Related Experiment Video
Updated: Feb 25, 2026
12:10
Semi-automated Optical Heartbeat Analysis of Small Hearts
Published on: September 16, 2009
12.9K
Performance of an open-source heart sound segmentation algorithm on eight independent databases.
Chengyu Liu1, David Springer, Gari D Clifford
1Department of Biomedical Informatics, Emory University, Atlanta, United States of America.
Physiological Measurement
|August 2, 2017
Summary
The hidden semi-Markov model (HSMM) accurately segments heart sounds, achieving high scores for S1, systole, S2, and diastole intervals. This robust algorithm is effective for analyzing diverse heart sound data.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Cardiology
Background:
- Heart sound segmentation is crucial for automated analysis and pathological event detection.
- Hidden Markov models (HMMs) show promise for noisy heart sound recordings.
- A logistic regression-based hidden semi-Markov model (HSMM) for heart sound segmentation has been recently developed.
Purpose of the Study:
- To evaluate the performance of the logistic regression-based HSMM heart sound segmentation method.
- To assess the algorithm's effectiveness using a diverse range of independently acquired heart sound data with varying quality.
Main Methods:
- A systematic evaluation scheme was developed using a new collection of over 120,000 seconds of heart sounds from 1297 subjects across eight independent databases.
- The HSMM segmentation method was evaluated on this comprehensive dataset.
- Standard metrics including sensitivity, specificity, accuracy, and the F-score were employed, along with an analysis of tolerance window effects.
Main Results:
- The HSMM algorithm demonstrated high accuracy on a large test dataset (102,306 heart sounds).
- Average F-score of 98.5% was achieved for S1 and systole segmentation.
- Average F-score of 97.2% was achieved for S2 and diastole segmentation.
- The F-score increased with larger tolerance window sizes.
Conclusions:
- The HSMM algorithm exhibits high segmentation accuracy on extensive and varied heart sound data, confirming its effectiveness.
- The developed evaluation framework and large open-access heart sound dataset offer valuable resources for algorithm testing and reproducible research.
