Breast glandularity and mean glandular dose assessment using a deep learning framework: Virtual patients study

Abstract

Breast dosimetry in mammography is an important aspect of radioprotection since women are exposed periodically to ionizing radiation due to breast cancer screening programs. Mean glandular dose (MGD) is the standard quantity employed for the establishment of dose reference levels in retrospective population studies. However, MGD calculations requires breast glandularity estimation. This work proposes a deep learning framework for volume glandular fraction (VGF) estimations based on mammography images, which in turn are converted to glandularity values for MGD calculations. Methods: 208 virtual breast phantoms were generated and compressed computationally. The mammography images were obtained with Monte Carlo simulations (MC-GPU code) and a ray-tracing algorithm was employed for labeling the training data. The architectures of the neural networks are based on the XNet and multilayer perceptron, adapted for each task. The network predictions were compared with the ground truth using the coefficient of determination (r 2 ). Results: The results have shown a good agreement for inner breast segmentation (r 2 = 0.999), breast volume prediction (r 2 = 0.982) and VGF prediction (r 2 = 0.935). Moreover, the DgN coefficients using the predicted VGF for the virtual population differ on average 1.3% from the ground truth values. Afterwards with the obtained DgN coefficients, the MGD values were estimated from exposure factors extracted from the DICOM header of a clinical cohort, with median(75 percentile) values of 1.91(2.45) mGy. Conclusion:We successfully implemented a deep learning framework for VGF and MGD calculations for virtual breast phantoms.

Virtual dosimetry study with three cone-beam breast computed tomography scanners using a fast GPU-based Monte Carlo code

“…Tables of D g N CT values are usually calculated with the homogeneous tissue approximation (i.e., where the phantom tissue is a homogeneous mixture of adipose and glandular tissues of given glandular fraction by mass), but they have been calculated also for anthropomorphic voxelized breast phantoms, where the voxels are either 100% adipose or 100% glandular, spatially distributed realistically in the breast phantom at a given value of glandular fraction . For DM and DBT dosimetry, the former approach was also adopted in Massera et al (2021aMassera et al ( , 2021b, Sarno et al (2021), Caballo et al (2022). Here, we adopt the latter method (as in Sarno et al 2022 and Mettivier et al 2022), using our set of patient-derived anthropomorphic breast phantoms.…”

Focus issue: Artificial intelligence in medical physics

“…Its estimation requires measurements of breast density, air kerma and Monte Carlo-based conversion coefficients. By use of deep neural networks Tomal et al [19] showed that both volumetric breast glandularity and mean glandular dose could be automatically assessed. The system was thoroughly validated by virtual anthropomorphic breast phantoms and other tools available in the literature.…”

Comparisons of glandular breast dose between digital mammography, tomosynthesis and breast CT based on anthropomorphic patient-derived breast phantoms

“…Such a homogeneous breast model is a simplification of the real condition of a 3D heterogeneous distribution of the glandular mass in the patient breast. MGD is adopted in worldwide quality assurance protocols both in DM and in DBT for assuring the accomplishment of regulatory dose limits and for imaging unit performance assessment [12][13][14][15][16] as well as for comparison of different solutions (DM vs DBT vs BCT) [17][18][19][20]. However, the adoption of MGD as a unique metric for dose assessment has critical aspects, both for its use in patient dose estimates and for the comparison of dose level of different imaging modalities.…”