Introduction
Artificial intelligence has recently been developed to improve polyp detection. The aim of this study was to evaluate the impact of real-time computer-aided detection (CADe) systems on the adenoma detection rate (ADR) in routine colonoscopy.
Aims & Methods
This monocentric randomized trial compared standard colonoscopy to CADe-assisted colonoscopy (GI GeniusTM, Medtronic). According to statistical calculations, 20 endoscopists from the Clinique Paris-Bercy (Charenton-le-Pont, France) had to include approximately 2,100 patients with a 1:1 randomization. The operator was not blinded to the study arm assigned to the patient. All consecutive patients aged ≥ 18 years who were scheduled for colonoscopy in our unit were screened during colonoscopy planning. The main criterion was ADR. The COLO-GENIUS trial is complete and registered with ClinicalTrials.gov, NCT04440865.
Results
A total of 2,104 patients were included from May 1, 2021 to May 1, 2022. 2,015 patients (979 men/1,036 women) were randomized (1,012 for standard colonoscopy vs. 1,003 for CADe-assisted colonoscopy). Both groups were comparable.
ADR increased from 33.7% in the standard group to 37.5% in the CADe group (estimated mean absolute difference: 4.1% [95% CI: 0.0%;8.1%], P=0.051), while the polyp detection rate (PDR) increased from 41.0% to 45.0% (estimated mean absolute difference; 4.3 % [95% CI: 0.0%;8.5%], P=0.048) and the mean number of adenomas per colonoscopy (APC) increased from 0.71±1.30 to 0.89±1.49 (0.17 [95% CI: 0.10;0.25], P<0.001). Difference was not significant for Advanced ADR which increased from 7.6% to 9.3% (estimated mean absolute difference:1.7% [95% CI: -0.8%;4.1%], P=0.18).
One bleeding event without deglobulization occurred (in the CADe group) after a large (> 2cm) polyp resection, resolved after a hemostasis clip was placed during a second colonoscopy.
Percentages and estimated absolute differences, for the primary outcome (ADR) and the secondary outcomes (PDR, ANDR, PSPDR and APC)
| Standard
N=1,012 | CADe
N=1,003 | Estimated absolute mean difference [95% CI] | P |
| ADR, n (%) | 341 (33.7%) | 376 (37.5%) | 4.1% [-0.0%;8.1%] a | 0.051 |
| Size of the largest adenoma |
|
|
|
|
| > 1 cm, n (%) | 60 (17.8%) | 71(19.0%) | 1.1% [-3.2%;5.3%] b | 0.63 |
| 5-9 mm, n (%) | 84 (24.5%) | 96 (25.7%) | 0.7% [-2.0%;-3.3%] b | 0.63 |
| < 5 mm, n (%) | 194 (57.4%) | 207(55.4%) | -1.7% [-8.7%;5.2%] b | 0.63 |
| PDR, n (%) | 414 (41.0%) | 451 (45.0%) | 4.3% [0.0%;8.5%] a | 0.048 |
| AdvADR, n (%) | 77 (7.6%) | 93 (9.3%) | 1.7% [-0.8%;4.1%] a | 0.18 |
| PSPDR, n (%) | 106 (10.5%) | 126 (12.6%) | 2.1% [-0.7%;4.8%] a | 0.14 |
| APC, mean (SD) | 0.71 (1.30) | 0.89 (1.49) | 0.17 [0.10;0.25] c | < 0.001 |
a Computed from a mixed effects logistic regression including a random intercept at the endoscopist level
b Computed from a mixed effects ordinal logistic regression including a random intercept at the endoscopist level
c Computed from a mixed effects Poisson regression including a random intercept at the endoscopist level
CADe: computer-aided detection; PDR: polyp detection rate; ADR: adenoma detection rate; AdvADR: advanced adenoma detection rate; PSPDR: proximal serrated polyp detection rate; APC: mean number of adenomas per colonoscopy Conclusion
This large randomized study performed in a non-academic center supports the benefits of CADe (GI GeniusTM, Medtronic) regarding ADR and APC, thus suggesting the systematic use of CADe in routine colonoscopy. However, the impact of CADe on ADR is low, especially among high detectors.
Disclosure
Competing interests in relation to the study: None declared.
Other competing interest:
David Karsenti, MD: Consultant for OLYMPUS; Support for attending meetings from ALFASIGMA and FUJIFILM
Franck Venezia, MD: Honoraria for lectures, presentations, speakers for ABBVIE, TAKEDA, AMGEN and JANSSEN-CILAG; Support for attending meetings from ABBVIE; Board for JANSSEN-CILAG and BIOGEN
Maryan Cavicchi, MD-PhD: Consultant for JANSSEN-CILAG, ABBVIE and MSD; Honoraria for lectures, presentations, speakers for MSD France, AMGEN, TAKEDA, PFIZER, MYLAN and TILLOTS; Support for attending meetings from FERRING, TAKEDA and MSD; Board for JANSSEN-CILAG, ABBVIE; Receipt of equipment, materials from CELTRION and BIOSYNEX
The other authors have declared no conflict of interest
