Artificial Intelligence-Aided Colonoscopies Have Higher Adenoma Detection Rates

Colonoscopies assisted by the artificial intelligence-based computer-aid detection (CADe) device ENDO-AID had a higher adenoma detection rate (ADR) than nonassisted colonoscopies, according to study results published in Gastrointestinal Endoscopy.

For the randomized controlled trial, patients (N=312) scheduled to undergo colonoscopy were randomly assigned 1:1 to undergo colonoscopy assisted by the CADe ENDO-AID (Olympus; n=155) or without CADe assistance (n=157). The primary outcome was the difference in ADR between aided and unaided colonoscopies.

The CADe and control cohorts included patients with a mean age [SD] of 62.99[10.26] and 64.71[11.79] years; 52.9% and 52.9% were men; Boston Bowel Preparation Scale (BBPS) scores were 7.37[1.35] and 7.23[1.24]; 31.6% and 35.7% were undergoing average-risk population screening; 32.9% and 31.8% were undergoing postpolypectomy surveillance; and 12.9% and 10.2% had a rectal bleeding indication for colonoscopy, respectively.

The intervention and control cohorts did not differ for insertion time (mean, 6.01 vs 6.43 min; P =.16) or withdrawal time (mean, 7.97 vs 7.81 min; P =.29), respectively.

Overall, 55.1% of the CADe group and 43.8% of the control group had a detected adenoma (P =.029), corresponding to an ADR of 56.8% and 44.6% (P =.031), respectively. The 2 groups did not differ for the proportion of patients with resected hyperplastic polyps (19.0% vs 17.7%; P =.71) or proportion of normal histological mucosa (4.1% vs 3.2%; P =.62) among the intervention and control cohorts, respectively.

The CADe colonoscopies detected more right colon adenomas (odds ratio [OR], 2.19; P =.002), polyps (OR, 1.96; P =.004), flat adenomas (OR, 1.96; P =.006), transverse colon adenomas (OR, 1.93; P =.03), nonneoplastic lesions (OR, 1.92; P =.02), adenomas sized less than or equal to 5 mm (OR, 1.87; P =.006), nonadvanced adenomas (OR, 1.77; P =.01), and adenomas in general (OR, 1.63; P =.031) compared with control individuals.

Overall, the mean adenomas per colonoscopy (APC) rates were 1.54 for the CADe colonoscopies and 0.92 for the non-CADe colonoscopies (OR, 1.23; P =.014). The CADe strategy had a superior APC rate for flat adenomas (OR, 1.37; P =.002), right colon adenomas (OR, 1.53; P =.002), transverse colon adenomas (OR, 1.49; P =.03), and small adenomas (£5 mm; OR, 1.36; P =.002) compared with control individuals. No significant group differences in APC detection rates for protruding adenomas, adenomas in the left colon, or larger adenomas were observed.

This study may have been limited by the fact that the ADR was high overall.

“This study shows that colonoscopy assisted by ENDO-AID CADe (OIP-1) increases the ADR and APC, suggesting that it prevents overlooking colorectal neoplastic lesions, especially because the system seems to be more sensitive for flat and diminutive adenomas (when compared to the control procedure),” the study authors noted.