation fold to prevent data leakage. Eight different supervised ML algorithms were compared using a nested 5-fold cross-validation method. Performance was evaluated using accuracy, sensitivity, specificity, F1-score, and AUC metrics. Additionally, the brain volumes of both groups were compared using statistical analysis. RESULTS: Classification performance was poor across all models. The Random Forest model achieved the highest AUC (0.627 ± 0.061), while the AdaBoost model recorded the highest overall accuracy (0.614 ± 0.023). Cerebellar and putamen volumes tended to be higher in the TTH group; however, these differences did not remain statistically significant after false discovery rate (FDR) correction. CONCLUSION: Volumetric MRI alone was insufficient for clinically meaningful differentiation between migraine and TTH; future studies should evaluate multimodal models integrating structural, clinical, and functional data.