Intelligent Auscultation System Based On Convolutional Non-negative Matrix Factorization

At present,cardiovascular disease is already the first human disease,and respiratory disease has also become one of the serious threats to human health.The stethoscope is the most commonly used medical device for clinical diagnosis of these two types of diseases,but the traditional stethoscope has low accuracy and cannot save sound signals.The auscultation process relies on the subjective experience of the doctor.It is subjective and strong.Foreign high-end electronic stethoscopes are expensive and cannot solve cardiopulmonary sounds.The key issue of mutual interference cannot automatically extract the pathological features of heart sounds to assist clinical diagnosis.In order to overcome this difficulty,this paper uses non-negative matrix factorization and convolutional non-negative matrix factorization,proposes two cardiopulmonary sound separation methods and studies and designs an intelligent auscultation system to improve the quality of auxiliary auscultation.The main research content of this article:1)Aiming at the characteristics of the timing structure of heart and lung sounds,this paper proposes a recursive sparse representation of heart and lung sounds separation method based on non-negative matrix decomposition.This method uses time-spectral matrix transformation and non-negative matrix decomposition to mine the frequency domain information and timing information of clean heart sounds and lung sounds to construct a recursive heart-lung sound dictionary.Based on the dictionary,sparse representations of heart and lung sounds are obtained,and finally heart and lung sound signals are reconstructed in the time domain to achieve separation of heart and lung sounds.2)In view of the fact that convolutional non-negative matrix decomposition can capture the time-series characteristics of repeated complex patterns of cardiopulmonary sounds,this paper proposes a recursive sparse representation of cardiopulmonary sound separation based on convolutional nonnegative matrix decomposition.This method uses convolutional non-negative matrix decomposition to mine frequency domain information and timing information with timing hysteresis of clean heart sounds and lung sounds,constructs a super-complete dictionary,and alternately optimizes the sparseness of the target subject ’s time spectrum during the decomposition process The expression matrix and the time-series activation matrix converge to the objective function,reorganize the frequency domain structure of heart sounds and lung sounds,and finally reconstruct the heart sound and lung sound signals in the time domain to achieve the separation of heart and lung sounds.3)In order to improve the quality of auscultation,an intelligent auscultation system was developed to provide auxiliary diagnosis.The system consists of a stethoscope,a mobile phone app and a back-end server.The signal is collected by a stethoscope,Bluetooth is transmitted to the mobile phone,and the mobile phone is sent to the backend through 4g or WiFi,and the processing returns the auxiliary inspection result.The system has functions such as user information management,health parameter management,waveform display and intelligent normal / abnormal classification auxiliary diagnosis.