Clinical summary
New artificial intelligence system: first validation study versus experienced endoscopists for colorectal polyp detection
| Title | New artificial intelligence system: first validation study versus experienced endoscopists for colorectal polyp detection |
| First author | Cesare Hassan (Digestive Endoscopy, Nuovo Regina Margherita Hospital, Rome, Italy) |
| Journal | Gut |
| Pub date | October 2019 (online), October 2020 (print) |
| DOI | 10.1136/gutjnl-2019-319914 |
Abstract
Objective
Validate GI Genius™ Intelligent Endoscopy Module using a dataset of white light endoscopy videos.
Study design
Validate the artificial intelligence system using images from 2,684 histologically confirmed polyps from 840 patients enrolled in a randomized controlled trial. Patients were randomized to either the validation or the training group.
Validation videos included 338 polyps from 105 patients. 49.7% were adenomas or sessile serrated adenomas (N=168).
Polyp video clips were created starting 5 seconds prior to the polyp visualization:
- True positive defined as a target polyp detected by the AI system
- False negative defined as no target polyp detected
Reaction time was compared between the AI and that of a panel of five expert endoscopists.
Key findings
Overall sensitivity per lesion was 99.7% (337 true positive, 1 false negative):
- Nearly 100% sensitivity of AI-assisted colonoscopy regardless of shape, size, location, and histology of the lesion did not impact detection ability.
- False positives were ≤1%.
- AI-assisted colonoscopy had a faster reaction time than the average of the endoscopists detection in 82% of polyps (AI-assisted colonoscopy was 1.27 + 3.81 seconds faster).
Conclusion
The validation of the GI Genius™ Intelligent Endoscopy Module produced high sensitivity, minimal false positives, and was faster than expert endoscopists at detecting polyps in the majority of polyps in the dataset.
