A novel method for discrimination between innocent and pathological heart murmurs

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Cardiology Signal Processing

Background:

• Distinguishing innocent murmurs (IM) from pathological murmurs, such as those in aortic stenosis (AS), is clinically significant.
• The subtle acoustic similarities between IM and mild AS murmurs present a diagnostic challenge.
• Conventional machine learning methods may not optimally capture the temporal characteristics of heart murmurs.

Purpose of the Study:

• To introduce and evaluate a novel Growing Time Support Vector Machine (GTSVM) for murmur classification.
• To enhance the characterization of innocent murmurs by emphasizing early signal components.
• To compare the diagnostic performance of GTSVM against conventional Support Vector Machine (SVM).

Main Methods:

• Development of a GTSVM algorithm focusing on early signal phases characteristic of innocent murmurs.
• Analysis of patient groups including normal individuals (NM), those with innocent murmurs (IM), and mild to severe aortic stenosis (AS).
• Performance evaluation using repeated random sub-sampling, comparing GTSVM with conventional SVM.

Main Results:

• The GTSVM achieved a mean classification rate of 88% and sensitivity of 86%.
• The conventional SVM achieved a mean classification rate of 84% and sensitivity of 83%.
• Statistical analysis confirmed a significant performance improvement with the GTSVM compared to the SVM.

Conclusions:

• The proposed GTSVM method offers superior performance in classifying heart murmurs compared to standard SVM.
• Tailoring the machine learning approach to emphasize early signal characteristics improves discrimination of innocent murmurs.
• GTSVM holds promise for more accurate non-invasive cardiac auscultation analysis.