X‐ray scatter correction in breast tomosynthesis with a precomputed scatter map library

Abstract

Purpose: To develop and evaluate the impact on lesion conspicuity of a software-based x-ray scatter correction algorithm for digital breast tomosynthesis (DBT) imaging into which a precomputed library of x-ray scatter maps is incorporated. Methods: A previously developed model of compressed breast shapes undergoing mammography based on principal component analysis (PCA) was used to assemble 540 simulated breast volumes, of different shapes and sizes, undergoing DBT. A Monte Carlo (MC) simulation was used to generate the cranio-caudal (CC) view DBT x-ray scatter maps of these volumes, which were then assembled into a library. This library was incorporated into a previously developed software-based x-ray scatter correction, and the performance of this improved algorithm was evaluated with an observer study of 40 patient cases previously classified as BI-RADS R 4 or 5, evenly divided between mass and microcalcification cases. Observers were presented with both the original images and the scatter corrected (SC) images side by side and asked to indicate their preference, on a scale from −5 to +5, in terms of lesion conspicuity and quality of diagnostic features. Scores were normalized such that a negative score indicates a preference for the original images, and a positive score indicates a preference for the SC images. Results: The scatter map library removes the time-intensive MC simulation from the application of the scatter correction algorithm. While only one in four observers preferred the SC DBT images as a whole (combined mean score = 0.169 ± 0.37, p > 0.39), all observers exhibited a preference for the SC images when the lesion examined was a mass (1.06 ± 0.45, p < 0.0001). When the lesion examined consisted of microcalcification clusters, the observers exhibited a preference for the uncorrected images (−0.725 ± 0.51, p < 0.009). Conclusions:The incorporation of the x-ray scatter map library into the scatter correction algorithm improves the efficiency of the algorithm. The observer study presented here is also the first test of the scatter correction algorithm with patient images and human observers, and demonstrates its potential to improve the clinical performance of DBT.

Characterization of scatter in digital mammography from use of Monte Carlo simulations and comparison to physical measurements

“…5 Two publications by Feng et al also suggest that glandularity can be ignored while performing scatter correction (albeit in tomosynthesis images, which are acquired without a grid). 20,21 Our findings support their findings, as well as contributing additional data in which the simulations included a grid.…”

The compressed breast during mammography and breast tomosynthesis:in vivoshape characterization and modeling

“…Several areas of research in DM and DBT, such as many of the approaches for scatter correction (Sechopoulos et al , 2007b; Feng and Sechopoulos, 2011; Feng et al , 2014; Diaz et al , 2014; Kim et al , 2015) and thickness correction (Snoeren and Karssemeijer, 2004; Kallenberg and Karssemeijer, 2012) algorithms, require simulating realistic 3D compressed breast shapes (some approaches for scatter correction such as the one proposed by Kim et al (2015) do not require this). Moreover, other areas of research like patient dosimetry (Dance, 1990; Dance et al , 2000; Sechopoulos et al , 2007a; Sechopoulos et al , 2012), breast density estimation (Pertuz et al , 2016; Gubern-Merida et al , 2014), image registration and segmentation (Richard et al , 2006; Hipwell et al , 2016), and 3D breast software phantoms (Bakic et al , 2002; Bakic et al , 2011; Wang et al , 2012; O’Connor et al , 2013; Hsu et al , 2013; Kiarashi et al , 2015) could also benefit from objective shape models of compressed breasts to improve their accuracy and relevance.…”

“…For example, 2D projections are reduced to a semi-elliptical approximation for the cranio-caudal (CC) mammography view (Dance, 1990; Boone and Cooper, 2000) or a subjective model for the medio-lateral oblique (MLO) view (Sechopoulos et al , 2007a). Also, the 3D curvature of the compressed breast is also usually modeled as a semicircle (Snoeren and Karssemeijer, 2004; Kallenberg and Karssemeijer, 2012) or as an arbitrary polynomial (Feng et al , 2014). In our previous work (Feng et al , 2013; Rodriguez-Ruiz et al , 2017) we developed a 2D model of the projection of the outer shape of compressed breasts in digital mammograms, based on the objective analysis of a previously acquired database of mammograms.…”

A scatter correction method for contrast-enhanced dual-energy digital breast tomosynthesis

“…The reason for this is that the wing data for the LE case is essentially zero due to the narrowness of the LE spsf Within the object region we approximate ${\widehat{s}}_{\mathit{\text{LE}}}^{\theta }false(x,yfalse)$ as a constant $k_{\mathit{\text{LE}}}^{\theta }$. Numerous studies (Sechopoulos et al, 2007; Feng et al, 2014; Boone and Cooper III, 2000) have shown a slow spatial variation of low energy scatter in the object region, but we use a constant because the LE spatial variation is so much smaller than the HE scatter variation that the constant approximation yields good results. The remaining steps follow as in the HE case.…”

“…The two types of projections were combined to obtain a scatter estimate. In another approach used for DBT patient-specific SC (Sechopoulos et al , 2007; Feng and Sechopoulos, 2011; Feng et al , 2014) precomputed, using offline MC, an extensive “library” of scatter maps for a variety of breast shapes and thicknesses. This method required an extensive library of (about 10 5 entries) and complex image analysis techniques to match a given breast to a library entry to obtain the scatter estimate.…”