| International Session（Panel Discussion）1(JSGE・JGES・JSGS) |
| Fri. November 1st 10:00 - 12:00 Room 7: Portopia Hotel South Wing Ohwada C |
| Artificial intelligence-assisted video colonoscopy for disease monitoring of ulcerative colitis | |||
| Yasuharu Maeda1, Shin-ei Kudo1, Noriyuki Ogata1 | |||
| 1Digestive Disease Center, Showa University Northern Yokohama Hospital | |||
|
|||
|
|||
| Introduction: The advent of artificial intelligence (AI)-augmented colonoscopy promises to mitigate the variability in diagnoses among endoscopists. Despite this potential, research to confirm if AI-assisted Mayo endoscopic subscore (MES) assessments can forecast clinical relapse or improves the diagnostic accuracy for non-experts remains absent. Methods: The AI model was trained on 74,713 images from 898 individuals who underwent colonoscopies across three institutions. In this prospective cohort analysis, we recruited 110 UC patients in clinical remission. Over a 12-month period post-colonoscopy, we observed patients for clinical relapse. A sub-analysis involving 124 videos was undertaken to ascertain if the AI could standardize diagnostic outcomes among six non-specialists. Results: The relapse rate among patients assigned an AI-determined MES of 1 was significantly elevated (24.5% [12 out of 49]), compared to those with an MES of 0 (3.2% [1 out of 31]), as indicated by the log-rank test (P = 0.01). During the follow-up, 16.2% (13 out of 80) of patients with an AI-determined MES of 0 or 1 experienced relapse, in contrast to 50% (10 out of 20) of patients with an MES of 2 or 3 (log-rank test, P = 0.03). The use of AI demonstrated superior consistency across both inter- and intra-observer evaluations compared to manual methods. Conclusion: The use of AI-based MES system can effectively categorize the relapse risk and refine the diagnostic capabilities of non-specialist practitioners. |
|||
|
Index Term 1: Airtificail intelligence Index Term 2: endoscopic remmision |
|||
| Page Top | |||