Unsupervised Malignant Mammographic Breast Mass Segmentation Algorithm Based On Pickard Markov Random Field
Abstract
In this paper, we present a probabilistic approach to segmentation
of malignant breast masses which have irregular
shape, spiculated margins and which may be embedded in
high density glandular tissue. First, we perform contrast enhancement
of the image using a simple logarithmic transformation.
Then, we derive a segmentation technique based
on a specific class of Markov random fields (MRFs) known
as Pickard random fields. As opposed to most MRF-based
methods which require complex and time-consuming computations,
our approach is simple, much faster and nearly unsupervised
as it only requires specification of the number of levels
on the MRF. Tests performed on 48 malignant masses extracted
from the INbreast database revealed that the proposed
approach yields superior results while being robust with respect
to the high variability of mammograms.