Model-based brain and tumor segmentation

Abstract

Combining image segmenrarion based on statistical classificarinn wirh a geometric prior has been shown rn signifrcanrly increase robustness and reproducibiliry. Using a probabilistic geometric model and image registration serves borh inirialization fj'probability densityfuncrions and de$nirion of sparial consrrainrs. A strong spatialprioz however: prevenrs segnientarion of srriicfares rhat ure not part of the model. Oirr driving application is the sgmenrarion of brain rissue and tumors from three-dimensional magneric resonance imaging (MRI). Our goal is a high-quality segmenration of borh healrhy tissire arid rumor: We presenr an exfension ro m exisrins expecrarion rnaxitnizarion ( E M ) sepnetzration algorirhnr rhar modifres U probabilistic brain atlas wirh an individual srrbject's information about tumor locarion obrained from subrrucrion of post-and pre-conrrasr MRI. The new' merhod handles various types of pathology, space-occrrpying mass rumors and infiltrating changes like edema. Preliminan results on five cases presenring rumor "pes wirh v e q different characrerisrics demonstrare the potential of rhe nebti technique for cliriical murine use forplanning and monitoring in neiirosnrgen, radiation oncology, and radiology

3D Brain Tumor Segmentation Using Fuzzy Classification and Deformable Models

“…Kaus et al [6] have proposed a method for automatic segmentation of small brain tumors using a statistical classification method and atlas registration. Moon et al [9] have also used the EM algorithm and atlas prior information for automatic tumor segmentation. These methods fail in the case of large deformations in the brain and they also require multichannel images (T1, T2, PD and contrast enhanced images) for classification.…”

Automated tumor segmentation using kernel sparse representations

“…However, it is sensitive to the contour initialization, and has a high computational cost due to its iterative nature. Model-based approaches [7] employ geometric priors to extend the Expectation Maximization (EM) algorithm to augment statistical classification. In relatively homogeneous cases such as low grade gliomas, the outlier detection framework proposed by Prastawa et al [2], [8] was shown to perform well.…”

Brain Tumors: How Can Images and Segmentation Techniques Help?

“…Also, bias field estimation is performed by including an extra term in all the mixture components that depends on the spatial positions of the voxels. Based on these ideas, (Moon et al, 2002) further extended the mixture modeling approach for segmenting brain tumors. These authors introduced a fully automatic method for segmenting MR images presenting tumor and edema, both mass-effect and infiltrating structures.…”

“…The initial fuzzy classification is used to guide locally the propagation direction and speed of a level-set-based contour, and is also used for automatically initializing the contour. (Prastawa et al, 2004) extended the approach described in (Moon et al, 2002). First, the detection of abnormal tissues was performed using a registered brain atlas (as a model of a normal brain) and a robust estimate for the location and dispersion of the normal brain tissues was calculated.…”