Evaluating false-positive detection in a computer-aided detection system for colonoscopy

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Computer-aided detection (CADe) systems enhance polyp identification during colonoscopy.
• False-positive (FP) activations remain a significant challenge, increasing endoscopist workload.

Purpose of the Study:

• To evaluate the impact of an updated CADe system on the rate and causes of FP.
• To compare FP rates and associated excessive time (ET) before and after CADe system enhancement.

Main Methods:

• Analysis of colonoscopy videos (pre- and post-update CADe system).
• 1:1 propensity score matching used for comparison.
• Quantification of FP occurrences and ET per colonoscopy.

Main Results:

• The updated CADe system showed a significant reduction in mean FPs (4.2 vs 18.1) and median ET (0s vs 16s).
• Mucosal tags, bubbles, and folds were key factors contributing to FP reduction.
• No significant decrease in true positive rate or adenoma detection rate was observed.

Conclusions:

• The updated CADe system effectively reduces FPs during colonoscopy.
• This advancement alleviates endoscopist burden without compromising polyp detection efficacy.