Artificial Intelligence Detection Program Assessment by a High Detector: Improved Adenoma Detection Rate, but Some Misses of Large, Flat Lesions
Douglas K. Rex, MD, MASGE, reviewing Rex DK, et al. Gastroenterology 2022 Apr 11.
Artificial intelligence (AI) programs for computer-aided detection (CADe) of colorectal polyps have increased adenoma detection rates (ADR) by about 10% in parallel randomized trials and reduced adenoma miss rates by half in tandem studies.
In a CADe program assessment by a single high-detecting endoscopist, consecutive lesions were characterized as detected first by CADe, detected simultaneously by CADe and the endoscopist, detected first by the endoscopist, or not detected by CADe but detected by the endoscopist. Based on the endoscopist’s assessment, ADR increased from 52.9% to 60.6% when lesions detected first by AI were included. Similarly, adenomas per colonoscopy increased from 1.7 to 2.1 by including lesions detected first by AI.
Only 2% of lesions were detected by the endoscopist and not by CADe. The mean size of lesions detected by the endoscopist and not by AI was larger than those detected clearly by AI first (9.9 vs 3.6 mm). The lesions were also more likely to be flat (95% of lesions detected by the endoscopist and not AI vs 54% of lesions detected first by AI) than sessile or pedunculated.
![d-rex](/wp-content/uploads/d-rex.jpg)
COMMENTThese data indicate that even for a high-level detector, the use of a CADe program increases ADR. Also, this report demonstrates the importance of continuing to retrain and retest CADe algorithms to ensure they detect the full spectrum of precancerous colorectal lesions.
Note to readers: At the time we reviewed this paper, its publisher noted that it was not in final form and that subsequent changes might be made.
CITATION(S)
Rex DK, Mori Y, Sharma P, Lahr RE, Vemulapalli KC, Hassan C. Strengths and weaknesses of an artificial intelligence polyp detection program as assessed by a high detecting endoscopist. Gastroenterology 2022 Apr 11. (Epub ahead of print) (https://doi.org/10.1053/j.gastro.2022.03.055)