Detection of Hepatocellular Carcinoma in CT Images Using Deep Learning

Abstract

Abstract – The purpose of this paper is to develop a method to detect hepatocellular carcinoma, namely liver cancer in CT (Computerized Tomography) images using the deep learning that is a kind of AI (Artificial Intelligence). Firstly, the learning and recognition programs were developed using Python as a programming language and TensorFlow provided by Google that is a machine learning library. The CT images of 30 clinical subjects were selected from the DICOM format data provided by Graduate School of Medicine of Ehime University. Then 150 sets of CT images were selected where one set consists of two CT images for early and late phases in the cases with hepatocellular carcinoma. In addition, 150 sets of CT images were also selected in the cases without hepatocellular carcinoma. The 450 sets of CT images to each the 150 sets, namely 900 sets in total were created by rotating each original CT image. Consequently, 1,200 sets of CT images (2,400 CT images) in total were used for the learning. Then validity and usefulness of the learning and recognition programs were proved by examining the calculated results. This time, the hepatocellular carcinoma could be detected with relatively high sensitivity of 92.2% even with a relatively small number of learning data, namely 1,200 sets of CT images