A Complex Digital Breast Phantom with 3D Tissue Compositions

  • Version: 1.0
  • Release date: July 6, 2015
  • Created by: Bin Deng <bdeng1 at nmr.mgh.harvard.edu>, and Qianqian Fang <fangq at nmr.mgh.harvard.edu>
  • License: The DigiBreast phantom and source data are in the public domain; the MATLAB scripts are covered under the BSD license. Please refer to the LICENSE_BSD.txt for proper use and redistribution of the contents.
  • Acknowledgement: The mammogram and glandularity source images were measured using a Philips MicroDose SI mammography system, contributed by Philips Healthcare.

1. Download
2. Introduction
3. What's in the package
3.1. Folder structure
3.2. DigiBreast phantom data
3.3. DigiBreast source data
3.4. Scripts
4. Tissue optical properties
5. Footnote
6. Reference

1. Download

Please download the latest release (Version 1) at our registration/download page.

2. Introduction

DigiBreast is a numerical breast phantom designed for 3D multi-physics simulations and validations of model-based image reconstruction algorithms for mammographically compressed breasts. The development of this phantom was described in Deng2015 with the original intent of testing a structure-prior guided image reconstruction algorithm for combined x-ray mammography and diffuse optical tomography (DOT) imaging. This digital breast phantom contains generic information such as 3D breast shapes and internal anatomical structures. We believe such breast phantom can address the needs for simulation-based validations for a wide range of model-based imaging modalities. Potential utilities of this digital phantom include, but not limited to, simulations of breast deformation, 2D and 3D x-ray breast imaging, and tomographic imaging of a compressed breast using tomographic optical, microwave, thermal and electrical impedance methods.

A unique aspect of this digital breast phantom is the inclusion of a realistic 3D glandularity map measured through a dual-energy x-ray mammography system, provided by Philips Healthcare. In comparison, conventional numerical breast phantoms represent various breast tissue constituents, i.e. the fibroglandular and adipose tissue, by piece-wise-constant regions (using a binary segmentation algorithm). Such representation removes the fine spatial details in the breast anatomical images, and results in loss of information. Statistical, or fuzzy segmentation methods avoid such information loss, and provide spatially-varying tissue volume fraction maps. In our previous works Fang2010, we have reported a joint x-ray/DOT image reconstruction algorithm utilizing a spatially varying tissue compositional model to improve DOT image resolution. This method was further characterized in Deng2015.

The DigiBreast phantom has limitations. While the breast shape is in 3D, the internal tissue compositional maps were derived from 2D x-ray measurements, thus, have an overall "cylindrical" shape along the sagittal direction. We, however, believe this approximation has negligible impact to most potential applications which deal with a mammographically compressed breast. This is because most of these methods utilizing a parallel-plate based measurement scheme and such scheme has an anisotropic spatial resolution - the horizontal/axial has significantly higher resolution than in the vertical/sagittal direction. Therefore, the focus in most of these imaging modalities are in the axial/horizontal view instead of the sagittal view. This limitation can be overcome in the future when 3D x-ray spectral imaging becomes available.

3. What's in the package

3.1. Folder structure

The DigiBreast package contains a "data" folder and a "script" folder, along with related documentation. In the JSON and UBJSON formatted packages, the data folder is replaced by either "json" or "ubjson". In either folder, MATLAB variables that encode the DigiBreast Phantom are saved as separate JSON and UBJSON formatted data files. These files can be loaded into MATLAB using the free JSONLab toolbox (http://iso2mesh.sf.net/jsonlab).

The package file structure is explained below.

├── AUTHORS.txt                   # Acknowledgement of contributions
├── README.txt                    # This file
├── data                          # DigiBreast data in MATLAB .mat format
│   └── DigiBreast.mat              # DigiBreast main data
│   └── DigiBreast_source.mat       # DigiBreast source data
├── json                          # DigiBreast data in JSON format (optional)
│   └── <<VariableName>>.json       # JSON files for each MATLAB variable
├── ubjson                        # DigiBreast data in UBJSON format (optional)
│   └── <<VariableName>>.ubj        # UBJSON files for each MATLAB variable
├── script                        # All related MATLAB scripts
│   ├── digibreast_meshrefine.m     # Creating the refined meshes at a given ROI
│   ├── digibreast_savejson.m       # Saving DigiBreast data in JSON and UBJSON
│   ├── digibreast_priors.m         # Creating tissue compositional maps
│   ├── digibreast_lesionprofile.m  # Creating a Gaussian-spherical tumor prior
│   └── digibreast_tablelookup.m    # Utility to lookup the optical properties
└── LICENSE_BSD.txt               # License file

