Research Of The Diastolic Of CAD Feature Extraction Based On DTW

Coronary artery disease(CAD) is one of the most important diseases which have long been serious threats to the health of people. And with the improved living standards, high fat intake, lack of exercise and other bad habits make patients younger, and increase the incidences. The "gold standard" of diagnosis of coronary artery disease is coronary angiography, but it’s invasive and expensive, not suitable for routine examination and extensive census. Electrocardiogram, echocardiogram and cardiac radionuclide diagnostics, multi-slice CT coronary were widely used in clinical practice as supplementary noninvasive examination, but because of the different mechanism of non-invasive examination, the performance scope is very limited. Since the development of modern signal processing technology, the traditional heart sound learning attracts more and more attentions. Especially in recent years, rapid development and application of speech signal processing technology opens up prospects for acoustic signal processing.This article studies diastolic signal characteristics extraction techniques of coronary artery diseases, based on dynamic time warping(DTW). It performed some research on coronary artery disease heart sound signal feature extraction and classification determination. Include the following aspects:1. Analysis of research status of heart sound of coronary artery disease.2. Heart sound signal samples collection: Collect apex heart sound signals of CAD, other heart diseases as well as normal healthy person.3. Based on pathological feature of enhanced high frequency of diastolic heart sound signal of coronary artery disease, extract diastolic signal, perform wavelet threshold noise reduction, and study sub-windowed frames, as 64 ms for each length.4. Automatic segmentation for heart sound signals.5. Analyze frequency domain characteristics differences of each frame, collect 180-250 Hz band energy as the feature for each frame, and make dimension as the number of frames.6. Based on the characteristics of coronary artery disease with enhanced high-frequency diastolic heart sounds, but different durations and positions, it applied improved Dynamic Time Warping(DTW) algorithm to calculate feature vectors and matching distances with different dimensions, and structure into same dimension vectors.7. Compare classification results of average distance classification, and the mean vector classification and Support Vector Machine(SVM) classification for specific and non-specific person. Find out the best initial template in the average distance classification by optimizing, and optimize SVM parameters to obtain better results.In this article, it started from the pathological features and conducted comprehensive study signal acquisition, preprocessing, feature extraction and classification analysis for clinical coronary artery disease. Based on the individual differences of heart rates and locations of pathological heart murmurs, it applied DTW to coronary artery disease heart sound signal processing, and use heart sound data collection to three categories of clinical validation, to obtain better classification results. Through experiments and research on each part for the development of clinical coronary artery disease heart sounds lossless diagnostic systems do basic work.