Computer-assisted detection for CT colonography: external validation

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Gastroenterology

Background:

• Computed tomography (CT) colonography is a key tool for colorectal cancer screening.
• Computer-assisted detection (CAD) systems aim to improve the accuracy of CT colonography interpretation.
• External validation is crucial to assess the generalizability of CAD systems.

Purpose of the Study:

• To externally validate a CAD system for CT colonography using an independent dataset.
• To evaluate the CAD system's performance across different sphericity filter settings.
• To determine the system's sensitivity and specificity for polyp detection.

Main Methods:

• Two readers analyzed 25 CT colonography datasets with known polyps.
• A CAD system, developed independently, was applied to the datasets.
• Performance was assessed by comparing CAD findings to known polyp locations, calculating true-positive, false-negative, and false-positive rates.

Main Results:

• The CAD system achieved 96% per-patient sensitivity in detecting polyps (median size 6mm).
• Sensitivity increased from 77% to 86% for polyps when lowering the sphericity setting from 0.75 to 0.50, particularly for smaller polyps (≤5mm).
• Lowering the sphericity setting significantly increased false-positive prompts, requiring more detailed review.

Conclusions:

• External validation confirmed the CAD system's high per-patient sensitivity (96%) for polyp detection in CT colonography.
• Adjusting the sphericity filter impacts the trade-off between sensitivity for small polyps and specificity.
• The findings support the potential utility of CAD systems in colorectal cancer screening via CT colonography.