Computer-aided detection for CT colonography: incremental benefit of observer training

Area of Science:

• Radiology
• Medical Imaging
• Gastroenterology

Background:

• Computer-assisted detection (CAD) software aids in interpreting medical images.
• CT colonography (CTC) is a key imaging modality for colorectal cancer screening.
• Observer performance in CTC interpretation can be influenced by training and software assistance.

Purpose of the Study:

• To evaluate the impact of focused training on radiologist performance using CAD for CTC interpretation.
• To quantify changes in sensitivity, specificity, and interpretation time after training.

Main Methods:

• Six radiologists with limited CTC experience received one day of focused training.
• Radiologists interpreted 20 patient datasets with CAD assistance before and after training.
• Statistical analysis included the binomial exact test and Wilcoxon's signed rank test.

Main Results:

• Per-polyp sensitivity increased by 18% overall after training (p<0.001).
• Sensitivity gains were most pronounced for 6-9 mm polyps (26% increase).
• Specificity decreased significantly (p=0.03), with a median increase of 3 false positives per 20 datasets.
• Interpretation time significantly increased from a median of 9.3 to 17.1 minutes (p=0.03).

Conclusions:

• One day of focused training enhances observer sensitivity in CAD-assisted CTC interpretation.
• The benefits of improved sensitivity are offset by decreased specificity and increased interpretation time.
• Training protocols should consider these trade-offs for optimal CAD implementation in CTC.