Fig. 1: Methodology pipeline.

a PD patients underwent 6 weeks of treadmill training. b T1/fMRI acquisition during resting state before and after the training. c fNIRS acquisition during normal and dual-task walking tasks. d EC analysis with TPDC for fMRI. e EC analysis with TPDC for fNIRS. f SVM classification based on rs-fMRI connectivity as the predictors. g SVM regression based on rs-fMRI connectivity to predict fNIRS connectivity. h All the feature combinations pass through nested cross validation. i Select the optimal SVM model based on the model performance. PD Parkinson’s disease, fNIRS functional near infrared spectroscopy, EC effective connectivity, fMRI functional magnetic resonance imaging, SVM support vector machine, TPDC time resolved partial directed coherence.