Key result
Three-layered neural network achieves 89% accuracy in differentiating normal from abnormal heart sounds.
Why the study?
The rise in cardiovascular diseases necessitates efficient diagnostic methods to differentiate between normal and abnormal heart sounds using phonocardiograms.
Does a deep learning algorithm using phonocardiograms accurately differentiate between normal and abnormal heart sounds?
Does a deep learning algorithm using phonocardiograms accurately differentiate between normal and abnormal heart sounds?
Absolute Event Rate: 89% vs 86%
A three-layered neural network using spectral roll-off feature extraction and up-sampling can detect abnormal heart sounds from phonocardiograms with 89% accuracy.
May support automated phonocardiogram screening; leaves open prospective clinical validation before practice change.
This research study aims to develop an algorithm using phonocardiograms and Deep learning to differentiate between normal and abnormal heart sounds. The increase in cardiovascular diseases (CVD's), primarily caused by unhealthy habits, necessitates efficient diagnosis. The study utilized 1055 data samples, with 833 being normal and 182 abnormal, each lasting 5 seconds. Logistic regression and a three-layered neural network were initially applied, yielding accuracies of 42% and 63%, respectively. By extracting features using spectral roll-off and reapplying the models, accuracies improved to 79% and 86%. Up sampling the data further increased the accuracies to 86% for logistic regression and 89% for the three-layered neural network. The findings demonstrate the potential of audio-based algorithms in diagnosing cardiovascular diseases accurately.
No takes yet. Share an insight, caveat, or question.
Sarada et al. (2023) studied Cardiovascular diseases (abnormal heart sounds) (n=1,055). Three-layered neural network with spectral roll-off and up sampling vs. Logistic regression was evaluated on Accuracy in differentiating between normal and abnormal heart sounds. A three-layered neural network using spectral roll-off feature extraction and up sampling achieved 89% accuracy in differentiating normal from abnormal heart sounds.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: