Benign and Malignant Diagnosis of Hepatocellular Carcinoma on CT Images Based on Fusion Network

Deep convolution neural network (CNN) is an important technology tool for classification of 3D-CT images. Because the training of 3D-convolution network requires a large number of labeled samples, this study proposes a method of multi-2D-convolution neural network fusion to classify benign and malignant hepatocellular carcinoma 3D-CT images. In view of the characteristics of liver CT images, this study proposes an algorithm of representing 3D-CT images from 2D-images, and then multi-network fusion method is used to distinguish two-dimensional images. In addition, this study, in conjunction with a 3A hospital, contributed a practical diagnostic data set containing more than 7000 samples of CT images of hepatocellular carcinoma. Experimental results on our datasets demonstrate validity of the proposed algorithm.