Combining UNet 3+ and Transformer for Left Ventricle Segmentation via Signed Distance and Focal Loss

Left ventricle (LV) segmentation of cardiac magnetic resonance (MR) images is essential for evaluating function parameters and diagnosing cardiovascular diseases (CVDs). Accurate LV remains a challenge because the large differences in structures different research subjects. In this work, network based on an encoder–decoder architecture automatic short-axis MR proposed. It combines UNet 3+ Transformer to jointly predict masks signed distance maps (SDM). can extract coarse-grained semantics fine-grained details from full scales, while used global features images. solves problem low accuracy caused by blurred edge information. Meanwhile, SDM provides shape-aware representation segmentation. The performance proposed validated 2018 MICCAI Ventricle Segmentation Challenge dataset. five-fold cross-validation evaluation was performed 145 clinical subjects, average dice metric, Jaccard coefficient, accuracy, positive predictive value reached 0.908, 0.834, 0.979, 0.903, respectively, showing better than that other mainstream ones.