Endoscopy has evolved as a diagnostic and therapeutic tool for gastrointestinal diseases. With the advanced development in imaging, there is significant improvement in detection of early gastrointestinal neoplasia. Recently, screening program has been implemented around the world on colorectal cancer and detection of early colorectal neoplasia had increased. Nowadays, detection and characterization of early gastrointestinal neoplasia is greatly improved with use of image enhanced endoscopy with magnification. However, the interpretation of the features to characterize early GI neoplasia required significant knowledge and focus of the endoscopist.
The recent development in Artificial Intelligence (AI) through deep learning allowed interpretation of image data in a short period of time, making immediate detection and characterization of early GI neoplasia possible during endoscopy. AI will enhance the workflow of diagnostic and therapeutic endoscopy in terms of quality, safety and efficiency. In collaboration with a local HK startup, we developed an AI driven system for quality control of diagnostic Esophago-gastro-duodenoscopy (EGD). In a prospective clinical trial, the completeness of diagnostic EGD was significantly higher with the use of Cerebro (AI system) among endoscopists trainee compared to those without using AI.
AI has also extended to application in improving detection of colorectal polyps and neoplasia. In a large prospective randomized trial conducted at endoscopy center of Prince of Wales Hospital, we confirmed that use of EndoAID could increase the overall adenoma detection rate among junior endoscopists in training especially for small-to-medium size and non-pedunculated adenomas, in different locations of colon and different levels of experience.
Our team is now developing an AI system in assisting the performance of gastric endoscopic submucosal dissection (ESD) for treatment of early gastric cancer. The initial experiment demonstrated a high efficiency in predicting trajectory during ESD. In future, this system will be able to provide real-time guidance towards the safest plane for submucosal dissection during ESD.
