This work sits somewhat apart from other imaging studies, which focus on improving image quality, segmentation, or radiation therapy planning. Here, the focus is not on the image but on the dose. Neighboring works tackle image reconstruction and processing, while this paper optimizes the computational side of dosimetry. It is the first systematic check of how phantom resolution affects MGD accuracy specifically for synchrotron phase-contrast breast CT.
From images of four post-mastectomy breast specimens, voxel phantoms were built with a resolution of about 0.12 mm in-plane and 0.65–0.83 mm slice thickness. The phantoms were then downsampled by increasing the in-plane voxel size and slice thickness independently. For each variant, the mean glandular dose was computed using Monte Carlo (EGSnrc/DOSXYZnrc). Comparison against the high-resolution reference used a criterion of deviation no more than 4.5%. After choosing the optimal resolution, the number of primary photon histories was reduced from 3×10⁹ to 1×10⁸, and the statistical uncertainty was monitored.
The dose turned out to be much more sensitive to in-plane voxel size than to slice thickness. Increasing the in-plane size by a factor of 2.5–3 changed MGD by 5–9%, whereas a comparable increase in slice thickness changed it by only 1–3%. A voxel size of 0.30 × 0.30 × 3.00 mm³ kept MGD deviation within 3.6–4.4% for all four specimens. The number of voxels dropped by a factor of 20–31. Further reducing the number of photon histories to 2×10⁸ kept the statistical uncertainty below 0.3%. Total computation time fell from about 1600 minutes to 72 minutes.
Unlike neighboring studies that improve image quality or automate segmentation, this work optimizes dosimetry itself. It is not a new imaging method but a way to make an existing dose calculation fast enough to fit into the clinical workflow. Anisotropic downsampling (more along the thickness, less in-plane) is tested here for the first time for synchrotron phase-contrast breast CT with a full 360° rotation.
The next step is to determine what quality of a preliminary low-dose scan is sufficient to build a reduced-resolution phantom. If such a scan can be made fast and safe, the dose for a patient could be estimated before the main examination. Porting the computation to high-performance clusters could cut the time to minutes. The approach applies not only to synchrotron facilities but also to conventional breast CT scanners.
The results will affect clinical dosimetry in phase-contrast breast CT programs. They are also useful for any center that needs fast, patient-specific image-based dose estimates.
Determine the minimum quality requirements for a preliminary scan used to build the phantom. Test reproducibility on a larger number of patients and at other X-ray beam energies.
An open question in medical physics is how detailed a patient model must be for accurate dose estimation. This work shows that excessive detail is not always necessary and that a compromise between accuracy and speed can be found. It is part of the broader shift from average phantoms to personalized, real-time dosimetry.
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
In this study, voxel phantoms of the breast were built from images of four mastectomy specimens, then the phantoms were downsampled by increasing the in-plane voxel size and slice thickness independently. For each configuration, the mean glandular dose (MGD) was calculated using the Monte Carlo method (EGSnrc/DOSXYZnrc) and compared against a high-resolution reference, with a tolerance of no more than 4.5% deviation. After selecting the optimal resolution, the number of photon histories was reduced from 3×10⁹ to 1×10⁸, and the statistical uncertainty was monitored. It turned out that the dose is more sensitive to the in-plane voxel size than to the slice thickness; an overall voxel size of 0.30×0.30×3.00 mm³ kept the MGD deviation within 3.6–4.4% for all samples, while the number of voxels dropped by a factor of 20–31. Reducing the number of photon histories to 2×10⁸ kept the statistical uncertainty below 0.3%, and the total computation time decreased from about 1600 to 72 minutes.
Where its strength is
A systematic investigation of how phantom resolution affects MGD accuracy specifically for synchrotron phase-contrast breast CT with a full 360° rotation, revealing an anisotropic sensitivity (in-plane matters more than slice thickness) and achieving a practical 22-fold speedup without substantial loss of accuracy.
Where it could go next
Worth trying
Why this may matter
Fast MGD calculation with maintained accuracy opens the door to real-time dosimetry before a diagnostic scan, which could improve radiation safety and personalization of breast CT protocols.
Nearby works in our archive
The archive is sparse around this work: the closest review is only 0.54, against a usual 0.64. It looks like few have drilled nearby.
The original phantom of one breast contained up to 95 million voxels, and after optimization only about 3 million remained. Yet the difference in the calculated dose was less than half a percent of the target radiation level.