Automated detection of abnormal heart sound signals using Fano-factor constrained tunable quality wavelet transform

Manual interpretation of heart sounds is insensitive and prone to subjectivity. Automated diagnosis systems incorporating artificial intelligence and advanced signal processing tools can potentially increase the sensitivity of disease detection and reduce the subjectiveness. This study proposes a novel method for the automated binary classification of heart sound signals using the Fano-factor constrained tunable quality wavelet transform (TQWT) technique. Optimal TQWT based decomposition can reveal significant information in subbands for the reconstruction of events of interest. While transforming heart sound signals using TQWT, the Fano-factor is applied as a thresholding parameter to select the subbands for the clinically relevant reconstruction of signals. TQWT parameters and threshold of the Fanofactor are tuned using a genetic algorithm (GA) to adapt to the underlying optimal detection performance. The time and frequency domain features are extracted from the reconstructed signals. Overall 15 unique features are extracted from each sub-frame resulting in a total feature set of 315 features for each epoch. The resultant features are fed to Light Gradient Boosting Machine model to perform binary classification of the heart sound recordings. The proposed framework is validated using a ten-fold cross-validation scheme and attained sensitivity of 89.30%, specificity of 91.20%, and overall score of 90.25%. Further, synthetic minority over-sampling technique (SMOTE) is applied to produce balanced data set which yielded sensitivity and specificity of 86.32% and 99.44% respectively and overall score of 92.88%. Our developed model can be used in digital stethoscopes to automatically detect abnormal heart sounds and aid the clinicians in their diagnosis. (c) 2020 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier B.V. All rights reserved.

Automated summary

Manual interpretation of heart sounds is unreliable, but automated systems incorporating AI and signal processing tools can improve disease detection sensitivity and reduce subjectivity. A novel method using TQWT for automated binary classification of heart sounds achieved high accuracy, validated through cross-validation and SMOTE for balanced data sets. This developed model can be integrated into digital stethoscopes to assist clinicians in diagnosing abnormal heart sounds.