A Novel Signals and Features Fusion Method Based on Fisher Vector for Severity Assessment in Parkinson's Disease

Parkinson's disease (PD) is a progressive neurode-generative disorder that requires accurate severity assessment to facilitate proper medication and interventions. This study proposes a novel method for PD severity assessment by fusing bilateral surface electromyography (sEMG) signals and features using Fisher vector. The sEMG signals were measured from the tibialis anterior (TA) and lateral gastrocnemius (GL) muscles in 28 PD patients. Preprocessing was performed to remove industrial frequency noise and low-frequency trend, followed by feature extraction using a customized feature generator that included time-domain and frequency-domain features. The extracted features and preprocessed sEMG signals were then vectorized using the Fisher vector and fused as input to the classifier. Our results demonstrated that considering bilateral features is more favorable than unilateral features for the PD severity assessment. Moreover, the proposed method significantly enhanced the classification accuracy, with a remarkable classification accuracy of 96.00% achieved through the combination of our proposed method with a decision tree classifier.