The Impact of Artificial Intelligence (AI) on the Quality of Upper Gastrointestinal (GI) Endoscopy

Overview

Oesophageal and gastric cancer are two of the six less survivable cancers, responsible for half of cancer deaths and a quarter of cancer cases. Six cancer charities have called for focused efforts to improve the poor outcomes for these cancers that have changed little in recent years (lesssurvivablecancers.org.uk). Improving endoscopy standards to minimise missed cancer cases will be an important contribution to improving oesophageal and gastric cancer outcomes. Endoscopy, flexible telescopic examination of the oesophagus, stomach and duodenum, is the method of choice for diagnosing upper gastrointestinal (UGI) cancer and its main purpose is usually to exclude cancer as the cause of peoples' symptoms. Over 1,000,000 endoscopies are undertaken each year in the UK but the test is not perfect and sometimes cancer or an abnormality that will turn into cancer is not found. When this happens, the cancer is known as a post-endoscopy upper gastrointestinal cancer (PEUGIC) or a 'missed' cancer. This is unfortunately a relatively common occurrence and 9% of people with UGI cancer in the UK (approximately 1400 per year) had an endoscopy that did not find their cancer in the three years before diagnosis. All people who undergo endoscopy will benefit from this research. Reducing the future number of cancers that are missed at endoscopy in England will be a direct benefit but preventing missed cancers will also help to improve the general quality of endoscopy. The rate of missing cancer at colonoscopy (post-colonoscopy colorectal cancer) has fallen from 9% in 2005 to 6.5% in 2013, unlike the PEUGIC rate that has increased between 2009 and 2018. Research has shown that endoscopists with longer procedure times and those who take more than four pictures during endoscopy have a higher abnormality detection rate for early cancer. In an attempt to help endoscopists, a novel AI called Cerebro has been developed as an endoscopy quality control tool. Cerebro gives the endoscopist real time feedback during an endoscopy, and aids them in the four following areas (Endovision AI 2022) 1. Ensures inspection completeness prompting the endoscopist on which areas have been missed. 2. Calculates the time spent at each landmark ensuring at least a 7-minute examination time. 3. Provides automatic photodocumentation which allows for better reporting 4. Prompts the endoscopist when further insufflation or washing is needed to improve views Variation in endoscopy quality in the UK will contribute to variations in missed cancer frequency and efforts to improve endoscopy quality, including using AI to standardise endoscopy quality, will hopefully reduce the frequency of PEUGIC in future and improve upper GI cancer outcomes. However, in order for AI use in endoscopy to be established its value in improving the quality of views needs studying.

Study design The research team propose to carry out a randomised crossover trial to assess the impact of AI on the quality of upper GI endoscopy Primary objective Quantifying the improvement that AI provides for a diagnostic endoscopy Primary Endpoint Number of sites successfully inspected during endoscopy (0-28) Secondary Endpoints Overall procedure time Individual site inspection time (0-28) Hypothesis The research team hypothesize that with the use of AI the number of sites successfully inspected during endoscopy will be increased and as a result there will be an increase in the overall procedure time and individual site inspection time. Plan An Artificial intelligence endoscopy quality control tool has been developed called Cerebro (Endovision, Hong Kong). Cerebro has been developed and validated in Hong Kong and Singapore but not previously used elsewhere in the world. Validation phase Although the AI has been validated in the Far East, there have been no validation studies carried out in the West. The research team will record 30 endoscopy videos for the AI to analyse to provide a score for completeness for the procedure. The same videos will be viewed by a panel of endoscopy experts to see if they are in agreement with the AIs completeness score for the procedure. All videos will be anonymised. Work phase one The research team plan to use Cerebro initially to assess the quality of endoscopy in independent and trainee endoscopists in the UK in a cross-sectional study of all endoscopists at Sandwell and West Birmingham NHS trust (SWBH) by recording participants success at examining all areas of the upper GI tract during diagnostic endoscopy as assessed by Cerebro in at least ten endoscopies, without providing feedback from Cerebro to the endoscopist during endoscopy at this stage. This analysis will provide data on the quality of endoscopy and particular areas of the upper GI tract that are not well examined in Western endoscopic practice that may contribute to missed cancer (PEUGIC). Work phase two Will involve participants undergoing diagnostic endoscopy at Sandwell and West Birmingham NHS trust (SWBH). We would then plan a randomised cross-over trial of Cerebro among all available endoscopists at SWBH. Following a period where Cerebro is used to assess endoscopists' baseline performance during at least 10 endoscopies without providing feedback to the endoscopists, the endoscopists of varying levels of experience and specialty will be randomised to receive feedback from Cerebro during all endoscopies or have Cerebro continue to record endoscopy view quality but not feedback to the endoscopist for three months. The two groups will then be crossed over. The learning curve to improve endoscopy view quality and sustainability of improvements in the absence of real-time feedback from Cerebro will then be assessed. Study population Inclusion criteria Subjects 18 years of age or above who are scheduled for outpatient diagnostic UGI endoscopy will be invited to part take in the study. Exclusion criteria The following subjects will be excluded from the study: inpatient/ emergency OGD referrals, referrals for therapeutic OGD (e.g. polypectomy, feeding tube insertion), previous gastro-duodenal surgery, any co-morbidity that may impair ability to provide information or give valid consent (e.g. dementia, cerebral vascular disease) or any major medical or neoplastic co-morbidity. Estimated duration 24 months