Design and development of electronic stethoscope for early screening of valvular heart disease prediction

Abstract

The significant human population of the world is suffering from valvular heart diseases and dies due to the lack of a simple predictive diagnosis system. Identifying the abnormalities in heart sound needs excellent auscultation skills and experience. Thus an electronic stethoscope is designed and developed using different microphonic heads. In this research work, hardware development of an electronic stethoscope interfaced with raspberry pi 4B and software development of the proposed CNN-based deep learning model is carried out. The Bluetooth-enabled electronic stethoscope is used to auscultate the heart sound analyzed with a developed deep learning Convolutional Neural Network-based Efficient Network model. The developed CNN model is designed with classifiers to predict valvular diseases accurately. The analysis result is almost in real-time processed and stored in the Cloud. The design provides a better way of studying PCG signal analysis, which eventually reduces the cost and makes the system compact. The proposed model has been trained through a standard and validated heart sound bank of normal and abnormal diseases and predicts the abnormality with accuracy. The proposed modified Efficient Net - B3 model scored an accuracy of 99.35 ± 0.34% on the test dataset with a sensitivity of 98.84 ± 0.07% and specificity of 98.23 ± 0.52%. The Selenium Web Driver tool with Google Drive API is used to automate the web application PCG signal analysis. Finally, the SQLite database has been used as a back-end server to store the patient record. The system is low-cost and portable, with data remotely accessible and tested with volunteers. The developed system can be used in rural areas where there is a lack of medical facilities exists and can be used to initiate primary screening of valvular diseases.