BRAIN TUMOR SEGMENTATION BASED ON U-NET WITH IMAGE DRIVEN LEVEL SET LOSS

This paper presents an approach for brain tumor segmentation based on deep neural networks. The proposes to utilize U-Net as architecture of the capture fine and soars information from input images. Especially, train network, instead using commonly used cross-entropy loss, dice loss or both, in this study, we propose employ a new function including Level set Dice function. level is inspired Mumford-Shah functional unsupervised task. Meanwhile, measures similarity between predicted mask desired mask. proposed then applied segment MRI images well evaluated compared with other approaches dataset nearly 4000 scans. Experiment results show that achieves high performance terms coefficient Intersection over Union (IoU) scores.