An accretion method to regularize digital breast tomosynthesis reconstruction

Fully3D 2023 Proceeding

L. Coito, K. Michielsen, I. Sechopoulos.

Digital breast tomosynthesis suffers from limited angle artifacts, which means that the representation of the distribution of fibro-glandular tissue is severely limited in the direction of the missing data. To recover the location of the tissue in the image, we present a regularization method based on the accretion of fibro-glandular tissue that is inspired by the particle accretion phenomenon with attractive interactions. The method is combined with a polychromatic reconstruction algorithm with material decomposition.

The key ingredient of our approach is the correspondence between the fraction of fibro-glandular tissue at each voxel with a number of particles with attractive interactions. In this manner, the particles within one voxel would be attracted to those of a neighbor, generating a redistribution among voxels and the tissue patterns present in breast phantoms.

Reconstruction performance was analyzed quantitative over 65 two-dimensional phantom slices. It was found that the relative error of the reconstructed glandularity was 11.4% on average, ranging between -48.8% and +32.4%, for glandularity values ranging between 0.07 and 0.62.
The Dice similarity coefficient of the fibro-glandular structures in the phantoms after reconstruction was 0.57 on average, ranging between 0.20 and 0.90.

Results indicate that the presented accretion method has shown to improve the results of the iterative reconstruction method by localizing some of the large structures of fibro-glandular tissue in the phantoms studied.

Tomographic Imaging

Overige afdelingen Imaging