A deep learning-based image processing pipeline for deep-brain structures: Application to atypical Parkinsonian disorders

Early diagnosis of atypical Parkinsonian disorders (APD) remains a major clinical challenge due to overlapping symptoms with Parkinson’s disease (PD). Segmentation of deep-brain structures from MRI may provide supportive imaging biomarkers. Here we present a fully automated, deep learning–based image processing pipeline to support differential diagnosis of APD. A region-based U-net segments 12 deep-brain structures by dividing MRI volumes into anatomically targeted regions, thereby reducing GPU demands and training time while maintaining competitive segmentation accuracy. Segmentation masks and volumetric features are then combined with T1-weighted MRI in a hybrid classification framework to explore subtype differentiation in APD. The proposed approach provides a computational framework that may contribute to improved differential classification of APD.