Notes:
The amount of fibroglandular tissue content in the breast as
estimated mammographically, commonly referred to as breast percent
density (PD%), is one of the most significant risk factors for
developing breast cancer. Approaches to quantify breast density
commonly focus on either semiautomated methods or visual
assessment, both of which are highly subjective. This software
package was developed to be a fully-automated density estimation
method that works on both raw (i.e., "FOR PROCESSING") and vendor
postprocessed (i.e., "FOR PRESENTATION") digital mammography
images,and has thus far been validated to work on GE Healthcare and
Hologic digital mammography systems*.
Briefly, the software first applies an edge-detection algorithm to
delineate the boundary of the breast and the boundary of the
pectoral muscle. Following the segmentation of the breast, an
adaptive multi-class fuzzy c-means algorithm is applied to identify
and partition the mammographic breast tissue area, into multiple
regions (i.e., clusters) of similar intensity. These clusters are
then aggregated by a support-vector machine classifier to a final
dense tissue area, segmentation. The ratio of the segmented
absolute dense area to the total breast area is then used to obtain
a measure of breast percent density (PD%).
The software generates both quantitative estimates of breast area,
dense area and PD% that are stored in a comma separated text file
(.csv, openable by Excel) as well as a .JPG image of the breast and
density segmentations overlaid on a window-levelled version of the
mammogram amenable for publication to a user-defined directory, in
addition to several optional files, as described in the Manual
Section.
* DISCLAIMER: Density estimation on mammograms from other vendors has not been
validated, therefore the performance and the quality of
segmentation is not guaranteed. However the breast segmentation
algorithm within LIBRA generally works well across all vendors and
thus may be of general use in a research context.ust against
markers, clips and calcifications that are of high intensity in
mammograms.
Changes:
Public Release 1.0.3 (Dec 23rd 2015)
- Improved the training of the SVM models for GE mammograms hence
more reliable density segmentation in low density breasts.
Public Release 1.0.2 (Dec 1st 2015)
- Improved stability of airtheshold estimation in breast
segmentation.
- Addressed failure in breast segmentation when spacing paddles
present.
- Re-implemented fuzzy-cmean clustering.
- And other bug fixes.
Public Release 1.0.1 (Nov 23rd 2015)
- More robust against variations in dicom header.
- Removed age as a hard requirement of the dicom.
- And other bug fixes.
Public Release 1.0.0 (Oct 31st 2014)
- First stable public release.
- Improved the run-time speed by sub-sampling the image histogram
using its CDF.
- Added regression tests in libra_demo.m and a ground truth mat
file in Sample_Data/.
- Added a libra_version.m for versioning purpose.
- Added a libra_compile.m for binary compilation.
- Added a unified libra.m that works on one dicom and multiple
dicoms in one input directory. (Spinned off from
libra_batchProcessing.m)
- Merged the GUI into libra.m and greatly enhanced functionality
and usability of the GUI.
- No longer supports output breast mask in nifti format.
- Fixed the aspect ratio of the output jpg images.
|