Ding Huang1, Jingyi Shen2, Jiaze Hong2, Yi Zhang2, Senjie Dai2, Nannan Du2, Mengting Zhang2, Daxin Guo3. 1. Medical Department, Ningbo Anorectal Hospital, Ningbo, Zhejiang, China. 2. The Second Clinical Medical College, Zhejiang Chinese Medical University, Hangzhou, Zhejiang, China. 3. Department of Gastroenterology, HwaMei Hospital, University of Chinese Academy of Sciences, Haishu District, Northwest Street 41, Ningbo, 315010, Zhejiang, China. airgzn666@163.com.
Abstract
BACKGROUND: This meta-analysis aimed to determine whether artificial intelligence (AI) improves colonoscopy outcome metrics i.e. adenoma detection rate (ADR) and polyp detection rate (PDR). METHODS: Two authors independently searched Web of Science, PubMed, Science Direct, and Cochrane Library to find all published research before July 2021 that has compared AI-aided colonoscopy with routine colonoscopy (RC) for detection of adenoma and polyp. RESULTS: This meta-analysis included 10 RCTs with 6629 individuals in AI-aided (n = 3300) and routine (n = 3329) groups. The results showed that both ADR (RR, 1.43; P < 0.001) and PDR (RR, 1.44; P < 0.001) using AI-aided endoscopy were significantly greater when compared with RC. The adenomas detected per colonoscopy (APC) (WMD, 0.25; P = 0.009), polyps detected per colonoscopy (PPC) (WMD, 0.52; P < 0.001), and sessile serrated lesions detected per colonoscopy (SSLPC) (RR, 1.53; P < 0.001) were significantly higher in the AI-aided group compared with the RC group. Subgroup analysis based on size, location, and shape of adenomas and polyps demonstrated that, except for in the cecum and pedunculated adenomas or polyps, the AI-aided groups of the other subgroups are more advantageous. Withdrawal time was longer in the AI-aided group when biopsies were included, while withdrawal time excluding biopsy time showed no significant difference. CONCLUSIONS: AI-aided polyp detection system significantly increases lesion detection rate. In addition, lesion detection by AI is hardly affected by factors such as size, location, and shape.
BACKGROUND: This meta-analysis aimed to determine whether artificial intelligence (AI) improves colonoscopy outcome metrics i.e. adenoma detection rate (ADR) and polyp detection rate (PDR). METHODS: Two authors independently searched Web of Science, PubMed, Science Direct, and Cochrane Library to find all published research before July 2021 that has compared AI-aided colonoscopy with routine colonoscopy (RC) for detection of adenoma and polyp. RESULTS: This meta-analysis included 10 RCTs with 6629 individuals in AI-aided (n = 3300) and routine (n = 3329) groups. The results showed that both ADR (RR, 1.43; P < 0.001) and PDR (RR, 1.44; P < 0.001) using AI-aided endoscopy were significantly greater when compared with RC. The adenomas detected per colonoscopy (APC) (WMD, 0.25; P = 0.009), polyps detected per colonoscopy (PPC) (WMD, 0.52; P < 0.001), and sessile serrated lesions detected per colonoscopy (SSLPC) (RR, 1.53; P < 0.001) were significantly higher in the AI-aided group compared with the RC group. Subgroup analysis based on size, location, and shape of adenomas and polyps demonstrated that, except for in the cecum and pedunculated adenomas or polyps, the AI-aided groups of the other subgroups are more advantageous. Withdrawal time was longer in the AI-aided group when biopsies were included, while withdrawal time excluding biopsy time showed no significant difference. CONCLUSIONS: AI-aided polyp detection system significantly increases lesion detection rate. In addition, lesion detection by AI is hardly affected by factors such as size, location, and shape.