Zebrafish are a significant model species for cardiovascular research, and accurate segmentation of the zebrafish ventricle is crucial for studying the cardiovascular system. However, the segmentation of the zebrafish ventricle currently relies on manual annotations and traditional machine learning methods, which suffers from heavy workloads and unsatisfied accuracies. In light of this, this study uses zebrafish heart beating videos as experimental data, testing and comparing seven cutting-edge deep learning models for automatic segmentation of the zebrafish ventricle. These models include UNet, PSPNet, UperNet, DeepLabV3plus, GCNet, SegFormer and KNet. We collected 178 zebrafish heart beating videos from normally cultured and exposed to different concentrations of Cresyl Diphenyl Phosphate, with exposure concentrations of 2.5, 10, 40, 160, and 640 µg/L, respectively. The training, validation, and test sets included 124, 18, and 36 videos, respectively. Our results demonstrated that KNet provided over 95.5% Dice coefficient on the test set, enabling rapid and precise segmentation of the zebrafish ventricle, which is a promising tool in zebrafish ventricle segmentation.