Reliable Automatic Organ Segmentation from CT Images using Deep CNN
Chang Liu, Shaozhi Wu, Su Wu, Ziheng Wang, Kai Xiao · 2019
In the field of biomedical image analysis, segmentation is critical for disease diagnosis and treatment planning. While manual segmentation is tedious, time consuming and subjective due to the large shape and appearance variance among different subject, accurate and reliable segmentation is very challenging for automatic segmentation methods at the same time. In this paper, we present our recent effort on developing a reliable segmentation algorithm in the form of a convolutional neural network. Our network architecture is inspired by the popular U-Net and its variation (3D U-Net) and has been carefully modified to maximize bladder and rectum segmentation performance. We transfer our data to four channels to augment the data and prevent overfitting. The Dice similarity coefficient (DSC) is used to evaluate the network's performance. The outcome of experiments demonstrates the superiority of the proposed method.