Deep Learning Based Web Application for Cardiovascular Abnormality Detection via Heart Sound Analysis Pekanbaru
Keywords:
Heart Sound, Phonocardiogram, Mel Spectrogram, Convolutional Neural Network, Resnet18, Web ApplicationAbstract
Cardiovascular diseases remain a leading contributor to global mortality and early identification is essential for timely intervention. Conventional auscultation is inexpensive but subjective and depends on clinical expertise. This study proposes a deep learning based web application that classifies heart sounds as normal or abnormal to support rapid screening from phonocardiogram recordings. Recordings from the PhysioNet Computing in Cardiology Challenge 2016 dataset were denoised using a Butterworth bandpass filter, segmented into cardiac cycles, and transformed into Mel spectrogram images. Two transfer learning backbones, VGG16 and ResNet18, were fine tuned for binary classification. Using a 70 percent training split with 15 percent validation and 15 percent testing, the best model, ResNet18, achieved 94.50 percent accuracy with 95.63 percent precision, 96.73 percent recall, and 96.17 percent F1 score on the held out test set. The web interface enables users to upload heart sound recordings and receive probability based predictions in real time. Internal usability testing indicated very good overall acceptance with an average score of 89.5 percent. The results suggest that a lightweight web deployment of spectrogram based deep learning can provide accessible decision support for early cardiovascular abnormality screening, particularly in resource constrained settings.
References
Roth, G. A., et al. (2020). Global burden of cardiovascular diseases and risk factors, 1990 to 2019: Update from the GBD 2019 study. Journal of the American College of Cardiology, 76(25), 2982 to 3021. https://doi.org/10.1016/j.jacc.2020.11.010
Clifford, G. D., Liu, C., Moody, B., Lehman, L. H., Silva, I., Li, Q., & Johnson, A. E. W. (2016). Classification of normal and abnormal heart sound recordings: The PhysioNet Computing in Cardiology Challenge 2016. Computing in Cardiology, 43, 1 to 4. https://doi.org/10.22489/CinC.2016.179-154
Li, F., et al. (2020). Classification of heart sounds using convolutional neural network. Applied Sciences, 10(11), 3956. https://doi.org/10.3390/app10113956
Li, J., et al. (2022). Improved MFCC and deep residual learning for heart sound classification. Frontiers in Physiology, 13, 1084420. https://doi.org/10.3389/fphys.2022.1084420
Chen, W., Zhou, Z., Bao, J., Wang, C., Chen, H., Xu, C., Xie, G., Shen, H., & Wu, H. (2023). Classifying heart sound signals based on CNN trained on MelSpectrum and Log MelSpectrum features. Bioengineering (Basel), 10(6), 645. https://doi.org/10.3390/bioengineering10060645
He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 770 to 778). https://doi.org/10.1109/CVPR.2016.90
Simonyan, K., & Zisserman, A. (2015). Very deep convolutional neural networks for large scale image recognition. arXiv:1409.1556.
