Segnet-based gland segmentation from colon cancer histology images

The morphology of glands is the main basis of colon cancer diagnosis and accurate segmentation of glands from histology images is a prerequisite for correct clinical diagnosis. A gland segmentation method based on Segnet is proposed in this paper. First, the Warwick-QU dataset is used and augmented for training Segnet. Second, the network parameters are optimized according to the training results. Finally, the trained Segnet is used to perform segmentation test on both Parts A and B of Warwick-QU. The results show that the presented method achieves segmentation accuracy of 0.882 on Part A and 0.8636 on Part B, and shape similarity 106.6471 on Part A and 102.5729 on Part B, respectively. Compared with the existing methods with respect to the same dataset, our method reaches the highest accuracy on whole.