Breast thickness map estimation and its associated correction in DBT imaging

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 243
  • Download : 0
DC FieldValueLanguage
dc.contributor.authorLee, Seoyoungko
dc.contributor.authorKim, Hyeongseokko
dc.contributor.authorLee, Hoyeonko
dc.contributor.authorCho, Seungryongko
dc.date.accessioned2021-08-03T08:30:22Z-
dc.date.available2021-08-03T08:30:22Z-
dc.date.created2021-08-02-
dc.date.issued2021-07-25-
dc.identifier.citationAAPM 2021 (63rd Annual Meeting & Exhibition)-
dc.identifier.urihttp://hdl.handle.net/10203/287012-
dc.description.abstractPurpose: Breast compression is still required during the digital breast tomosynthesis (DBT) image acquisition. Because of the rounded shape of a breast, it cannot be constricted into a single uniform thickness particularly near the periphery. Current segmentation-based thickness correction methods are error-prone and do not necessarily represent the x-ray physics. Methods: We propose a convolutional neural network (CNN) based breast thickness correction method. Not only a normalized log projection image but also a Euclidean distance map were prepared as network inputs. A distance map is generated by calculating the Euclidean distance of each internal pixel of the projection image to the closest point of the skin-line. We utilized patch-based learning to reduce computational costs whilst maintaining the image resolution of DBT projections. By providing distance map along with a log projection image, one can preserve the positional information of the patched image. We introduce a concept of the thickness map, which is a projection of a hypothetical binary volume mask that distinguishes inside and outside of a breast. We define a thickness corrected image by dividing the projection image with the corresponding thickness map in a pixel-wise manner. Thickness corrected images are the targets of the network training. Dataset for training and testing was constructed with help of the VICTRE phantom and MC-GPU simulation. Results: In the simulation study, the proposed network successfully corrected the breast thickness nonuniformity. A low root-mean-square deviation (NRMSE) down to 1.976% and high structural similarity (SSIM) up to 99.997% were obtained quantitatively. Conclusion: We have developed a CNN-based breast thickness correction method. By offering a distance map along with the projection, the proposed network successfully performed the breast thickness correction. Our method may benefit several applications including DBT image quality enhancement, scatter correction, density estimation, and multimodal imaging such as DBT plus diffuse optical tomography.-
dc.languageEnglish-
dc.publisherAmerican Association of Physicists in Medicine-
dc.titleBreast thickness map estimation and its associated correction in DBT imaging-
dc.typeConference-
dc.type.rimsCONF-
dc.citation.publicationnameAAPM 2021 (63rd Annual Meeting & Exhibition)-
dc.identifier.conferencecountryUS-
dc.identifier.conferencelocationVirtual-
dc.contributor.localauthorCho, Seungryong-
dc.contributor.nonIdAuthorLee, Hoyeon-
Appears in Collection
NE-Conference Papers(학술회의논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0