Breast Cancer Risk Estimation Using Parenchymal Texture Analysis in Digital Breast Tomosynthesis

Abstract

Mammographic parenchymal texture has been shown to correlate with genetic markers of developing breast cancer. Digital breast tomosynthesis (DBT) is a novel x‐ray imaging technique in which tomographic images of the breast are reconstructed from multiple source projections acquired at different angles of the x‐ray tube. Compared to digital mammography (DM), DBT eliminates breast tissue overlap, offering superior parenchymal tissue visualization. We hypothesize that texture analysis in DBT could potentially provide a better assessment of parenchymal texture and ultimately result in more accurate assessment of breast cancer risk. As a first step towards validating this hypothesis, we investigated the association between DBT parenchymal texture and breast percent density (PD), a known breast cancer risk factor, and compared it to DM. Bilateral DBT and DM images from 71 women participating in a breast cancer screening trial were analyzed. Filtered‐backprojection was used to reconstruct DBT tomographic planes in 1 mm increments with 0.22 mm in‐plane resolution. Corresponding DM images were acquired at 0.1 mm pixel resolution. Retroareolar regions of interest (ROIs) equivalent to 2.5 cm3 were segmented from the DBT images and corresponding 2.5 cm2 ROIs were segmented from the DM images. Breast PD was mammographically estimated using the Cumulus scale. Overall, DBT texture features demonstrated a stronger correlation than DM to PD. The Pearson correlation coefficients for DBT were r = 0.40 (p<0.001) for contrast and r = −0.52 (p<0.001) for homogeneity; the corresponding DM correlations were r = 0.26 (p = 0.002) and r = ‐0.33 (p<0.001). Multiple linear regression of the texture features versus breast PD also demonstrated significantly stronger associations in DBT (R2 = 0.39) compared to DM (R2 = 0.33). We attribute these observations to the superior parenchymal tissue visualization in DBT. Our study is the first to perform DBT texture analysis in a screening population of women, showing that DBT could potentially provide better breast cancer risk assessment in the future.