Effect of computer-aided detection for CT colonography in a multireader, multicase trial

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Gastrointestinal Diagnostics

Background:

• Computed tomographic (CT) colonography is a crucial tool for colorectal cancer screening.
• Interpreting CT colonography images can be challenging, requiring high reader accuracy.
• Computer-aided detection (CAD) systems aim to assist radiologists in image interpretation.

Purpose of the Study:

• To evaluate the impact of computer-aided detection (CAD) in second-read mode on radiologist accuracy.
• To assess CAD's effectiveness in interpreting computed tomographic (CT) colonographic images.
• To determine if CAD improves polyp detection rates and overall diagnostic performance.

Main Methods:

• A HIPAA-compliant study involving 100 colonoscopy-proved cases (52 positive for polyps, 48 negative).
• Nineteen blinded readers interpreted cases twice: once without CAD and once with CAD assistance.
• Segment-level and patient-level receiver operating characteristic (ROC) curve analyses were used to assess CAD's effect.

Main Results:

• Reader accuracy, measured by segment-level area under the ROC curve (AUC), significantly improved with CAD (0.758 with CAD vs. 0.737 without, P = .015).
• Sensitivity for detecting polyps 6 mm or larger increased across per-segment, per-patient, and per-polyp analyses with CAD.
• Sensitivity for large polyps (≥10 mm) and small adenomas also improved with CAD, though specificity slightly decreased (by 0.025, P = .05).

Conclusions:

• Computer-aided detection (CAD) significantly enhances radiologist performance in CT colonography interpretation.
• CAD improves sensitivity for detecting polyps of various sizes, including small adenomas.
• While CAD offers substantial benefits in detection, a minor reduction in specificity was observed.