Screening mammography: test set data can reasonably describe actual clinical reporting

Area of Science:

• Radiology
• Medical Imaging
• Screening Mammography

Background:

• Accurate interpretation of screening mammograms is crucial for early breast cancer detection.
• Evaluating the reliability of test set readings compared to real-world clinical reporting is essential for quality assurance.

Purpose of the Study:

• To determine how well test set readings represent actual clinical reporting in screening mammography.
• To assess the impact of different test conditions on the representativeness of readings.

Main Methods:

• Compared actual clinical reports with three test set conditions (clinical with priors, lab with priors, lab without priors).
• Assessed agreement using Kendall coefficient of concordance and analyzed region-of-interest figures of merit, sensitivity, and specificity.
• Established intraobserver variability baseline with a separate group of readers.

Main Results:

• Moderate agreement (W = 0.69-0.73) was observed between actual clinical reporting and test set conditions, especially when prior images were available.
• Agreement was lowest when prior images were excluded from test set readings.
• Test set conditions showed higher median figures of merit, potentially due to increased sensitivity.

Conclusions:

• Test set readings can achieve reasonable agreement with actual clinical reporting in mammography.
• Caution is advised as test set conditions may lead to inflated sensitivity compared to clinical practice.