Effectiveness of a deep-learning polyp detection system in prospectively collected colonoscopy videos with variable bowel preparation quality

Abstract

Background: Colonoscopy is the gold standard for polyp detection, but polyps may be missed. Artificial intelligence (AI) technologies may assist in polyp detection. To date, most studies for polyp detection have validated algorithms in ideal endoscopic conditions.

Aim: To evaluate the performance of a deep-learning algorithm for polyp detection in a real-world setting of routine colonoscopy with variable bowel preparation quality.

Methods: We performed a prospective, single-center study of 50 consecutive patients referred for colonoscopy. Procedural videos were analyzed by a validated deep-learning AI polyp detection software that labeled suspected polyps. Videos were then re-read by 5 experienced endoscopists to categorize all possible polyps identified by the endoscopist and/or AI, and to measure Boston Bowel Preparation Scale.

Results: In total, 55 polyps were detected and removed by the endoscopist. The AI system identified 401 possible polyps. A total of 100 (24.9%) were categorized as “definite polyps;” 53/100 were identified and removed by the endoscopist. A total of 63 (15.6%) were categorized as “possible polyps” and were not removed by the endoscopist. In total, 238/401 were categorized as false positives. Two polyps identified by the endoscopist were missed by AI (false negatives). The sensitivity of AI for polyp detection was 98.8%, the positive predictive value was 40.6%. The polyp detection rate for the endoscopist was 62% versus 82% for the AI system. Mean segmental Boston Bowel Preparation Scale were similar (2.64, 2.59, P=0.47) for true and false positives, respectively.

Conclusions: A deep-learning algorithm can function effectively to detect polyps in a prospectively collected series of colonoscopies, even in the setting of variable preparation quality.

Colorectal cancer (CRC) is the third most common cancer in the United States, with an estimated annual incidence of 101,420 colon and 44,180 rectal cancers in 2019.1 Most CRC arises from precancerous adenomas, with an evolution from a polyp to invasive carcinoma over a mean period of 10 years.2 Colonoscopy is the gold standard screening method for CRC. This procedure allows detection and removal of precancerous adenomas in a timely manner, and thus decrease the incidence of interval CRC. Every 1% increase in the adenoma detection rate (ADR) is associated with a 3% decrease in the risk of interval CRC, and 5% decrease in interval CRC-related mortality.3 ADR is an established quality metric for colonoscopy.4,5 The American Society of Gastrointestinal Endoscopy (ASGE) and the American College of Gastroenterology (ACG) recommend a target ADR of 25% (30% for men and 20% for women).6 Polyps can go undetected during a colonoscopy for several reasons, including inadequate visual inspection of the mucosa secondary to poor insufflation, poor quality of preparation, location behind mucosal folds, or lack of recognition by the endoscopist. Small (<5 mm) and sessile polyps can be particularly hard to recognize.7–9

There is evidence that a wide variety of technological innovations can potentially increase ADR, including the use of high-definition endoscopes, wide-angle viewing colonoscopes, and various devices to flatten folds such as Endocuff and EndoRings.10 Specific endoscopic techniques have also been shown in some studies to increase the ADR, including extending withdrawal time and right colon retroflexion.11,12 Of particular interest to the field of artificial intelligence (AI)-assisted colonoscopy, several studies have shown that an “extra set of eyes” (ie, nursing staff, fellow trainees) observing the video screen during colonoscopy can increase the ADR.13–16

Several studies have recently been published demonstrating the potential of AI for colon polyp detection.17–19 However, many of the published studies in this area have focused on optimal conditions for AI software training and subsequent polyp detection, with AI software being trained and tested on still images, or curated videos from nonconsecutive cases. In this study, we sought to evaluate the performance of AI with a deep-learning algorithm for polyp detection in a real-world setting of a routine colonoscopy, with variable bowel preparation quality.