Clinical consequences of computer-aided colorectal polyp detection.

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Computer-aided detection (CADe) systems aim to improve polyp detection during colonoscopies.
• Previous studies suggest CADe benefits, but potential biases and clinical outcome impacts require further investigation.

Purpose of the Study:

• To evaluate the real-world effectiveness of a deep learning-based CADe system in colon polyp detection.
• To compare the adenoma detection rate (ADR) and polyp detection performance between endoscopists with and without CADe assistance.

Main Methods:

• A multicenter trial utilizing a CADe system (convolutional and recurrent neural networks) for real-time polyp detection.
• Blinded endoscopists were monitored, with CADe detections prompting immediate reinspection.
• Adenoma detection rates and polyp detection rates were measured pre- and post-study, with histological confirmation.

Main Results:

• CADe sensitivity for polyp detection was comparable to human endoscopists (94.6% vs 96.0%), but superior for adenomas.
• CADe led to a 1.1% increase in ADR per patient, primarily from small adenomas (<5mm).
• Non-neoplastic polyp detection increased by 4.9%, procedure time rose by 6.6 minutes, and surveillance intervals changed in 2.3% of patients.

Conclusions:

• The additional adenoma detection gain from CADe was minimal, with rare changes to patient surveillance intervals.
• CADe increased the detection of non-neoplastic polyps, potentially adding to the workload for inspection and polypectomy.