Investigate brain volumes with machine learning algorithms to differentiate PD and PSP patients

Motivation: Investigate brain regions extracted with MRI with machine learning approach. Goal(s): The aim of the study was to identify the mainly brain regions involved in both diseases to show the differences between Progressive Supranuclear Palsy and Parkinson's Disease. Approach: Using Freesurfer we extracted morphological data from MRI and we analyzed with a nested cross validation using XGBoost algorithm and a feature selection using SHapley Additive exPlanations. Results: The difference between PSP and PD were in subcortical such as Left Accumbens and Left Cerebellum White Matter and in different cortical thickness with an AUC of 0.87 after feature selection. Impact: This study provides first evidence of alterations in subcortical volume and cortical thickness between Parkinson's disease patients and Progressive Suplanuclear Palsy patients using a rigorous approach combining nested cross validation, XGBoost and SHAP as feature selection.