Computer-aided detection-assisted colonoscopy: classification and relevance of false positives.

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Computer-aided detection (CADe) systems aim to improve colonoscopy accuracy.
• False positives (FPs) generated by CADe can reduce colonoscopy efficiency by increasing examination time.
• Understanding the causes and clinical relevance of CADe FPs is crucial for optimizing its use.

Purpose of the Study:

• To classify the causes and clinical relevance of CADe FPs during colonoscopy.
• To determine the distribution of these FPs in a real-world setting.
• To assess the impact of CADe FPs on colonoscopy withdrawal time.

Main Methods:

• Post-hoc analysis of 40 colonoscopy videos from a randomized trial comparing CADe use.
• A modified Delphi process involving 4 expert endoscopists to define FP classification domains.
• Manual examination and classification of FPs by 2 expert endoscopists during the withdrawal phase.

Main Results:

• The primary causes of CADe FPs were artifacts from the bowel wall (88%) and bowel content (12%).
• The mean number of FPs per colonoscopy was 27.3, with 5.7% requiring additional exploration time (0.7% of withdrawal time).
• FPs due to bowel wall artifacts were significantly more frequent than those from bowel content (P < .001).

Conclusions:

• CADe FPs in colonoscopy are predominantly caused by bowel wall artifacts.
• Despite a high occurrence, CADe FPs contribute minimally to overall examination time.
• Most FPs are rapidly identified and discarded by endoscopists, showing negligible clinical impact on withdrawal time.