Computational Modeling of Auscultation-based valve monitoring

Cardiac auscultation is the practice of listening to heart sounds to make a medical diagnosis and has historically been used to detect a wide range of heart valve diseases (HVDs). It is advantageous in that it is inexpensive, non-invasive and can be performed anywhere. However, the development of sophisticated angiography techniques, and the relatively poor repeatability of auscultation-based diagnosis, due to a heavy reliance on individual proficiency, have led to a decline in this clinical skill among physicians. The issue is further compounded by a lack of practical training offered in med schools.

While modern angiography techniques like 4D CT or echo are more sensitive and specific, they are expensive, invasive and involve either radiation or potentially toxic contrast agents, which make it difficult to employ them on a routine basis. On the contrary, auscultation is a safe and practical technique which offers a wealth of information and can potentially aid in the management of patients with HVDs. Its major limitation of accuracy subjectivity can be alleviated with the help of machine-learning based pattern recognition and possibly enable a new safe, at-home diagnostic technique.

In a paper I published in Frontiers in Physiology, I show how the fluid-dynamics associated with healthy and stenotic aortic valves result in characteristic sounds (listen to the sounds!) familiar to physicians and present a supervised learning algorithm which can recognize these characteristics and enhance the accuracy of auscultation-based diagnosis.