The signal curves in perfusion dynamic contrast enhanced MRI (DCE-MRI) of cancerous breast tissue reveal valuable
information about tumor angiogenesis. Pathological studies have illustrated that breast tumors consist of different subregions,
especially with more homogeneous properties during their growth. Differences should be identifiable in DCEMRI
signal curves if the characteristics of these sub-regions are related to the perfusion and angiogenesis. We introduce
a stepwise clustering method which in a first step uses a new similarity measure. The new similarity measure (PM)
compares how parallel washout phases of two curves are. To distinguish the starting point of the washout phase, a linear
regression method is partially fitted to the curves. In the next step, the minimum signal value of the washout phase is
normalized to zero. Finally, PM is calculated according to maximal variation among the point wise differences during
washout phases. In the second step of clustering the groups of signal curves with parallel washout are clustered using
Euclidean distance. The introduced method is evaluated on 15 DCE-MRI breast datasets with different types of breast
tumors. The use of our new heterogeneity analysis is feasible in single patient examination and improves breast MR
diagnostics.
A novel approach is introduced for clustering tumor regions with similar signal-time series measured by dynamic
contrast-enhanced (DCE) MRI to segment the tumor area in breast cancer. Each voxel of the DCE-MRI dataset is
characterized by a signal-time curve. The clustering process uses two describer values for each pixel. The first value is
L2-norm of each time series. The second value r is calculated as sum of differences between each pair of S(n-i) and S(i)
for i = {0...n/2} where S is the intensity and n the number of values in a time series. We call r reverse value of a time
series. Each time series is considered as a vector in an n-dimensional space and the L2-norm and reverse value of a vector
are used as similarity measures. The curves with similar L2-norms and similar reverse values are clustered together. The
method is tested on breast cancer DCE-MRI datasets with N = 256 x 256 spatial resolution and n = 128 temporal
resolution. The quality of each cluster is described through the variance of Euclidean distances of the vectors to the mean
vector of the corresponding cluster. The combination of both similarity measures improves the segmentation compared
to using each measure alone.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.