This paper presents a new homogeneity measure for
variational segmentation with multiple level set functions.
We propose to modify the quadratic homogeneity
measure to trade off the convexity of the function against
a faster rate of convergence. We tested in two series of
experiments the performance of this new homogeneity
force at converging to appropriate partitioning of brain
MRI data sets, over a large range of image spatial
resolution and image quality, in terms of tissue
homogeneity and contrast.
Although numerous methods to segment brain MRI for
extraction of white matter (WM), gray matter (GM) and
cerebro-spinal fluid (CSF) have been proposed over the
past two decades, little work has been done to evaluate
and compare the performance of different segmentation
methods on real clinical data sets as well as the
performance of a single segmentation method on different
clinical data sets.