Research On Diagnosis Algorithms Of Parkinson’s Disease Based On Speech Features

Parkinson’s disease is a degenerative disease of the nervous system,which occurs in people over 50 years old and has the characteristics of difficulty to be diagnosed and irreversible.At present,there is no exact and effective treatment for Parkinson’s disease in the world.The symptoms of Parkinson’s disease are easily confused with physical dysfunction due to age,so there are plenty of missed diagnosis and misdiagnosis.The earlier the medical intervention for Parkinson’s disease,the better the effect of slowing the progress of Parkinson’s disease,so a new method for the diagnosis of Parkinson’s disease that is fast,safe,convenient,and accurate is very important for the prevention and treatment of Parkinson’s disease.The current new diagnosis methods for Parkinson’s disease include diagnosis of Parkinson’s disease based on hand-drawn features,gait analysis and facial expression analysis.These studies have achieved certain results,but there are also problems such as complicated process detection and physical damage to the patient’s body,which is not easy to carry out for plenty of Parkinson’s disease detection.More than 90% of Parkinson’s patients have different pronunciations from normal people,so speech can be used as a characteristic marker to diagnose Parkinson’s disease.The diagnosis of Parkinson’s disease based on voice features has the advantages of speed,safety and high accuracy.It can detect Parkinson’s disease on a large scale,so it has very important research significance and practical value.This research aims at detecting Parkinson’s disease with speech features,starting with speech data collection and analysis and feature classification models,the following studies are carried out:(1)Some hybrid models composed of three different feature dimensionality reduction algorithms and three different machine learning classification algorithms were trained and classified by the Parkinson’s speech data set in the UCI database,thus comparative analysis of the classification results of different combination models,and compared and analyzed the classification results of different combination models,and got the best combination classification model.(2)Most of the Parkinson’s classifications based on speech features improves classification accuracy from two aspects: feature selection and classifier optimization,but they does not pay attention to the analysis of the overall samples.This paper proposes a classification model that combines the sample selection method(NNGIR)with three classifiers.The overall performance of the model is improved by removing noise samples and performing samples reduction.The model proposed in this paper has achieved good results,not only improving the accuracy of classification,but also reducing the running time of the algorithm,thus comprehensively improving the performance of the model.(3)The existing public Parkinson’s speech databases are all from foreign languages,and there are no Chinese characteristics of Parkinson’s patient classification research.In this paper,the speech of 20 Chinese Parkinson’s patients were collected for speech signal analysis,speech feature extraction and classification research.The results prove that the classification model trained from the speech data set in the UCI database is not suitable for the classify Parkinson’s Chinese features.To get an accurate Parkinson’s classification model of Chinese features,it is also necessary to enrich Chinese Parkinson’s speech data.