Breast ultrasound images offer several attractive properties which make
them suitable and an interesting tool for breast cancer detection. However,
due to their intrinsic high noise rate and low contrast properties, mass detection
and segmentation becomes a challenging task. In this paper, a
semi-automated two-stage breast mass segmentation method is proposed.
Initially, ultrasound images are segmentated using Support Vector Machines
or Discriminant Analysis with a multi-resolution pixel descriptor
extracted using non-linear diffusion, band-pass filtering and scale-space
curvature. A set of heuristic rules complements the initial segmentation
task, selecting the ROI in a fully automated manner. In the second segmentation
stage, two different methods are used to attempt a refined segmentation
of the area retrieved in the first stage. The first uses an AdaBoost
algorithm, using curvature measures and non-linear diffusion of
the original image at lower scales, and in the second active contours are
applied to improve the spatial resolution of the ROI.