Conference Paper

Atlas-based segmentation of brain MR images using least square support vector machines

Dept. of Electr. & Electron. Eng., Shiraz Univ. of Technol., Shiraz, Iran
DOI: 10.1109/IPTA.2010.5586779 Conference: Image Processing Theory Tools and Applications (IPTA), 2010 2nd International Conference on
Source: IEEE Xplore

ABSTRACT This study presents an automatic model based technique for brain tissue segmentation from cerebral magnetic resonance (MR) images. In this paper, support vector machine (SVM) based classifier, as a new and powerful kind of supervised machine learning with high generalization characteristics, is employed. Here, least-square SVM (LS-SVM) in conjunction with brain probabilistic atlas as a priori information is applied to obtain class probabilities for three tissues of cerebrospinal fluid (CSF), white matter (WM) and grey matter (GM). The entire process of brain segmentation is performed in an iterative procedure, so that the probabilistic maps of brain tissues will be updated at any iteration. The quantitative and qualitative results indicate excellent performance of the applied method.

0 Bookmarks
 · 
64 Views
  • [Show abstract] [Hide abstract]
    ABSTRACT: Ground-truth annotations from the well-known Internet Brain Segmentation Repository (IBSR) datasets consider Sulcal cerebrospinal fluid (SCSF) voxels as gray matter. This can lead to bias when evaluating the performance of tissue segmentation methods. In this work we compare the accuracy of 10 brain tissue segmentation methods analyzing the effects of SCSF ground-truth voxels on accuracy estimations. The set of methods is composed by FAST, SPM5, SPM8, GAMIXTURE, ANN, FCM, KNN, SVPASEG, FANTASM, and PVC. Methods are evaluated using original IBSR ground-truth and ranked by means of their performance on pairwise comparisons using permutation tests. Afterward, the evaluation is repeated using IBSR ground-truth without considering SCSF. The Dice coefficient of all methods is affected by changes in SCSF annotations, especially on SPM5, SPM8 and FAST. When not considering SCSF voxels, SVPASEG (0.90 ± 0.01) and SPM8 (0.91 ± 0.01) are the methods from our study that appear more suitable for gray matter tissue segmentation, while FAST (0.89 ± 0.02) is the best tool for segmenting white matter tissue. The performance and the accuracy of methods on IBSR images vary notably when not considering SCSF voxels. The fact that three of the most common methods (FAST, SPM5, and SPM8) report an important change in their accuracy suggest to consider these differences in labeling for new comparative studies. J. Magn. Reson. Imaging 2014. © 2014 Wiley Periodicals, Inc.
    Journal of Magnetic Resonance Imaging 01/2014; · 2.57 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: In healthcare applications, there is tremendous growth in using the computer assistance for effective and fast diagnostic. There are various modalities such as Magnetic resonance imaging (MRI), computed tomography (CT), digital mammography, and others, to provide an insight of subject's body, noninvasively in order to facilitate diagnostic stakeholders to take decision in diagnosis. Being an important step of imaging systems in diagnostic, MRI imaging has been active area for researchers in computational intelligence and image processing. One of the most important problems in image processing and analysis is segmentation and same is true for biomedical imaging. The main objective of segmentation is separating the pixels associated with different types of tissues like white matter (WM), gray matter (GM) and cerebrospinal fluid (CSF). In this paper, we attempted to optimize the feature set constructed from more than three different types of features. It is well-known fact that, long feature vector representation can be boosting the performance. However, irrelevant feature elements from the long feature vector can become hurdle in convergence of classifier. The optimization feature vector is accomplished using genetic algorithm (GA) with an objective function of maximizing the sum of precision and recall. In addition to the elimination of the feature elements, some elements were also weighted to reduce their effect in the feature matching score. This overall process can also be considered as “fusion of features” for MRI segmentation.
    Advance Computing Conference (IACC), 2013 IEEE 3rd International; 01/2013
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper introduces a new brain Magnetic Resonance Imaging segmentation framework that combines a powerful multiresolution/multiscale image analysis technique with a robust weakly used ensemble learning paradigm. Firstly, the image is proceeded with the anisotropic diffusion filter to reduce the noise. Then, Stationary Wavelet Transform (SWT) is applied to get multiresolution/multiscale texture information. During the SWT stage, three levels of decomposition are used and four statistical features are computed around every voxel of each resulting sub-band. The feature extraction step allows to describe each voxel through a feature vector of 60 dimensions. Finally, the extracted features are used to feed a Random Forest classifier. To train and test this classifier, we make use of the Internet Brain Segmentation Repository database. The achieved results showed that our system outperforms other state of art methods for the segmentation of Gray Matter, White Matter, and Cerebrospinal Fluid
    Formal Pattern Analysis & Applications 04/2014; · 0.74 Impact Factor