“School of Cognitive Sciences”
Back to Papers HomeBack to Papers of School of Cognitive Sciences
| Paper IPM / Cognitive Sciences / 11388 |
|
||||
| Abstract: | |||||
|
We propose a constrained, three-dimensional,
nonparametric, entropy-based, coupled, multi-shape
approach to segment subcortical brain structures from
magnetic resonance images (MRI). The proposed method
uses PCA to develop shape models that capture structural
variability. It integrates geometrical relationship between
different structures into the algorithm by coupling them
(limiting their independent deformations). On the other
hand, to allow variations among coupled structures, it
registers each structure separately when building the shape
models. It defines an entropy-based energy function, which
is minimized using quasi-Newton algorithm. To this end,
probability density functions (pdf) are estimated iteratively
using nonparametric Parzen window method. In the
optimization algorithm, constraints are used to improve
segmentation quality. These constraints are extracted from
training data. Sample results are given for the segmentation
of caudate, hippocampus, and putamen, illustrating highly
superior performance of the proposed method compared to
the most similar methods in the literature.
Download TeX format |
|||||
| back to top | |||||