3.2. DigiBreast phantom data

DigiBreast.mat is a MATLAB mat-file containing all essential components of the 3D digital breast phantom used in the simulation study as presented in the Deng2015 paper. It contains 4 data structures - ForwardMesh, ReconMesh, LesionCentroids, and OpticalProperties. This phantom is built on the source images included in DigiBreast_source.mat, and a 2 cm slab was added toward the chest wall.

a MATLAB structure containing three fields, namely "node", "elem",and "glandularity".

  • ForwardMesh.node: the node coordinate list of the forward mesh
  • ForwardMesh.elem: the tetrahedral element list of the forward mesh
  • ForwardMesh.glandularity: a struct containing the following fields:
    • ForwardMesh.glandularity.truth: the measured glandularity at each node
    • ForwardMesh.glandularity.empirical: the nodal glandularity list using an empirical segmentation algorithm
    • ForwardMesh.glandularity.dualgaussian: the nodal glandularity list using a dual-gaussian segmentation method
    • ForwardMesh.glandularity.thresholdp2: the nodal glandularity list using a threshold segmentation method with a 0.2% threshold
    • ForwardMesh.glandularity.threshold2: the nodal glandularity list using a threshold segmentation method with a 2% threshold

a MATLAB structure containing two fields, namely "node" and "elem".
  • ReconMesh.node: the node coordinate list of the reconstruction mesh
  • ReconMesh.elem: the tetrahedral element list of the reconstruction mesh

a MATLAB structure with two fields, "adipose" and "fibroglandular", containing the [x,y,z] lesion centroids (in mm) of the two simulated lesion locations (as used in the Deng2015 paper) within either adipose or fibroglandular tissue vicinity.

a 4x9 cell array with optical properties (HbO, HbR, HbT, SO2, scattering power and amplitude, reduced scattering coefficients at 690 nm and 830 nm) of adipose and fibroglandular tissues, as well as of malignant lesions. These optical properties are estimated based on mean values of reconstruction optical images for our previous clinical study published in Fang2011. Optical properties are all properly labeled within the variable, and should be easy to interpret. A function included in this package, "digibreast_tablelookup.m", can also be used to look up for any particular optical properties of a certain tissue type.

3.3. DigiBreast source data

DigiBreast_source.mat is a MATLAB mat-file that contains the anonymized and down-sampled (1 mm pixel resolution) 2D images of the original mammogram, glandularity and thickness maps. All original images are clinical measurements from a normal breast in the cranio-caudal view (CC view) using a Philips dual-energy mammographic system - MicroDose SI. The MAT-file includes variables Mammogram, Glandularity, ThicknessMap, and Registration. Users can choose to use our readily-built 3D DigiBreast phantom in DigiBreast.mat, or to create their own 3D realistic breast phantoms using different meshing settings based on the 2D source images in DigiBreast_source.mat.

A digital breast mammogram in the CC view (335x307 in 1x1 mm pixels). The mammogram has been masked to exclude skin region.

Fibroglandular tissue volume fraction map (335x307 in 1x1 mm pixels) derived directly from the MicroDose SI measurement. This is the "ground truth" glandularity referred in the Deng2015 paper. By stacking vertically and repeating this image, we can map the forward mesh nodes into this 3D glandularity profile using the Registration data structure below, and produce the subfield ForwardMesh.glandularity.truth in DigiBreast.mat.

the measured breast thickness map at each pixel location (335x307 in 1x1 mm pixels).

a 12 x 3 matrix representing the mapping between the mammogram image space and optical probe space for multi-modal imaging purposes. The odd-numbered rows are 6 key-points (x/y/z coordinates) in the mammogram-voxel-space, and the even-numbered rows are the corresponding key-points in the optical probe space (the same as the mesh coordinate space).

