Parkinson’s Disease Classification Using Machine Learning Approaches and Resting-State EEG

Purpose: Diagnosis of Parkinson’s disease (PD) is generally based on family medical history, physical examination, and response to medication. Objective tools using machine learning algorithms have been developed to aid in PD diagnosis; however, feature extraction is time consuming, computationally intensive, and difficult to implement in clinical settings. This study compared the performance of two methods, namely a support vector machine (SVM) and convolutional neural network (CNN), in the classification of patients with PD based on resting-state electroencephalography (EEG). Methods: In total, 39 patients with PD and 40 healthy controls participated in the experiment. Mean frequency, relative power, coherence, sample entropy, and multiscale entropy were calculated as features. Results: The accuracies of the SVM using 548 selected features and that of the CNN using 2992 extracted features were 88.88% and 98.66%, respectively. The accuracy of the CNN using raw data was 97.54%. Furthermore, the CNN model using features required 1272 s for training and 0.07 s for testing, whereas the CNN model using raw data required 994 s for training and 0.32 s for testing. Conclusion: Our results imply that a CNN model taking raw data as inputs can automatically select the salient features, thereby reducing the required training time and achieving high classification performance.