Artificial intelligence and machine learning in gastroenterology and hepatology

Abstract

This thesis investigated the role of artificial intelligence and machine learning in field of gastroenterology and hepatology. In particular, the first six studies were to explore the role of artificial intelligence and deep learning in gastrointestinal endoscopy followed by four studies on the use of machine learning on diagnosis and risk prediction in various gastroenterology and hepatology diseases. The first study is the first real-time use of deep learning model on colonoscopy in Hong Kong. It demonstrated that use of deep learning model can increase detection of colonic polyps in a prospective cohort as well as a retrospective cohort. The second study is the first multi-centre randomized controlled trial on the use of deep learning model on colonoscopy in Asia. It showed that deep learning model can significantly improve adenoma and polyp detection rate as well as the number of adenoma and polyp per patient. The third study explored the deep learning model in prediction of histology of colonic polyp, in particular large polyp. It showed that deep learning model was accurate in determination of the histology of large polyp, which allowed endoscopist to choose the appropriate endoscopic intervention of that particular polyp. The fourth study assessed the accuracy of deep learning model in prediction of histology of gastric lesions as well as its effects in the learning curve of junior endoscopist in assessment of gastric lesion. It showed that deep learning model was accurate in prediction of histology and it actually improved the performance of junior endoscopist regarding the gastric polyp histology assessment endoscopically. The fifth study assessed the effect of artificial intelligence on the polyp surveillance interval. It showed that a significant proportion of patients will be assigned to higher risk group due to the fact that AI actually detected more polyps. Therefore, they would require a shorter surveillance interval according to the current guideline. The next two studies were about the use of machine learning for the diagnosis of gastrointestinal diseases. They showed that advanced machine learning can actually outperform traditional logistic regression in prediction of gastric cancer after Helicobacter pylori eradication as well as prediction of post-colonoscopy colorectal cancer after colonoscopy. The next one was about the application of machine learning for the prognosis of gastrointestinal disease. It showed that advanced machine learning model can outperform the traditional mortality prediction score in determination of one year mortality for patient with advanced HCC on immunotherapy. The final study investigated the use of time series machine learning model to estimate the impact of COVID-19 on endoscopy service. It showed that advanced machine learning model can accurately predict the impact of COVID-19 on endoscopy service. This series of studies demonstrated the application of artificial intelligence and machine learning in gastrointestinal endoscopy as well as diverse gastrointestinal diseases, which can outperform traditional approach.