Parkinson's Disease Diagnosis And Prediction Based On Hand Drawing And Speech

Parkinson's disease,also known as tremor paralysis,is the second largest neurological disease after alzheimer's disease.The traditional examination method for Parkinson's disease is to determine the unified Parkinson's disease rating scale.The diagnosis process needs to be completed by both doctors and patients,which has a certain subjectivity.At the same time,due to the inconvenience of Parkinson's patients,traditional diagnosis process not only consumes the time and energy of patients,but also occupies increasingly tense doctors and therapeutic resources.In this situation,the realization of early diagnosis and prediction of Parkinson's disease is of great significance for the treatment of Parkinson's disease.The early onset of Parkinson's disease is characterized by different degrees of speech disorder,static tremor,muscle stiffness and other motor symptoms.Therefore,this dissertation tries to use the hand tremor information and voice injury information of Parkinson's patients to realize the early diagnosis and prediction.The specific research work is as follows:Firstly,this dissertation introduces the relevant theoretical basis and technology,such as,the image processing technology for hand-drawn pictures preprocessing and the speech processing technology for speech preprocessing.At the same time,the speech feature extraction algorithms,including linear and nonlinear feature extraction algorithms,are introduced.What's more,the UPDRS regression analysis method and multi-task learning method are introduced,and the deep learning technology is also introduced.Secondly,this dissertation carries out the design of Parkinson's disease diagnosis scheme based on hand-drawn images.Through data analysis,it is found that the hand dysfunction of Parkinson's disease patients has a great impact on the hand-drawn images of patients.The traditional diagnosis method of Parkinson's disease based on hand-drawn pictures usually extracts relevant features from the hand-drawn pictures and then applies the traditional classification algorithm to realize the diagnosis of Parkinson's disease.Due to the relatively small number of features extracted from handdrawn pictures,the accuracy of the training model in the diagnosis of Parkinson's disease is very low,so the traditional diagnosis methods of Parkinson's disease usually have poor diagnostic performance.To solve this problem,deep learning tools is introduced in this dissertation and a convolutional neural network model is built to realize the early diagnosis of Parkinson's disease.Experimental results show that the CNN network established in this dissertation can effectively improve the accuracy of Parkinson's disease diagnosis.Finally,a prediction scheme of Parkinson's disease level based on speech is proposed.In this dissertation,by analyzing the speech data of Parkinson's patients published by UCI,the implied information of the speech data set of Parkinson's patients are further studied.The traditional speechbased Parkinson's disease prediction method is a single task.Due to the relevant information between tasks cannot be utilized,traditional prediction methods usually have poor prediction performance.Based on this,this dissertation proposes a multi-task regression algorithm to predict the UPDRS of Parkinson's disease.Experimental results show that the multi-task regression algorithm proposed in this dissertation can effectively improve the prediction accuracy of UPDRS for Parkinson's patients.