A Discriminative Level Set Method with Deep Supervision for Breast Tumor Segmentation

•A novel, deeply supervised network is devised for breast tumor segmentation.•A discriminative level-set method is integrated at different stages of the network.•A new feature-discriminator is added to the level set method’s energy function.•Feature-discriminator improves model ability to identify hazy regions preciously.•The ensembling in the level set group suppresses the false-positive.•The proposed model can locate the more reliable malignant regions.•Experiments on multiple datasets affirm the success and effectiveness of the model.