Introduction: Artificial intelligence (AI) with deep learning is revolutionizing patient care across medicine. In Gastroenterology, AI systems are helping endoscopists identify polyps in real-time. Several randomized control trials have tested the efficacy of Computer-Aided Detection (CADe) system in adenoma and polyp detection. We aimed to assess the impact of CADe on adenoma detection rates (ADR) at our institution.
Methods: This is a cross-sectional study that took place at a University Hospital between November 2021 and March 2022. We constructed a de-identified database with patients over the age of 45 that underwent screening and surveillance colonoscopies. Incomplete studies secondary to poor bowel preparation were excluded. We compared ADR, Polyp Detection Rate (PDR), total procedure time, withdrawal time, adenoma detected per colonoscopy (APC), and polyp detected per colonoscopy (PPC) between colonoscopies performed with and without CADe.
Results: A total of 64 colonoscopies were evaluated, 32 of them were done with CADe, and 32 without it. ADR was 53% with CADe and 43% without (odds ratio 1.45, 95% CI 0.5442–3.9013; p=0.4537). Polyp detection rate was 78% with CADe, 62% without CADe (odds ratio 2.1429, 95% CI 0.7118–6.4512; p=0.1753). Average total procedure time was 25 minutes 24 seconds (SD ± 7 minutes) with CADe, and 23 minutes 41 seconds without (SD ± 9 minutes) (p=0.42), average withdrawal time was 16 minutes 43 seconds (SD ± 6 minutes) for CADe and 14 minutes 49 seconds (SD ± 8 minutes) without CADe (p=0.32). APC were 1.48 (SD ± 1.15) with CADe and 0.90 (SD ± 1.3) without CADe (p=0.48). PPC were 2 (SD ± 2.38) and 1.90 (SD ± 2.69) respectively (p=0.49).
Discussion: Several randomized control trials have proven that the use of CADe increases ADR without increasing withdrawal time. In our study, ADR with CADe was found to be higher compared to an already good ADR without CADe, and procedure, as well as withdrawal time were mildly increased with the use of CADe. However, the results were not significant, likely due to a low sample size. A larger study would be needed in order to show significant differences within the two groups.
