TH‐A‐103‐10: Improved Segmentation of Low‐Contrast Fibroglandular Structures in High‐Noise Breast CT Volumes for XCAT Modeling


Conference Paper

Purpose: This work improves the accuracy and realism of automated breast computed tomography (bCT) tissue segmentation by refining the detection of low‐contrast fibroglandular structures to produce high‐resolution realistic computer‐generated (XCAT) breast phantoms from empirical human subject data. Methods: Previous work by Hsu et al. [Med. Phys. 38, 5756‐5770 (2011)] produced high‐resolution realistic computer‐generated breast phantoms from empirical human subject data but challenges were encountered with the accurate segmentation of fine, low‐contrast glandular structures. The current work addresses those challenges. A 3‐D anisotropic diffusion algorithm was used to denoise fourteen bCT datasets. After breast masking, two adipose non‐uniformity correction techniques were applied. The first has been described by Altunbas, et al. [Med. Phys. 34, 3109‐3118 (2007)]. The second approach employed an original technique using higher‐order polynomials to correct for residual adipose non‐uniformity. Histogram thresholding then produced initial gland and skin segmentations. This was followed by a novel glandular linking and extension protocol based on skeletonization of the skin and glandular segmentations and a pixel gray‐level‐weighted distance transform. Skin mask definition and glandular density differentiation completed the segmentation. Results: Volumetric denoising reduced the standard deviation of the adipose background by an average of 60.4%. The Altunbas method corrected for radially symmetric, quadratic non‐uniformities in breasts with circular coronal cross sections, but performed poorly on high‐density breasts and breasts with asymmetric adipose non‐uniformity. Follow‐up correction using the novel method improved adipose uniformity by an average of 24.6%. The new fibroglandular linking and extension protocol improved the detection of low‐contrast fibroglandular structures, including Cooper's ligaments. The total number of fibroglandular tissue islands was also reduced. Conclusion: The semi‐automated bCT segmentation protocol improved low‐contrast glandular fiber detection in high‐noise reconstructions. Linking of disparate fibroglandular tissue islands and capture of Cooper's ligaments will contribute to the overall accuracy and realism of empirically‐derived XCAT breast phantoms. This work was supported by NIH Grant 5R01‐CA‐134658. © 2013, American Association of Physicists in Medicine. All rights reserved.

Full Text

Duke Authors

Cited Authors

  • Wells, J; Segars, P; Dobbins, J

Published Date

  • January 1, 2013

Published In

Volume / Issue

  • 40 / 6

Start / End Page

  • 527 -

International Standard Serial Number (ISSN)

  • 0094-2405

Digital Object Identifier (DOI)

  • 10.1118/1.4815727

Citation Source

  • Scopus