人工智能辅助结肠镜检测结直肠息肉的临床研究

Objective To explore the application value of artificial intelligence(AI)-assisted diagnosis in colonoscopy, and to provide evidence-based references for optimizing the endoscopic workflow of early colorectal cancer screening and reducing the missed diagnosis and misdiagnosis rates of polyps. Methods Endoscopic image data of 2,000 patients who underwent routine white-light colonoscopy with complete pathological biopsy results in our hospital from January 2024 to December 2024 were enrolled. The images were divided into a training set(1,600 images) and a validation set(400 images) at an 8∶2 ratio. Two modes were adopted for the validation set: manual endoscopic interpretation alone(artificial group) and AI-assisted endoscopic interpretation(AI group). Taking postoperative pathological results as the gold standard for diagnosis, the polyp detection efficiency, polyp classification accuracy, adenoma missed diagnosis rate, unnecessary biopsy rate and interpretation time consumption of the two groups were compared. Results The sensitivity, specificity, and area under the curve(AUC) of polyp detection in the AI group were 96.88%, 96.81%, and 0.975, respectively, which were higher than those in the artificial group(85.71%, 87.15%, and 0.902)(P<0.05). The accuracy rates of adenomatous polyps and hyperplastic polyps in the AI group were 93.50% and 91.75%, which were higher than 79.25% and 76.00% in the artificial group(P<0.05). The rate of missed diagnosis of adenoma and unnecessary biopsy in the AI group were 3.13% and 5.36%, which were lower than those in the artificial group(14.29% and 21.43%), and the average time of single colonoscopy image interpretation was shorter(P<0.05). Conclusion The CNN-based AI-assisted colonoscopy system can improve the detection accuracy and pathological classification accuracy of colorectal polyps, reduce missed adenoma diagnosis and excessive biopsy of non-adenomatous polyps, and shorten the time consumption of endoscopic diagnosis. It can help junior endoscopists improve diagnostic proficiency, and is worthy of promotion and application in colonoscopy screening at hospitals of all levels.