Research On The Intelligent Predictive Risk Of Stroke Based On Speech Features

Stroke is a disease of sudden cerebral vascular blockage or rupture.In Chinese adults,stroke is the first cause of death or disability.At the same time,stroke not only has a high fatality rate and disability rate,but also easy to recur.At present,in medicine,the treatment of stroke is mainly used after the occurrence of stroke,the treatment method lacks timeliness,which leads to the unsatisfactory treatment effect of stroke.For the prevention and treatment of stroke,the medical community generally believes that prevention is the best treatment for stroke diseases.In view of the problems encountered in the prevention and treatment of stroke diseases,this paper proposes an intelligent stroke risk prediction method based on speech features.This method mainly studies the following aspects:(1)In this paper,speech features are used as feature sets for intelligent stroke prediction.Before the onset of stroke,the most obvious symptoms of patients are poor speech and difficulty,accompanied by dyslexia and other symptoms.In this paper,we extract speech features from the original speech data by acquiring the voice of stroke patients and normal people.In speech features,Mel Frequency Cepstrum Coefficient features can well characterize the speech characteristics of the original speech.MFCC speech feature is used as the feature set of stroke intelligent prediction.(2)In this paper,we use the relevant data preprocessing methods to preprocess the speech features.In the acquisition of MFCC speech features,there are many problems,such as missing values and redundant features.In this paper,we use synthetic minority oversampling technique,principal components analysis and median substitution to pre-process the speech features to solve the problems of feature data.(3)In this paper,we use a variety of modeling methods to predict the risk of stroke intelligently.After the pretreatment of MFCC speech features,the training set of MFCC speech features is used to model and train the logit regression and the test set of MFCC speech features is used to evaluate the model.Then,the training set of MFCC speech features is used to model and train the convolutional neural network and the test set of MFCC speech features is used to evaluate the model.Finally,the test set of MFCC speechfeatures is used to evaluate the model.The training set of deep feature acquired by product neural network is used to model and train logistic regression machine,and the test set of deep feature is used to evaluate the model.The experimental results show that in terms of prediction precision,the convolution neural network is used to establish prediction model,and the prediction precision is 80%.Logit regression was used to establish the prediction model,and the prediction precision was56%.Using convolution neural network and logical regression to establish the prediction model,the prediction precision is 81%.