AI-Assisted Colonoscopy Linked to Improved Adenoma Detection Rate

artificial intelligence
artificial intelligence
Researchers compared the adenoma detection rate in artificial intelligence-assisted colonoscopy vs conventional colonoscopy.

Artificial intelligence (AI)-assisted colonoscopy improved the overall adenoma detection rate (ADR) in asymptomatic patients undergoing colorectal cancer (CRC) screening, according to a study in Clinical Gastroenterology and Hepatology.

The randomized controlled trial (ClinicalTrials.gov Identifier: NCT04422548) compared AI-assisted colonoscopy with standard colonoscopy for adenoma detection at 6 university-affiliated endoscopy centers in Hong Kong, Beijing, Inner Mongolia, Jilin, and Xiamen, China.

Participants were asymptomatic adults aged 45 to 75 years undergoing CRC screening by direct screening colonoscopy or fecal immunochemical test (FIT)-based screening from November 2019 to August 2021.

The patients were randomly assigned 1:1 to AI-assisted colonoscopy or conventional colonoscopy (CC). The AI polyp detection system used was Eagle Eye, v. 5.1 (Xiamen Innovision, Xiamen, China).

The colonoscopies were conducted by 2 nonexpert attending endoscopists (colonoscopy experience <5000) and 2 expert attending endoscopists (colonoscopy experience ≥5000) at each of the 6 centers. Patients in the AI group had the same colonoscopy procedure as that of the CC group, except that the AI system was turned on for real-time polyp detection during the intubation and withdrawal phases.

The primary outcome was overall ADR, which was defined as the proportion of patients with at least 1 adenoma detected during colonoscopy. Secondary outcomes were the mean number of adenomas per colonoscopy (APC), ADR according to the experience of endoscopists, adenoma size, morphology, histology, colonoscopy intubation time, and withdrawal time.

A total of 3059 patients were enrolled, of whom 1519 participants (mean age, 57.49±7.55 years; men, 46.5%) were assigned to the AI group and 1540 patients (mean age, 57.03±7.43 years; men, 47.3%) were in the CC group and included in the intention-to-treat (ITT) analysis.

The median intubation time (minutes) (4.48; [IQR, 3.33-6.30] vs 4.17 [IQR, 3.17-5.76], P <.001) and the median withdrawal time (minutes) (8.25; [IQR, 6.67-12.00] vs 7.78 [IQR, 6.45-11.00], P =.004) were significantly longer in the AI group vs the CC group in the ITT analysis.

The overall ADR detection rate in the AI group was significantly higher compared with the CC group (39.9% vs 32.4%, P <.001) in the ITT analysis. The APC in the AI group also was significantly increased (0.59±0.97 vs 0.45±0.81, P <.001).

AI-assisted colonoscopy improved ADR detection for the expert attending endoscopists (42.3% vs 32.8%, P <.001) and nonexpert attending endoscopists (37.5% vs 32.1%, P =.023) vs CC in subgroup ITT analysis.

Detection of nonadvanced adenomas in the AI group was significantly improved compared with the CC group (32.3% vs 26.7%, P =.001). Advanced adenoma detection also was significantly higher in the AI group vs the CC group (6.6% vs 4.9%, P =.041).

AI-assisted colonoscopy significantly increased the detection of adenomas less than 5 mm (16.5% vs 11.5%, P <.001), adenomas greater than or equal to 10 mm (6.5% vs 4.7%, P =.033), and nonpedunculated adenomas (27.6% vs 21.8%, P <.001) vs CC.

The researchers noted that the AI system used did not undergo dedicated training with a large dataset of sessile serrated lesions (SSL), and therefore the sensitivity and specificity for SSL of the AI system is unclear.

“This large-scale, multicenter, randomized study showed the benefits of AI-assisted colonoscopy in asymptomatic subjects undergoing CRC screening,” the study authors wrote. “AI-assisted image analysis has already been applied in mammography for the screening of breast cancer, as well as in 3-D low-dose computed tomography for the screening of lung cancer. It is time for us to consider generalizing the use of AI-assisted endoscopy in the gastrointestinal tract.”

Reference

Xu H, Tang RSY, Lam TYT, et al. Artificial intelligence-assisted colonoscopy for colorectal cancer screening: a multicenter randomized controlled trial. Clin Gastroenterol Hepatol. Published online July 18, 2022. doi:10.1016/j.cgh.2022.07.006