Quantitative Assessment Of Parkinson’s Motor Symptoms Based On Deep Learning

Parkinson’s Disease(PD)is a degenerative disease of the nervous system and the second most common disease in the world.According to the Movement Disorders Society-Sponsored Revision of the Unified Parkinson’s Disease Rating Scale(MDS-UPDRS)is a quantitative assessment of Parkinsonian tremor(0,1,2,3,4),which is crucial for the treatment of Parkinson’s disease.However,due to the influence of environmental and psychological factors on patients,the tremor amplitude fluctuates constantly.In clinical practice,clinicians assess tremor severity in a short period of time,and manual tremor labeling relies on clinician experience.Effective quantitative assessment of the severity of parkinsonism is of great significance for the elderly and their families.In order to solve the above problems,this paper adopts the corresponding method to study the automatic quantitative scoring of Parkinson’s tremor.(1)Automatic quantification of tremor based on wearable inertial sensor and Machine Learning(ML)algorithm is affected by manual labeling by clinicians.This study proposes an automatic label correction method judged by clinicians to improve the quantitative assessment of parkinsonism tremor.In view of the serious overlap of dynamic feature ranges between different severity levels.This paper proposes an outlier correction algorithm(PCA-IQR)based on Principle Component Analysis(PCA)and Interquartile Range(IQR)statistical rules.Fuzzy boundaries between different severity scores are learned to optimize the labels.Then,according to the improved feature vectors,a Support Vector Machine(SVM)with radial basis function kernel is proposed to classify the severity of tremor.The classification models of Radial Basis Function(RBF)support vector machine,K-Nearest Neighbor(KNN)and Linear Kernel support vector machine are compared.Experimental results show that this method has high classification performance and good model generalization ability(accuracy:97.93%,precision: 97.96%,sensitivity: 97.93%,F1-score: 97.94%)for tremor quantitative evaluation.(2)With the advent of deep learning and the improvement of computing power,deep learning and artificial intelligence methods have recently been used in mobile and wearable sensors for feature extraction and classification.The feature extraction process is automatic,thus eliminating the problem that manual feature extraction may miss some important features and improving the overall classification accuracy.The superior performance of deep neural networks largely relies on a large amount of training data to avoid overfitting.Labeled data for real-world time series applications can be limited,and collecting and labeling large amounts of medical data is often difficult.Therefore,data augmentation is essential for the successful application of deep learning models on time series data.In this paper,the data augmentation method is used to solve the problem of data imbalance.In addition,this paper proposes a hybrid deep model,including coupled Convolutional Neural Networks(CNN)and Long Short-Term Memory networks(LSTM)with attention mechanism,it provides reliable classification test results for Parkinson’s tremor assessment.This method can effectively solve the long-term dependence problems such as information redundancy and information loss,and provide optimized feature vectors for sample classification,so as to improve the performance of classifier.At the same time,the dependence of manual feature extraction is reduced,and the classification accuracy of Parkinson’s tremor quantization 5 is 98.32%.