CADe in endoscopy - Endoscopy Feb. 26

Introduction

Colonoscopy quality is often judged by the adenoma detection rate (ADR), because better detection is linked to lower risk of future colorectal cancer. AI-based computer-aided detection (CADe) systems reliably increase ADR in the short term. But many clinicians have a practical concern: if endoscopists start depending on AI, will their “native” detection skills worsen when AI is not used? In other words—does CADe cause deskilling over time?

Problem statement

Most CADe studies show an immediate ADR benefit, but they are usually short-duration trials and rarely address what happens months to years later in real-world practice. The unanswered question is:

After CADe is implemented, do endoscopists maintain their performance in standard (non-CADe) colonoscopy, or does performance drift downward because AI is doing the “thinking”?

This study specifically examined whether CADe leads to skill transfer (endoscopists learn and improve even without AI) or deskilling (performance drops without AI).

What the study did:

• Single-center, prospective, real-world study over 3 years (2021–2023).
• CADe was installed in half of the colonoscopy rooms, and patients were distributed across CADe vs non-CADe rooms as part of routine workflow.
• Endoscopists were grouped based on their baseline detection performance:
• High detectors (already meeting quality benchmark)
• Low detectors (below benchmark)
• The key focus: How detection performance changed over time, both with CADe and without CADe, to see if skills improved or deteriorated.

Key findings clinicians should know

1. CADe improved detection when it was used

Across both strong and weaker detectors, AI support increased the ability to find adenomas and polyps—especially subtle lesions.

2. No evidence of deskilling in non-CADe colonoscopy

The most clinically important finding: once CADe was introduced, performance in standard colonoscopy did not fall over time. In other words, using CADe did not make endoscopists worse when they scoped without AI.

3. High detectors showed meaningful skill transfer

High-performing endoscopists not only benefited during CADe use, but also appeared to internalize improvements—their non-CADe performance stayed strong and in some analyses improved over time.

4. Low detectors improved with AI—but showed limited learning without AI

Low detectors clearly benefited when CADe was on, but their unassisted learning curve did not show the same degree of sustained improvement. This suggests CADe helps them “in the moment,” but may not automatically translate into durable skill gains without additional training support.

5. CADe particularly helps with subtle lesions

Detection gains were strongest for lesions that are easier to miss—such as flat lesions and sessile serrated lesions—supporting CADe as a quality-enhancing tool, not just a “polyp counter.”

Practical conclusion for clinicians

CADe can be implemented without fear of deskilling. Over 3 years of real-world use, endoscopists did not lose their baseline detection ability in standard colonoscopy.

However, the study also suggests a key implementation lesson:

• High detectors may naturally translate AI support into lasting skill improvement.
• Low detectors may need structured feedback/training (beyond simply turning on AI) to convert CADe assistance into sustained independent performance.

What this means for your practice / unit

• If your unit is considering CADe, this study is reassuring: AI support does not appear to erode core skills.
• CADe can be positioned not only as a detection aid but as a quality framework tool, especially for improving detection of subtle lesions.
• For training programs: CADe should ideally be paired with targeted coaching, particularly for low detectors, to ensure long-term uplift even without AI.

One-line GastroAGI “Clinical Takeaway”

CADe boosts detection and—importantly—does not cause long-term deskilling; performance in non-CADe colonoscopy is maintained over time.