Related Experiment Video
Updated: Jul 11, 2025
05:32
Retrospective Cardiac Gating with A Prototype Small-Animal X-ray Computed Tomograph
Published on: February 21, 2025
301
An edge-device-compatible algorithm for valvular heart diseases screening using phonocardiogram signals with a
Shichao Ma1, Junyi Chen1, Joshua W K Ho1
1School of Biomedical Sciences, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Pokfulam, Hong Kong SAR, China; Laboratory of Data Discovery for Health Limited (D24H), Hong Kong Science Park, Hong Kong SAR, China.
Computer Methods and Programs in Biomedicine
|November 11, 2023
Summary
A new lightweight AI model accurately detects heart murmurs using mobile phones. This artificial intelligence (AI) approach enhances early screening for valvular heart disease with improved accuracy and speed.
Area of Science:
- Cardiology
- Artificial Intelligence
- Mobile Health
Background:
- Population-level screening for valvular heart disease is crucial.
- Mobile-phone-based auscultation offers scalable screening.
- Accurate, lightweight AI models are needed for mobile deployment.
Purpose of the Study:
- To develop a lightweight deep learning model for heart murmur classification.
- To improve AI model accuracy using self-supervised learning (SSL) on unlabeled data.
- To enable efficient on-device inference for mobile health applications.
Main Methods:
- A lightweight convolutional neural network (CNN) with significantly fewer parameters was designed.
- Self-supervised learning (SSL) was employed for pre-training on unlabeled phonocardiogram (PCG) data.
- A mobile application prototype was developed for in-device inference and fine-tuning.
Main Results:
- The lightweight model achieved 98.65% accuracy in 10-fold cross-validation.
- SSL pre-training boosted accuracy to over 99.4% and improved robustness to noisy data.
- On smartphones, the model performed inference in 0.03-0.37s, consuming less power than standard models.
Conclusions:
- A lightweight, accurate phonocardiogram classifier was developed.
- The model demonstrates near real-time performance on standard mobile devices.
- This technology facilitates efficient, large-scale screening for heart conditions.
