Research On Some Problems For Computer-aided Heart Disease Diagnosis

The heart disease automatic diagnose which mainly consists of three aspects:the physiological signal preprocessing, the feature extraction and the automatic diagnosis of disease is an important part of computer-aid medical diagnosis. Firstly, the ECG (Electrocardiograph, ECG) signal processing algorithms are studied. Secondly, the feasibility and reliability of automatic heart beat classification model which only used the ECG signals was researched. Lastly, the feasibility and reliability of the heart disease automatic diagnose model used medical knowledge (such as physiological data) was researched. The main works are as follows:(1) The processing algorithms for ECG were researched which consists of the preprocessing algorithms and the feature waves detection algorithms. According to the deficiencies of mathematic morphological derivative (MMD) algorithm in feature waves detection, the conception of double-layer mathematic morphological derivative (DSMD) was proposed, and then DSMD combined with the local feature conversion was applied to feature waves detection. Compared to the MMD algorithm, the algorithm proposed in this paper improves the accuracy of detection.(2) A4-class automatic heartbeat classification model was built in this paper which used ECG signal and the C-SVM classifier. On basis of the feature waves location of ECG signal,42features were extracted from ECG signals included the time domain, the wavelet domain, the high order statistics domain as the inputs of multi-classification C-SVM model. The high order statistics of wavelet coefficients were substituted for wavelet coefficients, through which the features were simplified and the accuracy was improved. In order to improve the robustness of the model, the PSO algorithm was used to optimize the parameters of the model. The second ballot was proposed in decision-making phase. The accuracy of the model was reached up to98%.(3) A valid heart disease automatic diagnose model was set up used the Cleveland database which contains13physiological data, and the genetic algorithm was used to simplified the features. The results shows when iteration reach up to1000only4features was left to build the model which had the same accuracy of the model used the entire features. This4features show great important in the heart disease automatic diagnosis.