Related Experiment Video
Updated: Aug 5, 2026
10:46
Efficient Derivation of Human Cardiac Precursors and Cardiomyocytes from Pluripotent Human Embryonic Stem Cells with Small Molecule Induction
Published on: November 3, 2011
A Multi-Task Deep Learning Framework for Characterizing Beating Behavior and Synchrony in Cardiomyocyte Clusters
Tianxin Wang1, Xinjie Liu2, Fangshuo Zhang1
1School of Integrated Circuits, Shandong University, Jinan 250100, China.
Bioengineering (Basel, Switzerland)
|July 28, 2026
Summary
We developed CardioSegNet, a deep learning model for analyzing cardiomyocyte cluster beating. This tool reveals local synchronization patterns, aiding cardiac function evaluation and cardiotoxicity screening.
Area of Science:
- Cardiovascular Research
- Biomedical Engineering
- Computational Biology
Background:
- Beat-level synchrony in cardiomyocyte clusters is crucial for cardiac function.
- Existing invasive methods and computer vision techniques have limitations for analyzing dense cardiomyocyte clusters.
- Accurate characterization of cardiomyocyte cluster dynamics is needed for in vitro studies.
Purpose of the Study:
- To develop an advanced computational framework for analyzing cardiomyocyte cluster beating characteristics from microscopic imaging.
- To introduce CardioSegNet, a deep learning model for high-accuracy segmentation and dynamic parameter extraction.
- To quantify inter-cluster synchrony and investigate synchronization patterns.
Main Methods:
- Developed CardioSegNet, a multi-task deep learning model with attention mechanisms for semantic segmentation, contour detection, and distance transform.
- Utilized a watershed algorithm for precise cardiomyocyte cluster segmentation.
- Applied the Pixel-Difference method to extract time-series beating signals and dynamic parameters (amplitude, period, frequency, Beat Rate Irregularity).
- Introduced PeriodAwareNAPTDij to quantify inter-cluster synchrony.
Main Results:
- CardioSegNet achieved a Dice coefficient of 0.8868 and HD95 of 93.02 µm on an independent test set, indicating strong segmentation performance.
- Analysis revealed that cardiomyocyte populations exhibit local subgroups with high internal synchrony, not uniform global synchronization.
- The degree of synchronization between clusters was found to be positively correlated with their physical distance.
Conclusions:
- The developed label-free analytical pipeline, powered by CardioSegNet, offers an efficient tool for myocardial function evaluation.
- This method is suitable for cardiotoxicity screening in vitro.
- The findings highlight complex, localized synchronization patterns within cardiomyocyte cultures.