No improvement in detection of advanced colorectal neoplasias in AI-assisted colonoscopy
Click Here to Manage Email Alerts
Key takeaways:
- There was no difference in detection of advanced colorectal neoplasias in AI-assisted (34.8%) vs. standard (34.6%) colonoscopy.
- Computer-aided detection increased detection of nonpolypoid and small lesions.
Colonoscopies performed with artificial intelligence did not improve detection of advanced colorectal neoplasias, although there was enhanced detection of nonpolypoid and smaller lesions, according to data.
“Detection rates for adenoma and serrated polyps have been associated with post-colonoscopy [colorectal cancer] incidence and improvement in these quality indicators is expected to enhance the preventative effectiveness of CRC screening,” Carolina Mangas-Sanjuan, MD, PhD, of the department of gastroenterology at General University Hospital of Alicante in Spain, and colleagues wrote in Annals of Internal Medicine. “Systems relying on artificial intelligence using deep-learning technology have been linked to improved adenoma detection rates in different clinical settings and also helped to reduce adenoma miss rates.”
They continued: “A limitation, however, is that ADRs may increase due to enhanced detection of small polyps and nonadvanced adenomas, whereas improved detection of advanced and more clinically significant lesions by the artificial intelligence systems has not been established.”
In a multicenter, parallel, randomized controlled trial, Mangas-Sanjuan and colleagues analyzed data from 3,213 people (mean age, 61 years; 53.4% men) with positive fecal immunochemical tests who underwent AI-assisted colonoscopy (n = 1,610) or standard colonoscopy (n = 1,603) from April 2021 through March 2022.
The primary outcome was advanced colorectal neoplasia detection rate, while secondary outcomes included mean number of advanced neoplasias, adenomas, serrated lesions, polyps and advanced serrated lesions per colonoscopy.
Researchers reported no difference in the detection of advanced colorectal neoplasias (34.8% with AI vs. 34.6% standard; adjusted risk ratio [aRR] = 1.01; 95% CI, 0.92-1.1) or mean number of advanced colorectal neoplasias detected per colonoscopy (0.54 vs. 0.52; aRR = 1.04; 99.9% CI, 0.88-1.22). There also was no difference in ADR between groups (64.2% with AI vs. 62% standard; aRR = 1.06; 99.9% CI, 0.91-1.22).
However, computer-aided detection did increase the mean number of nonpolypoid lesions (0.56 vs. 0.47) and proximal adenomas (0.94 vs. 0.81) detected per colonoscopy, as well as lesions 5 mm and smaller, including polyps (1.68 vs. 1.4), adenomas (1.12 vs. 0.97) or serrated lesions (0.25 vs. 0.19).
“Computer-aided detection was not associated with improved detection of advanced colorectal neoplasias,” Mangas-Sanjuan and colleagues concluded. “Artificial intelligence applications are in a dynamic phase. Our results show the need for improvement in this technology, using larger and more variable data sets to train deep-learning systems and for further evaluations of these new systems in large, adequately powered randomized controlled trials.”
Collapse
Read more about
Click Here to Manage Email Alerts