Application of Convolutional Neural Networks in Gastrointestinal and Liver Cancer Images: A Systematic Review

Abstract

Background: The use of artificial intelligence in the interpretation of images and the diagnosis of gastrointestinal and liver cancers has been evaluated. A convolutional neural network (CNN), a machine-learning algorithm similar to deep learning, has shown its capability to recognize specific features that can detect cancer. Aims: This review aimed at assessing the application of CNN in examining gastrointestinal and liver images in the diagnosis of cancer and explore the accuracy level of CNNs used. Methods: PubMed, EMBASE, and the Web of Science were systematically searched. Studies using CNNs to analyze endoscopic, pathological, or radiological images of gastroenterological or liver cancers were identified according to the international consensus standards with the aim of detecting or staging cancer. Two independent reviewers extracted the data for the study reports. The accuracy of CNNs in detecting cancer or early stages of cancer was analyzed. The primary outcomes of the review were analyzing the type of cancer, and identifying the type of images that showed optimum accuracy in cancer detection. Results: A total of 22 articles that met the selection criteria and were consistent with the aims of the study were identified. The studies covered cancers of the esophagus, stomach, pancreas, liver and biliary system and colon. It also covered risk factors and pre-cancerous conditions such as Helicobacter pylori infection, liver cirrhosis and colonic polyps. The studies were performed in Japan (n = 6), China (n = 6), the United States (n = 5), and Hong Kong, France, Switzerland, Germany, the United Kingdom, and Bangladesh (n = 1 each). The studies aimed at identifying lesions (n = 5), classification (n = 9), and segmentation (n = 8). Several methods were used to assess accuracy of the CNN and the overall level was satisfactory. Conclusions: The role of CNNs in analyzing images and as tools in early detection of gastrointestinal or liver cancers and classifying cancers has been demonstrated in these studies. Although a few limitations have been identified in these studies, overall there was an optimal level of accuracy of the CNNs used in segmentation and classification of images of gastrointestinal and liver cancers.