Machine learning approach effectively discriminates PD from progressive supranuclear palsy: multi-level indices of rs-fMRI

Motivation: Employ machine learning approach based on multi-level indices of rs-fMRI to discriminate Parkinson's disease (PD) from progressive supranuclear palsy (PSP). Goal(s): To investigate the optimal machine learning model and corresponding rs-fMRI based multi-level indices combination. Approach: 58 PD and 52 PSP patients were involved. Multi-level indices of rs-fMRI was applied to discriminate PD from PSP using machine learning approaches, including K-nearest neighbor (KNN), logistic regression (LR), multilayer perceptron (MLP), and support vector machine (SVM). Results: LR and SVM based on multi-level indices of rs-fMRI exhibited significantly superior classification performance than KNN and MLP for distinguishing PD from PSP. Impact: PD and PSP have similar clinical syndrome but were treated differently. Our finding suggested LR and SVM based on multi-level indices of rs-fMRI can effectively differentiate PD from PSP. It would help the treatment selection for PD and PSP patients.