A task-based quality control metric for digital mammography.

Area of Science:

• Medical Imaging Physics
• Radiological Sciences
• Observer Performance Modeling

Background:

• Digital mammography requires robust quality control (QC) metrics.
• Current QC methods may not adequately capture system performance across diverse platforms.
• Task-based observer models offer a promising approach for evaluating image quality and detectability.

Purpose of the Study:

• To tune and validate an observer model for predicting the detectability index (dʹ) in digital mammography.
• To assess the potential of the observer model's dʹ as a task-based QC metric.
• To evaluate the model's performance across different digital mammography systems.

Main Methods:

• A reader study was performed to tune observer model parameters (noise power spectrum, modulation transfer function, contrast).
• A non-prewhitening observer model, including an eye-filter and internal noise, was used to predict dʹ.
• The model was validated using a test phantom with varying disc sizes imaged on six different digital mammography systems.

Main Results:

• A strong correlation (Pearson r=0.96) was observed between measured and modeled dʹ values.
• The model demonstrated robustness with low coefficients of variation in dʹ (0.07–0.10) across systems.
• Threshold thickness measurements showed minimal variation, with standard deviations ranging from 0.001 to 0.017 mm.

Conclusions:

• The observer model accurately predicts the detectability index (dʹ) for digital mammography.
• The model's dʹ shows potential as a system-independent, cross-platform QC metric.
• This approach offers a reliable method for task-based quality assessment in digital mammography.