Construction Of Disease Classification Tree Based On Voice Diagnosis

With the gradual application of computer technology in the medical field,using computer to analysis and diagnosis patient's voice is low cost and easy to operate.Voice diagnosis has the advantages of simple,rapid,no damage and no pain,more and more researchers start to focus on relevant researches.The classes of diseases that affect vocalization mainly include neurological,pulmonary and vocal organ.This paper focuses on the typical diseases of the three major classes,analyzed the differences between pathological voices and health voices.In this paper,the feature extraction,feature selection and feature fusion are studied for the binary classification between different classes,by decomposing the multi-classification problem into binary classification problems,constructs the disease classification tree based on voice diagnosis.First of all,by analyzing the differences between pathological voices and healthy voices,the voice features based on prior knowledge and dictionary learning are extracted.For the imbalance problem of samples among different classes,the under-sampling and over-sampling methods are designed respectively for heathy samples and lung cancer samples.The experiments show that the proposed method effectively mitigates the misclassification of minority classes caused by imbalance of samples.Then,to solve the problem of high dimension of voice features,feature selection methods are adopted to eliminate redundant features,and the best feature subset is selected to effectively reduce the feature dimension while ensuring classification accuracy.Based on the optimization of single voice features,analyzing the differences of binary classification among different pronunciations,the multi pronunciation fusion methods on feature level and decision level are proposed to make full use of different pronunciations features,which effectively improves the accuracy of classification.Finally,to solve the problem of the misclassification of minority classes in the multi-classification,this paper decomposes the multi-classification problem into binary classification problems.Comparing the differences of binary classification among the four classes,referencing to the experiments of the feature selection and feature fusion,two disease classification tree models based on MVM and OVR are proposed.Compared with the random forest model,the experiments show that,the accuracy of the disease classification tree based on MVM increased 5.22%.For the disease classification tree based on OVR,the accuracy of lung cancer increased to 68.82%,and G-mean reached 69.07%,the experiment results prove the effectiveness of the algorithm.