Relationship between Roe and Metz simulation model for multireader diagnostic data and Obuchowski-Rockette model...

Area of Science:

• Medical Imaging
• Radiology
• Statistical Analysis in Medicine

Background:

• Diagnostic radiology studies involve multiple tests and readers evaluating patient cases.
• Reader performance is often assessed using receiver operating characteristic (ROC) curves and methods like Obuchowski and Rockette.
• Existing simulation models (e.g., Roe and Metz) lack direct links to performance outcomes, hindering realism evaluation.

Purpose of the Study:

• To derive relationships between Roe and Metz (RM) simulation model parameters and Obuchowski-Rockette (OR) model performance outcomes.
• To enable empirical evaluation of the realism of the RM simulation model.
• To assess the performance of OR parameter estimates, particularly for sample size estimation.

Main Methods:

• Derivation of mathematical relationships between RM model parameters and empirical area under the ROC curve (AUC) outcomes.
• Utilizing these derived relationships to connect simulation parameters to reader-performance metrics.
• Application of the derived relationships to evaluate OR parameter estimation and confidence bounds.

Main Results:

• Established explicit links between RM simulation parameters and Obuchowski-Rockette reader-performance outcomes (e.g., empirical AUC).
• Demonstrated the capability to assess the realism of the RM simulation model's parameters.
• Provided a framework for evaluating the performance of Obuchowski-Rockette parameter estimates.

Conclusions:

• The derived relationships are crucial for validating simulation models in diagnostic radiology research.
• This work facilitates more accurate assessment of statistical methods used for reader performance analysis.
• The findings support improved sample size estimation in reader studies through better variance component assessment.