The application of panoramic segmentation network to medical image segmentation

In recent years, the image segmentation method based on deep learning has achieved outstanding performance in the field of medical image, but there are still some issues which need to be solved. In semantic segmentation, corresponding category is classified for each pixel in the image, while the instance segmentation conducts detection and segmentation of target in area-of-interest. With the continuous development of deep learning, these two tasks are increasingly integrated to realize panoramic image segmentation. This paper proposes a panoramic image segmentation network. Firstly, the bisenet network architecture is integrated to the image segmentation branch; secondly, in the image detection branch, the mask-rcnn network architecture is employed for this network; thirdly, these two branches use the same backbone network for simultaneous training and mutual improvement. Finally, we apply the proposed method in both the street view database and LiTS medical database. The results of massive experiments show that our algorithm has great performance.