Biomedical Engineering Reference
In-Depth Information
(1) This method is easily extended to segmentation of vectorial images with
integration of the multiple channels information in the homogeneity mea-
sure. This property, described in detail in [45], has potential applications in
segmentation of multiprotocols MRI brain data sets or any co-registered
multimodality data sets where combination of spatial information can
assist the definition of a particular organ contours.
(2) This method is extensible to multiphases segmentation using a system of
n
coupled dynamic PDEs with
{
φ
1
,...,φ
n
}
level set functions defining 2
n
phases in the segmented data. Extensive description of the multiphase
method is provided in [41]. Potential applications of the multiphase for-
mulation include segmentation of brain MRIs into multiple tissue types.
An illustration of this application is provided in Fig. 2.5.
2.3
Joint Image Registration
and Segmentation
2.3.1
Motivations
Combining registration and segmentation has been motivated by the need to
incorporate prior information to guide and constrain the segmentation process.
The quality of the images acquired by the various medical screening modalities
is often poor due to the presence of multiple noise sources in the acquisition
system, degradation of data content during reconstruction processes (e.g., tomo-
graphic reconstruction with Radon transform), motion and respiratory artifacts
introduced by motion of the patient, and inherent limitations of system acqui-
sition accuracy. The combination of these factors degrade the signal to noise
ratio of the data, limit the spatial resolution, introduce inhomogeneities in the
tissue appearance across volumetric slices, and deteriorate boundary defini-
tions between specific organs and their surrounding tissues. These issues are
encountered with other medical imaging modalities such as ultrasound, MRI,
PET and SPECT and CT.
In the context of brain MRI segmentation for example, incorporation of atlas
information to assist the segmentation task of a particular data set has been a
very successful and popular approach for many years as reviewed in [46]. For
organs with very characteristics shapes such as cardiac ventricles, the corpus
