Interpretive Performance and Inter-Observer Agreement on Digital Mammography Test Sets.

Area of Science:

• Radiology
• Medical Imaging
• Breast Imaging

Background:

• Digital mammography interpretation is crucial for breast cancer screening.
• Assessing radiologist performance and agreement is vital for quality assurance.
• Understanding factors influencing diagnostic accuracy is an ongoing research area.

Purpose of the Study:

• To evaluate radiologist interpretive performance on digital mammograms.
• To assess inter-observer agreement among radiologists interpreting mammograms.
• To investigate if radiologist characteristics impact performance and agreement.

Main Methods:

• Twelve radiologists independently interpreted 1000 digital mammograms (12 cancer cases).
• Performance metrics included recall rate, cancer detection rate (CDR), positive predictive value (PPV), sensitivity, specificity, false positive rate (FPR), and area under the ROC curve (AUC).
• Inter-radiologist agreement was measured; radiologist characteristics (experience, fellowship, volume) were analyzed.

Main Results:

• Interpretive performance varied: recall rate 7.5%, CDR 10.6/1000, PPV 15.9%, sensitivity 88.2%, specificity 93.5%, FPR 6.5%, AUC 0.93.
• Radiologists interpreting >3000 mammograms annually showed higher CDR and sensitivity (p=0.064).
• Inter-radiologist agreement showed 77.2-88.8% agreement and kappa 0.27-0.34; characteristics did not affect agreement.

Conclusions:

• Radiologist performance met American College of Radiology screening goals, despite inter-observer variability.
• Higher annual screening mammogram interpretation volume may be associated with improved radiologist performance.
• Further research may explore optimizing training and quality control for digital mammography interpretation.