An 8-layer residual U-Net with deep supervision for segmentation of the left ventricle in cardiac CT angiography

•We proposed an 8-layer residual U-Net with deep supervision for segmentation of the left ventricle in cardiac CT angiography. Experimental results exhibited that our method had high segmentation accuracy and robustness for different left ventricle shape, size, and image contrast.•We used online data augmentation composed of sequential random rotation, scaling, and shear transformation in the training process, which contributes to improve the generalization and robustness of our method.•We annotated the data by an interactive semi-supervised algorithm of graph cut, confirmed by cardiologists, which is more convenient and objective than the manual annotation.