Quantifying Lesion Contrast-to-Noise Ratio In Simulated Digital Mammography Using the Victre Pipeline
Abstract
Purpose
To quantify lesion CNR in simulated digital mammography using the VICTRE pipeline and assess the effects of lesion size, lesion density, breast density, imaging energy, scatter, and ROI methodology.
Methods
Four anthropomorphic digital breast phantoms generated with VICTRE were selected to represent clinically distinct breast-density categories (fatty to extremely dense). A single spiculated mass was inserted at a fixed anatomical location in each phantom. Lesion size was varied across six α values (1.0–4.0 mm), and lesion density was varied across five values (1.06–1.25 g/cm³). Monte Carlo simulations generated projections at 24 kVp and 40 kVp, with exposure normalization informed by clinical AEC behavior. CNR was computed as the absolute difference between mean lesion and background signals divided by the background standard deviation. Two ROI methodologies were evaluated: (1) structured background ROIs placed superior or lateral to the lesion, including a fixed-geometry lateral ROI to stabilize comparisons across lesion size, and (2) an automated lesion-centered approach in which background ROIs were defined locally around the lesion while excluding lesion pixels. For each method, ROIs were held fixed across energies and scatter conditions.
Results
Across ROI methodologies, 24 kVp yielded higher CNR than 40 kVp, and primary-only images yielded higher CNR than primary-plus-scatter images (p < 0.001). Fixed-geometry structured ROIs and lesion-centered background ROIs demonstrated monotonic increases in CNR with lesion size and density. In contrast, structured ROIs whose geometry varied with lesion size produced greater CNR variability and less consistent monotonic behavior, reflecting sensitivity to local background heterogeneity.
Conclusion
These VICTRE simulations demonstrate strong, systematic CNR dependence on imaging energy and scatter, while observed lesion size and density trends depend on ROI methodology. Lesion-centered background sampling provides a physically local and reproducible approach for CNR evaluation in simulation-based mammography studies and establishes a baseline for subsequent image-quality and modeling investigations.