基于最小二乘法椭圆拟合在心音识别中的应用研究

Least Squares Method-Based Ellipse Model for Heart Sound Classification Research

摘要: 心音识别是诊断心脏疾病的重要方法之一，其识别精度使医护人员为患者提供更加准确的治疗手段。鉴于此，为提高识别精度，本研究提出了基于支持向量机(SVM)对心音信号特征生成分类边界曲线，考虑到分类边界曲线的形状，建立了基于分类边界曲线的最小二乘法椭圆模型，优化了心脏疾病诊断的复杂性。其算法步骤为：首先根据心音信号的频率特征生成基于支持向量机的分类边界曲线，然后建立基于分类边界曲线的最小二乘法椭圆模型对分类边界曲线进行椭圆拟合。为验证在心音识别应用中椭圆模型的有效性，对在线心音数据库和临床心音的实验分析，即对188例房颤、181例主动脉瓣返流、257例二尖瓣返流、325例正常声音和150例肺动脉瓣狭窄心音进行检测，其分类精度分别为91.7%、98.8%。98.4%、99.8%和98.7%，证明其分类具有较高的精度。

Abstract: Heart sound classification is one of the important methods to diagnose heart disease, and its iden-tification accuracy enables medical staff to provide more accurate treatment for patients. In view of this, in order to improve the recognition accuracy, this study proposes to generate classification boundary curve based on support vector machine (SVM). Considering the shape of classified boundary curve, a least square elliptical model based on classified boundary curve is established, which optimizes the complexity of heart disease diagnosis. Firstly, the classification boundary curve based on support vector machine is generated according to the frequency characteristics of the heart sound signal, and then the least square elliptical model based on the classification boundary curve is established to perform ellipse fitting on the classification boundary curve. To verify the effectiveness of elliptical model in heart sound recognition, an experimental analysis of online heart sound database and clinical heart sounds was carried out. 188 cases of atrial fibrillation, 181 cases of aortic regurgitation, 257 cases of mitral regurgitation, 325 cases of normal sound and 150 heart sounds of pulmonary stenosis were detected, and the classification accuracy was 91.7%, 98.8%, 99.8%, and 98.7%, proving that its classification has high precision.