CT Image Segmentation of Liver Tumor Based on Improved Convolution Neural Network

Abstract

In this paper, an improved cross entropy loss function CNN structure is proposed to segment images, and a group of original three-phase images of liver tumors are input into the network for training. Finally, the dice coefficient (DSC) of the segmentation results of this method is 96%, the accuracy is 86% and the recall rate is 89%.The results show that the improved cross entropy loss function CNN structure is more conducive to the segmentation of liver tumors, with a higher accuracy. Meanwhile, it is proved that the algorithm converges when the correlation conditions are satisfied.