This paper describes an automated algorithm for segmentation
of brain structures (CSF, white matter, and gray matter) in MR
images. We employ machine learning, i.e. k-Nearest Neighbors, of features
derived from k-means, Canny edge detection, and Tourist Walks
to fully automate the seeding process of the Random Walker algorithm.
We test our methods on a dataset of 12 diabetes patients with atrophy
and varying degrees of white matter lesions provided by the MRBrainS13
Challenge, and find encouraging segmentation performance.

Automated Walks using Machine Learning for Segmentation

This study investigates a method for brain tissue segmentation from 3D T1 weighted (T1w) MR images via convex relaxation with a hierarchical ordering constraint. It employs a multi-atlas-based initialization from 5 training images and is tested on 12 T1w MR images provided by the MICCAI 2013 MRBrainS segmentation challenge. The registered atlas images, fully segmented into eight different brain structures, are used to formulate shape and intensity priors for a Maximum A-Posteriori (MAP) energy that is subsequently minimized with a dual hierarchical max-flow computation. The algorithm makes use of a hierarchical label ordering constraint to regularize label families individually and its inherently globally optimal results guarantee robust segmentations. Major parts of the image processing pipeline are implemented using General-Purpose Programming on Graphics Processing Units (GPGPU) for a substantial increase in computation speed.

Multi-Atlas-based Segmentation with Hierarchical Max-Flow

 A fully automatic brain segmentation method is presented. First the skull is stripped using a model-based level set on T1-weighted inversion recovery images, then the brain ventricles and basal ganglia are segmented using the same method on T1-weighted images. The central white matter is segmented using a regular level set method but with high curvature regulation. To segment the cortical gray matter, a skeleton-based model is created by extracting the mid-surface of the gray matter from a preliminary segmentation using a threshold-based level set. An implicit model is then built by defining the thickness of the gray matter to be 2.7 mm. This model is incorporated into the level set framework and used to guide a second round more precise segmentation. Preliminary experiments show that the proposed method can provide relatively accurate results compared with the segmentation done by human observers. The processing time is considerably shorter than most conventional automatic brain segmentation methods. 

Fully automatic brain segmentation using model-guided level sets and skeleton-based models

We propose a new MAP-based technique for the unsupervised segmentation of different brain structures (white matter, gray matter, etc.) from T1-weighted MR brain images. In this paper, we follow a procedure like most conventional approaches, in which T1-weighted MR brain images and desired maps of regions (white matter, gray matter, etc.) are modeled by a joint Markov-Gibbs Random Field model (MGRF) of independent image signals and interdependent region labels. However, we specifically focus on the most accurate model identification that can be achieved. The proposed joint MGRF model accounts for the following three descriptors: i) a 1st-order visual appearance descriptor(empirical distribution of signal intensity), ii) a 3D probabilistic shape prior, and iii) a 3D spatially invariant 2nd-order homogeneity descriptor. To better specify the 1st-order visual appearance descriptor, each empirical distribution of signals is precisely approximated by a Linear Combination of Discrete Gaussians (LCDG) having both positive and negative components. The 3D probabilistic shape prior is learned using a subset of 3D co-aligned training T1-weighted MR brain images. The 2nd-order homogeneity descriptor is modeled by a 2nd-order translation and rotation invariant MGRF of 3D T1-weighted MR brain region labels with analytically estimated potentials. The initial segmentation, based on a 1st-order visual appearance and 3D probabilistic shape, is then iteratively refined using a 3D MGRF model with analytically estimated potentials. Experiments on twelve 3D T1-weighted MR brain images confirm the high accuracy of the proposed approach.

MAP–Based Framework for Segmentation of MR Brain Images Based on Visual Appearance and Prior Shape

Segmentation of the human cerebrum from magnetic resonance images (MRI) into its component tissues has been a defining problem in medical imaging. Until recently, this has been solved as the tissue classification of the T1-weighted (T1-w) MRI, with numerous solutions for this problem having appeared in the literature. The clinical demands of understanding lesions, which are indistinguishable from gray matter in T1-w images, has necessitated the incorporation of T2-weighted Fluid Attenuated Inversion Recovery (FLAIR) images to improve segmentation of the cerebrum. Many of the existing methods fail to handle the second data channel gracefully, because of assumptions about their model. In our new approach, we explore a model free algorithm which uses a classification technique based on ensembles of decision trees to learn the mapping from an image feature to the corresponding tissue label. We use corresponding image patches from a registered set of T1-w and FLAIR images with a manual segmentation to construct our decision
tree ensembles. Our method is efficient, taking less than two minutes on a 240x240x48 volume. We conduct experiments on five training sets in a leave-one-out fashion, as well as validation on an additional twelve subjects, and a landmark validation experiment on another cohort of five
subjects.

