Medical Image Segmentation Using Deep Learning and Blending Loss

Semantic segmentation is an important task in medical-supporting. The purpose of semantic segmentation í to identify pre-defined objects, and its pixel-by-pixel location. The most popular method in semantic segmentation was using Convolutional neural network which has considerably improved semantic image segmentation. This work investigates a Blending loss which incorporates into traditional methods. Three popular algorithms are U-Net, PSPNet and FPN are examined carefully to investigate upgrading performance after combining new objective function. Moreover, we did experiments on two medical datasets to avoid bias and verify performance of the new method.