Automatic Heart Sound Analysis

Researchers in Tokyo have proposed a new method of interpreting heart sounds that does not rely on any degree of segmentation for analysis. While is it relatively easy to determine the presence of a severe heart murmur, traditional heart sound analysis requires segmentation to properly analyze and classify the murmur. Severe murmurs often distort fundamental heart sounds and prevent their segmentation, impeding any analysis. The method proposed in the article circumvents this problem by eliminating the need for segmentation of heart sounds, instead down-sampling and running an entire segment of raw heart sounds through a complex computational system, which returns a list of individual heart sounds each with either a normal or abnormal classification. The method was primed with a set of “training” sounds acquired over the internet. In addition to not requiring sound segmentation of any kind, the result is a form of automated auscultation. Auscultation performed by medical personnel requires a vast amount of training and experience, and the number of people skilled in auscultation is slowly declining. Such proposed automated auscultation would rectify this problem. Shortcomings of the method include its current ability only to pinpoint exactly when the normal and abnormal sounds appear in the cardio cycle, but I feel that with a much larger sample size the ability of the proposed method to classify and analyze heart issues can only exponentially improve.

This article was of interest to me because we have just reviewed the cardiovascular system in VTPP 435, and complications involving the heart are a major cause of death worldwide. A childhood friend of mine has a slight heart murmur and had a pacemaker implanted a few years back, so any novel research that may improve our understanding or interpretation of heart complications is of particular interest to me.

Link: "http://www.biomedical-engineering-online.com/content/10/1/13"

John Gruetzner

VTPP 435 - 502