Skip to Main Content
We propose a partial directed coherence (PCD) method based on a sparse multivariate autoregressive (mAR) model to investigate patterns of information flow in electroencephalography (EEG) recordings in Parkinson's disease (PD) patients performing a visually-guided motor task. The use of a sparsity constraint on the mAR matrix addresses issues such as sample size, model order selection and number of parameters to be estimated, particularly when the number of EEG channels used is large and the window size is small in order to capture dynamic changes. The proposed PDC-based information flow analysis demonstrated distinctly altered patterns of connectivity between PD patients off medication and healthy subjects, particularly with respect to net information outflow from the left sensorimotor (L Sm) region, which might indicate excessive spreading of activity in the diseased state. Disrupted patterns of connectivity in PD were partially restored by levodopa medication. In addition, PDC-based analysis proved to be more sensitive to temporally-dynamic connectivity changes as compared to traditional spectral analysis, which might be influenced primarily by large-scale changes. We suggest that the proposed sparse-PDC method is a suitable technique to investigate altered connectivity in Parkinson's disease.