Investigating simulation-based metrics for characterizing linear iterative reconstruction in digital breast...

Area of Science:

• Medical imaging physics
• Digital breast tomosynthesis (DBT)
• Image reconstruction algorithms

Background:

• Accurate image quality assessment is crucial for optimizing digital breast tomosynthesis (DBT) reconstruction parameters.
• Traditional metrics like root-mean-square-error (RMSE) may not fully capture perceptual image quality relevant to clinical tasks.
• Developing robust simulation-based metrics is essential for characterizing parameter dependencies in DBT image reconstruction.

Purpose of the Study:

• To adapt and investigate simulation-based image quality metrics for characterizing parameter dependences in linear iterative image reconstruction for DBT.
• To evaluate the performance of standard RMSE, gradient RMSE, and a region-of-interest (ROI) Hotelling observer (HO) metric.

Main Methods:

• Three 2D DBT simulation-based metrics were investigated: image RMSE, gradient RMSE, and ROI-HO observer for signal-known-exactly/background-known-exactly (SKE/BKE) and signal-known-exactly/background-known-statistically (SKE/BKS) tasks.
• Simulation studies varied voxel aspect ratio and regularization strength for Tikhonov-regularized least-squares optimization.
• Metrics were applied to bar pattern phantoms for visual assessment comparison and to ACR phantom data for clinical task relevance (lesion and microcalcification detection).

Main Results:

• Image RMSE sensitivity to mean pixel value limited its applicability for DBT reconstruction assessment.
• Gradient RMSE demonstrated insensitivity to mean pixel value and better correlation with visual assessment of bar patterns.
• ROI-HO metric showed increasing trends with regularization strength, saturating at intermediate levels, indicating diminishing returns for signal detection.

Conclusions:

• Gradient RMSE appears to correlate better with visual assessment than standard RMSE for DBT image reconstruction.
• The ROI-HO metric reflects visual trends in phantom reconstructions concerning regularization strength.
• Further data collection is necessary to fully establish the utility of these simulation-based metrics for DBT image quality assessment.