Biomedical Engineering Reference
In-Depth Information
effectively as in other scans, although curves driven by regional information
are still robust in segmenting the overall profile.
5. From Figures 18 and 19, we can see that all six methods listed in [102] have
approximately the same tendencies. However, our method demonstrates a
slightly different performance in some scans. This is mainly because the
six other methods are pixel classification-based per se, and our method em-
ploys an integration approach, and so it is expected to perform differently
from the other six algorithms.
5.4. Conclusions
In this section we have proposed a new approach to segmenting the cerebral
cortex. We formulated the segmentation problem using a modified Chan-Vese
model, and combined it with statistical coupled edge detection. The knowledge
of the thickness of the cortex is considered to improve the accuracy of the results.
Convincing results are demonstrated with both simulated and real MR brain data.
It should be mentioned that our approach shares some features with Zeng's
approach [59]. Both methods use a level set method ribbon brain model. However,
we consider region information and modify the Chan-Vese model to accommodate
the ribbon structure of the brain, which improves the robustness the algorithm.
Furthermore, we only use one level set function, instead of two, which dramatically
decreases the computational expense in 3D cases.
Future work will include explicit modeling of the MR bias field to further
improve robustness in case of RF inhomogeneities. Also, it is well known that
level set methods are topologically independent, which may be disadvantageous
in brain segmentation. Implementation of topology-preserving level set methods
[97] may also further improve segmentation results.
6.
PRIOR-BASED CARDIAC VALVE SEGMENTATION USING
A GEODESIC SNAKE
6.1. Introduction
The algorithm presented in this section was originally developed for a project
for reconstructing the mitral valve leaflet from 3D ultrasound image sequences,
which would lead to a better understanding of cardiac valve mechanics and the
mechanisms of valve failure. Segmenting the valve efficiently could also aid
surgeons in diagnosis and analysis of cardiac valve disease.
Among medical imaging techniques, ultrasound is particularly attractive be-
cause of its good temporal resolution, noninvasiveness, and relatively low cost.
In clinical practice, segmentation of ultrasound images still relies on manual or
semiautomatic outlines produced by expert physicians, especially when it comes
to an object as complex as a cardiac valve. The large quantities of data in 3D vol-
