Enhancing heart disease diagnosis through precise heart sound including infrasound

Abstract

Recent research has been pushing the frontier of diagnosing cardiovascular diseases, which have a high mortality rate globally. The need for swift responses to acute diseases has underscored the importance of continuous health monitoring and the development of monitoring diagnostic technology. Our work focused on an overlooked aspect—the microphones used to measure heart sounds. Existing ones, particularly capacitive MEMS types, measure only heart sounds primarily within the audible frequency range, potentially limiting the diagnostic accuracy. To overcome these limitations, we used microphones capable of measuring down to infrasound areas. We also designed an experimental environment to capture the precise heart sounds. The crux of our approach was using the full information of heart sounds through infrasound measurements. A convolutional neural network-based deep learning model was employed for algorithm development and validated using the PhysioNet 2016 CinC’s open-source heart sound data. We conducted an experiment on the impact of low-frequency components on heart sound diagnosis using a new infrasound dataset measured at Seoul St. Mary's Hospital of the Catholic University of Korea. Accuracy and sensitivity, respectively, improved on average by 2% and 4%, showing that datasets including low-frequency components generally performed better.