In the Journal of Applied Physics, researchers developed a method to identify aortic valve stenosis using complex network analysis that is accurate, simple to use, and low-cost.
Aortic valve stenosis occurs when the aortic valve narrows, constricting blood flow from the heart through the artery and to the entire body. In severe cases, it can lead to heart failure. Identifying the condition can be difficult in remote areas because it requires sophisticated technology, and diagnoses at early stages are challenging to obtain.
“Many rural health centres don’t have the necessary technology for analysing diseases like this,” said author M.S. Swapna, of the University of Nova Gorica and the University of Kerala. “For our technique, we just need a stethoscope and a computer.”
The diagnostic tool works based on the sounds produced by the heart. The organ creates a “lub” noise as it closes the mitral and tricuspid valves, pauses as ventricular relaxation occurs and the blood fills in, then makes a second noise, “dub,” as the aortic and pulmonary valves close.
Swapna and her team used heart sound data, collected over 10 minutes, to form a graph. This was then split into sections, with each part representing with a node on the graph. If the sound in that portion of the data was similar to another section, a line was drawn between the two nodes.
In a healthy heart, the graph showed two distinct clusters of points, with many nodes unconnected. In contrast, a heart with aortic stenosis contained many more correlations and edges.
“In the case of aortic stenosis, there is no separation between the ‘lub’ and ‘dub’ sound signals,” explained Swapna.
The researchers used machine learning to examine the graphs and identify those with and without disease, achieving a classification accuracy of 100%. Their method takes the correlation of each point under consideration, making it more accurate than others that only consider the strength of the signal, and it does so in less than 10 minutes. As such, it could be useful for early-stage diagnoses.
So far, the method has only been tested with data, not in a clinical setting. The authors are developing a mobile application that could be accessed worldwide. Their technique could also be used to diagnose other conditions.
“The proposed method can be extended to any type of heart sound signals, lung sound signals, or cough sound signals,” said Swapna.
Source: American Institute of Physics
