Virtual colonoscopy: effect of computer-assisted detection (CAD) on radiographer performance.

Area of Science:

• Medical Imaging
• Gastroenterology
• Artificial Intelligence in Medicine

Background:

• Virtual colonoscopy (VC) is a crucial tool for colorectal cancer screening.
• Accurate polyp detection is essential for effective VC screening.
• Computed-assisted detection (CAD) systems offer potential to enhance diagnostic accuracy.

Purpose of the Study:

• To evaluate the impact of a CAD system on polyp detection rates by trained radiographers during VC.
• To compare the effectiveness of CAD as a "second reader" versus a "concurrent reader".

Main Methods:

• Four radiographers interpreted 62 VC examinations (150 polyps) in four sessions.
• Examinations were read without CAD, with CAD as a "second reader", and with CAD as a "concurrent reader" after a 6-week interval.
• Polyp detection, false-positive rates, and interpretation times were recorded and compared.

Main Results:

• "Second reader" CAD significantly improved polyp detection by 12% (48% to 60%, p<0.001).
• No significant improvement was observed with CAD as a "concurrent reader" (p=0.20).
• Interpretation time increased moderately with "second reader" CAD (21.7 to 29.6 min, p<0.001) but not with "concurrent reader" CAD (19.1 min, p=0.12).

Conclusions:

• CAD, utilized as a "second reader", substantially enhances radiographer performance in detecting polyps during virtual colonoscopy.
• The increase in interpretation time associated with "second reader" CAD is moderate and clinically acceptable.
• CAD as a "concurrent reader" did not yield significant improvements in polyp detection.