6th International Workshop on Breast Densitometry and Breast ...

6 th <strong>International</strong> <strong>Workshop</strong> on Breast Densitometry
and Breast Cancer Risk Assessment
P48
A FULLY-AUTOMATED METHOD FOR QUANTIFYING FIBROGLANDULAR
TISSUE AND BACKGROUND PARENCHYMAL ENHANCEMENT IN BREAST MRI
Shandong Wu, Susan P. Weinstein, Emily F. Conant, Despina Kontos
Department of Radiology, University of Pennsylvania, Philadelphia PA USA
Purpose: The goal is to provide a methodological framework for automated quantitative analysis of the
fibroglandular tissue (FGT) and background parenchymal enhancement (BPE) in breast magnetic
resonance imaging (MRI) for breast cancer risk assessment and risk reduction interventions in high-risk
women.
Background: Breast MRI provides 3D scanning and therefore can eliminate the tissue superposition
problem in mammography. Recent studies suggest that FGT and BPE are associated with breast cancer
risk assessment, specifically for high-risk women. However, currently FGT and BPE are mainly evaluated
qualitatively by visual assessment, which suffers from subjective results and inter- and intra-reader
variability.
ABSTRACTS
Method: Our fully automated method for quantifying FGT and BPE includes three steps (Fig. 1). First,
an integrated scheme of synergistic edge extraction has been developed to detect chest wall line so that
segment the whole breast. Second, the FGT is estimated based on our proposed fuzzy-c-means (FCM)-
Atlas method. Third, BPE is estimated through identifying the enhancing voxels in the DCE subtraction
(sub) images by measuring the relative intensity change relative to the intensity in the pre-contrast (pre)
images. We define an enhancement ratio R = Intensity_sub / Intensity_pre × 100 and the voxels that have
a greater value than a predefined threshold are identified as the enhancing voxels. Based on the
segmentation, we compute four measures: absolute volume of FGT (|FGT|), percentage of |FGT| relative
to the whole breast volume (FGT%), absolute volume of BPE (|BPE|), and percentage of |BPE| relative to
the whole breast volume (BPE%). Our method has been evaluated using a representative dataset of 60
bilateral breast MRI scans (120 breasts) through comparing to the manual segmentation obtained from
two experienced breast imaging radiologists by using our customized user-interactive software
(“MRwizard V.1.0”).
Results: The inter-reader Pearson correlation coefficient (r) is 0.95 for FGT% and 0.96 for |FGT|. When
compared to the average of the two readers, the proposed FGT segmentation method achieves a
correlation of r=0.88 for FGT% and r=0.84 for |FGT|; the bilateral correlation between left breasts and
right breasts is r=0.90 for FGT% and r=0.86 for |FGT|. For BPE-related measures, the inter-reader
correlation is r=0.80 for |BPE | and r=0.78 for BPE%, respectively. We parameterize the enhancement
ratio threshold ranging from 0% to 200% and the corresponding BPE is estimated. It is found that there is
a range, i.e., 30% - 80%, rather than a single value of the enhancement ratio that high correlations (r = 0.8
to 0.9) are achieved between the automated BPE estimation and the manual segmentation in terms of the
4 BPE-related measures (i.e., |BPE| and BPE% by 2 readers). Our fully automated method runs timeefficiently
at ~5 minute for each 3D MR scan (56 slices), compared to ~65 minutes needed for the manual
segmentation.
Conclusions: We have developed and validated a fully automated method for quantifying FGT and BPE
in breast MRI. It has been used successfully as a quantitative imaging biomarker for assessing the
response of risk-reducing salpingo-oophorectomy in BRCA1/2 mutation carriers and can also be used to
support other clinical applications.
83