Research And Implementation Of Abnormal Heart Sound Recognition Algorithm

Remote auscultation makes the cardiovascular disease patients access to the medical service without going out,which greatly reduces the patient’s medical cost.Abnormal heart sound recognition is an important part of the remote auscultation and it is of great significance for the implementation of initial patient screening and long-term home monitoring.Focusing on remote auscultation,this article aims to research and develop a wireless electronic stethoscope with noise reduction function and the abnormal heart sound recognition algorithm.The main contents of this article include:1)Noise reduction of heart sound: Due to the periodic misalignment of the adaptive filter’s parameters caused by the high amplitude of the first and second heart sounds,a pre-process method is introduced to improve the conventional dual microphone noise reduction algorithm,which can decline the parameters’ misalignment and increase the signal-to-noise ratio of the heart sounds.2)Heart sound segmentation: According to the interference of heart sound murmur,the calculation of the heart sound’s Shannon envelope is improved and the locations of the first and second heart sounds are determined by using second-order difference and logistic regression classifier.The segmentation algorithm’s accuracy of normal and abnormal heart sounds are 87.40% and 77.01% respectively.3)Abnormal heart sound recognition: A logistic regression classifier is used to classify the normal and abnormal heart sounds,using the similarity of the systolic and diastolic signal in time and frequency domain as features.The algorithm achieves an accuracy of 87.50% and the sensitivity and the specificity are 90.74% and 80.77% respectively.4)Hardware design: Develop a wireless electronic stethoscope with dual microphone noise reduction function based on the STM32 platform,including heart sound collection and wireless data transmission.Some corresponding applications are also implemented.This article examines the effectiveness of the adaptive noise reduction,the heart sound segmentation and the abnormal heart sound recognition algorithm through some experiments.All algorithms mentioned above can be implemented in software and achieve the expected result.