Automatic Hepatocellular Carcinoma Diagnosis using Graph Convolutional Network

Abstract

Blood tests are used to screen a risk group for hepatocellular carcinoma. Various studies have utilized artificial intelligence to diagnose hepatocellular carcinoma using blood test records. However, most studies suffer from performance degradation due to insufficient data. In this paper, we propose a novel graph convolutional network-based computer-aided diagnosis model to address the data insufficiency problem. The proposed method assists training by converting data into graphs representing the relationships among the features. As a result, our diagnosis model has improved 4% accuracy compared to existing approaches with 89.3% accuracy.