Spinocerebellar Ataxia Type 12 classification through Structural MRI: Insights from Regional Brain Volume and Cortical Thickness Analysis

Spinocerebellar ataxia (SCA) type 12 is a neurodegenerative hereditary disorder of progressive nature with a particularly high prevalence in India. This investigation evaluates the potential application of structural magnetic resonance imaging (sMRI) data in the automated classification of SCA12. We acquired 32 SCA12 patients and 28 healthy controls for binary classification. Feature extraction was performed via three brain atlases - Automated Anatomical Labeling (AAL3), Neuromorphometrics atlas, and Destrieux atlas. Subsequently, the ReliefF feature selection algorithm was employed to select the most informative features for analysis. Finally, two machine learning models, a linear Support Vector Machine (SVM) and radial basis kernel SVM (RBSVM), were implemented. Both models underwent training with a forward feature selection approach combined with leave-one-out cross-validation. The model achieved an optimal classification accuracy of $88.33 \%$ when features were constructed using gray matter volumes using the AAL3 atlas. It was observed that gray matter, white matter, and cortical thickness are better biomarkers than CSF volume for differentiating SCA12 patients. Additionally, biomarkers identified via feature selection may enhance the efficiency of clinical diagnosis.