Exploring universal segmentation models for automatic quantification of cardiac functional parameters from zebrafish heartbeat videos.

Quantifying cardiac functional parameters is crucial for assessing the toxicity of environmental chemicals on the cardiovascular system. Current methodologies for evaluating zebrafish cardiac function largely rely on tedious manual annotations and inaccurate semi-automatic or automatic measurements, hindering accurate and comprehensive functional evaluation. In this paper, we propose a framework for automatically quantifying cardiac functional parameters from zebrafish heartbeat videos by exploring universal segmentation models. We benchmarked 20 state-of-the-art deep segmentation models for automated segmentation of zebrafish ventricles and pericardia. The best-performing model, Mask2Former, was selected to segment ventricles and pericardia from the heartbeat videos. Seven cardiac functional parameters for zebrafish embryos, including heart rate, stroke volume, cardiac output, maximum ventricular area, ejection fraction, diastole to systole ratio, and pericardial arc length, were then computed based on the quantification of ventricular changes and pericardial morphologies. Experiments on 178 zebrafish heartbeat videos reveal that the trained Mask2Former exhibited remarkably superior performance, attaining an IoU of 93.46