3.4. Scripts

Generate a Gaussian-sphere lesion profile at defined centroid.


To generate a Gaussian lesion profile that represents the volume fractions of a 5 mm FWHM lesion located within the adipose vicinity as shown in Deng2015 on the forward mesh

 plotmesh([ForwardMesh.node lesionprofile],ForwardMesh.elem,'z=15',...

Refine the input mesh within a spherical region centered at centroid.


To generate the refined mesh used in Deng2015 (see Table 1 in the paper for details)

 % interpolation of glandularity maps in the refined mesh
 mesh.value=[ForwardMesh.glandularity.truth ForwardMesh.glandularity.dualgaussian];
 plotmesh([mesh_refined.node mesh_refined.value(:,1)],mesh_refined.elem,'z=15',...
 plotmesh([mesh_refined.node mesh_refined.value(:,2)],mesh_refined.elem,'z=15',...

Generate tissue compositional priors for the DigiBreast phantom.


To generate 2-compositional normal tissue priors using glandularity map derived from dual gaussian segmentation algorithm

 plotmesh([ForwardMesh.node priors.normal(:,1)],ForwardMesh.elem,'z=15',...
 title('Adipose tissue volume fractions');colorbar;
 plotmesh([ForwardMesh.node priors.normal(:,2)],ForwardMesh.elem,'z=15',...
 title('Fibroglandular tissue volume fractions');colorbar;

To generate 3-compositional normal and lesion tissue priors using the same glandularity map derived from dual gaussian segmentation algorithm

 plotmesh([ForwardMesh.node priors.lesion(:,1)],ForwardMesh.elem,'z=15',...
 title('Adipose tissue volume fractions');colorbar;
 plotmesh([ForwardMesh.node priors.lesion(:,2)],ForwardMesh.elem,'z=15',...
 title('Fibroglandular tissue volume fractions');colorbar;
 plotmesh([ForwardMesh.node priors.lesion(:,3)],ForwardMesh.elem,'z=15',...
 title('Lesion tissue volume fractions');colorbar;

Export DigiBreast mesh data into JSON and UBJSON files.

Search optical properties from the OpticalProperties table by tissue and property names.



4. Tissue optical properties

With the provided fibroglandular tissue volume fraction map (variable Glandularity within DigiBreast_source.mat), users can freely build your own breast phantom by multiplying the optical properties of fibroglandular, adipose, and cancerous tissues to the volume fractions at each pixel/node.

Tissue type HbO (μM) HbR (μM) μs’ (mm−1)
at 690 nm at 830 nm
Adipose 13.84 4.81 0.851 0.713
Fibroglandular 18.96 6.47 0.925 0.775
Malignant 20.60 6.72 0.957 0.801

5. Footnote

The DigiBreast phantom main and source data is in the public domain. The MATLAB scripts under the "script" sub-folder have a BSD license. See LICENSE_BSD.txt for details.

Some of the scripts included in this package requires the installation of the "iso2mesh" and "JSONLab" toolboxes. To download these toolboxes:

If you use this DigiBreast phantom main or source data in your publication, please cite the phantom version number (currently Version 1) to avoid conflict to any further updates of this mesh. If you use DigiBreast data in your research, the authors are appreciated if you can cite the Deng2015 paper below in your related publications.

6. Reference

[Deng2015] B. Deng, D.H. Brooks, D.A. Boas, M. Lundqvist, and Q. Fang, "Characterization of structural-prior guided optical tomography using realistic breast models derived from dual-energy x-ray mammography," Biomedical Optics Express 6(7): 2366-79 (2015).

[Fang2010] Q. Fang, R.H. Moore, D. B. Kopans, D.A. Boas DA, “Compositional-prior-guided image reconstruction algorithm for multi-modality imaging,” Biomedical Optics Express, 1(1), 223-235 (2010)

[Fang2011] Q. Fang, J. Selb, S.A. Carp, G. Boverman, E.L. Miller, D.H. Brooks, R.H. Moore, D.B. Kopans and D.A. Boas, "Combined optical and X-ray tomosynthesis breast imaging," Radiology 258(1): 89-97 (2011).

Powered by Habitat