Quantitation And Screening Of Bradykinesia In Parkinson’s Disease Using Motion Capture

Parkinson’ s disease(PD)is a common neurodegenerative disease in the aged populations,often accompanied by varying degrees of motor and non-motor symptoms.Bradykinesia is one of the main clinical manifestations of Parkinson’s disease,and also serves as a main basis for clinical screening and diagnosis.However,the evaluation results of the scale mainly depend on doctors’ personal perception and clinical experience,with varying degrees of subjectivity and individual differences.Therefore,quantitative evaluation of motor symptoms in PD patients is a difficult clinical challenge in PD diagnosis and treatment.Based on the optical 3D motion capture device,this paper deeply analyzed the bradykinesia symptoms of Parkinson’s patients by capturing the fine movements of the upper and lower limbs of Parkinson’ s patients.In view of the highprecision characteristics of the motion capture system,denoising,preprocessing,and extraction of relevant features were performed on the recorded original signals,and a series of quantitative evaluation models for Parkinson’s bradykinesia symptoms were established.The main contents of this paper are summarized as follows:(1)A quantitative screening system based on high-precision optical motion equipment were designed and constructed,the corresponding experimental paradigm for Parkinson’s patients were designed,and a bradykinesia database of multiple subjects was established for further quantitative evaluation.(2)The physiological and clinical characteristics of bradykinesia were deeply explored.Based on the criteria required by UPDRS scale,features were extracted manually from motion capture data.Feature extraction was mainly based on amplitude,speed and smoothness,which could be distinguished between healthy subjects and Parkinson’s patients with different degrees of severity.(3)Various supervised learning methods were used to establish a quantitative assessment model for the binary screening between Parkinson’s patients and healthy subjects and for the classification of Parkinson’s patients with different severity.In view of the class imbalance in the experimental data,the CS-SVM model was discussed to have a better effect than the undersampled and oversampled methods.The experimental results proved the effectiveness of manual feature extraction and the accuracy of quantitative evaluation.(4)Furthermore,through deep learning,it is discussed that data augmentation and end-to-end LSTM and GRU network methods can improve accuracy of the classification and optimize the evaluation effect.The final experimental results prove that the data processing,features extraction and learning methods in this study are capable of distinguishing different degrees of bradykinesia of the experimental subjects,and verify the feasibility and reliability of the optical motion capture for the quantitative assessment of Parkinson’ s disease,which have great practical significance.