Clinical summary

Abstract

Objective

Evaluate the safety and efficacy of computer-assisted detection (CADe) in the real-time identification of colorectal neoplasia during colonoscopy. (Artificial Intelligence for Colorectal Adenoma Detection: AID Study).

Study design

A multicenter, randomized study where patients underwent colonoscopy for colorectal cancer screening, post-polypectomy surveillance, workup following positive fecal immunohistochemical test, or GI symptoms. 700 subjects were randomized in a 1:1 fashion to colonoscopy with and without CADe using GI-Genius™ intelligent endoscopy module (15 patients were excluded from final analysis, N= 685).

Key findings

341 patients were randomized to CADe, 344 to standard colonoscopy. No statistical differences were detected between the two groups for reason for colonoscopy, adequate bowel preparation, or cecal intubation rate. 49.2% of patients were men; mean age 61.32 ± 10.2 years.

CADe was 30% more likely to detect adenomas/colorectal cancer than colonoscopy without (RR=1.3, 95% CI 1.14-1.45):

• 54.8% of colonoscopy patients with CADe were diagnosed (n=187)
• 40.4% of control patients were diagnosed (n=139)

CADe was 28% more likely to detect proximal (proximal to the splenic flexure) adenomas/colorectal cancer than colonoscopy without:

• 36.1% CADe compared to 28.2% control (RR=1.28, 95% CI=1.03-1.59)

CADe was more likely to detect both non-polypoid and polypoid lesions than control:

• 26.6% CADe versus 18.4% control in non-polypoid lesions (RR=1.42, 95% CI=1.09-1.79)
• 37.3% CADe versus 26.5% control in polypoid lesions (RR=1.36, 95% CI=1.12-1.62)

Sessile serrated lesion detection was not statistically significant between the two groups. Adenomas per colonoscopy (APC) were higher for the CADe group compared to control:

• 1.07 ± 1.54 average APCs for CADe arm versus 0.71 ± 1.20 average APCs for control arm (Incidence Rate Ratio=1.46, 95% CI=1.15-1.86)

Withdrawal times were the same between the two groups (417± 101 seconds for CADe vs. 435 ± 149 seconds for controls, p=0.1). 27.1% of overall polyp resections were non-neoplastic; the CADe and control arms were not statistically different in identification of non-neoplastic lesions (RR=1.00, 95% CI = 0.90-1.12, p=0.940).

Conclusion

The study revealed that computer-assisted detection (CADe) in the real-time identification of colorectal neoplasia during colonoscopy had an increased adenoma detection rate than the control group. The use of CADe did not impact the withdrawal time, nor was there a change in the detection of non-neoplastic lesions.