EEG analysis of Parkinson's disease using time–frequency analysis and deep learning

Abstract

This study proposed two EEG analysis methods for diagnosis and monitoring of Parkinson’s disease. By combining time–frequency analysis with deep learning, tunable Q-factor wavelet transform with deep residual shrinkage network (TQWT-DRSN) and the wavelet packet transform with deep residual shrinkage network (WPT-DRSN) are applied to classify four kinds of clinical sleep EEG data in Shaanxi Provincial People's Hospital, which included different types of diseases, Parkinson's disease, REM sleep disorder, Parkinson's disease with REM sleep disorder, and select a group of normal people as a control group. For 2-class classification tasks, the accuracies achieved 99.92% on Parkinson’s disease predicting. In 3-class classification and 4-class classification tasks, the accuracies of WPT-DRSN are 97.81% and 92.59%, which are higher than 95.20% and 90.46% of TQWT-DRSN. The results showed that methods proposed in this paper can be used to monitor the condition of Parkinson's disease, and has certain guiding significance for the early diagnosis, effective treatment and prognosis judgment of Parkinson's disease.