MR Brain Segmentation using Decision Trees

This work presents an automatic brain MRI segmentation method which can classify brain voxels into one of three main tissue types: gray matter (GM), white matter (WM) and Cerebro-spinal Fluid (CSF). Intensity-model based classification of MR images has proven problematic. The statistical approach does not carry any spatial, textural and neighborhood information in it. We propose to use a computationally fast and novel feature-set to facilitate voxel wise classification based on regional intensity, texture, spatial location of voxels in addition to posterior probability estimates. Information available through T1-weighted (T1), T1-weighted inversion recovery (IR) and T2-weighted FLAIR (FLAIR) MRI sequences was also leveraged. An aggregate overlap of 90.21% for all intracranial structures was reported between the automatic classification and available expert annotation as measured by the DICE coefficient.

Gaussian Intensity Model with Neighborhood Cues for Fluid-Tissue Categorization of Multi-Sequence MR Brain Images

We present a method for automated brain-tissue segmentation through voxelwise classification. Our algorithm uses manually labeled training images to train a support vector machine (SVM) classifier, which is then used for the segmentation of target images. The classification incorporates voxel intensities from a T1-weighted scan, an IR scan, and a FLAIR scan; features to encode the voxel position in the image; and Gaussian-scale-space features and Gaussian-derivative features at
multiple scales to facilitate a smooth segmentation.
An experiment on data from the MRBrainS13 brain-tissue-segmentation challenge showed that our algorithm produces reasonable segmentations in a reasonable amount of time.

Automated Brain-Tissue Segmentation by Multi-Feature SVM Classification

This work is integrated in the MICCAI Grand Challenge: MR Brain Image Segmentation 2013. It aims for the automatic segmentation of brain into Cerebrospinal fluid (CSF), Gray matter (GM) and White matter (WM). The provided dataset contains patients with white matter lesions, which makes the segmentation task more challenging. The proposed algorithm uses multi-sequence MR images to extract meaningful features and learn a Random Decision Forest that classifies each voxel of the image. The results show that it is robust to the presence of the white matter lesions, and the metrics show that the overall results are competitive.

Automatic Brain Tissue Segmentation of Multi-sequence MR Images Using Random Decision Forests

An automated method of MR Brain image segmentation is presented. A block based Expectation Maximization method is presented for the tissue classification of MR Brain images. The standard Gaussian Mixture Model is the most widely used method for MR Brain Image Segmentation and Expectation Maximization algorithm is used to estimate the model parameters. The Gaussian Mixture Model considers each pixel as independent and does not take into account the spatial correlation between the neighbouring pixels. Hence the segmentation result obtained using standard GMM is highly sensitive to Inensity Non-Uniformity and noise. The image is divided into blocks before applying EM since the GMM is preserved in the local image blocks. Also, Nonsubsampled Contourlet Transform is employed to incorporate the spatial correlation among the neighbouring pixels. The method is applied to the 12 MR Brain volumes of MRBRAINS13 test data and the White Matter, Gray Matter and CSF structures were segmented.

Modified Expectation Maximization Method for Automatic Segmentation of MR Brain Images

In this paper, we validate our proposed segmentation algorithm called Bayesian-based adaptive mean-shift (BAMS) on real mul-timodal MR images provided by the MRBrainS challenge. BAMS is a fully automatic unsupervised segmentation algorithm. It is based on the adaptive mean shift wherein the adaptive bandwidth of the kernel for each feature point is estimated using our proposed Bayesian approach [1]. BAMS is designed to segment the brain into three tissues; white matter (WM), gray matter (GM) and cerebrospinal fluid (CSF). The performance of the algorithm is evaluated relative to the manual segmentation (ground truth). The results of our proposed algorithm show the average Dice index 0.8377±0.036 for the WM, 0.7637±0.038 for the GM and 0.6835 ±0.023 for the CSF.