Automatic segmentation of subcortical brain structures in MR images using information fusion

This paper reports a new automated method for the segmentation of internal cerebral structures using an information fusion technique. The information is provided both by images and expert knowledge, and consists in morphological, topological, and tissue constitution data. All this ambiguous, complementary and redundant information is managed using a three-step fusion scheme based on fuzzy logic. The information is first modeled into a common theoretical frame managing its imprecision and incertitude. The models are then fused and a decision is taken in order to reduce the imprecision and to increase the certainty in the location of the structures. The whole process is illustrated on the segmentation of thalamus, putamen, and head of the caudate nucleus from expert knowledge and magnetic resonance images, in a protocol involving 14 healthy volunteers. The quantitative validation is achieved by comparing computed, manually segmented structures and published data by means of indexes assessing the accuracy of volume estimation and spatial location. Results suggest a consistent volume estimation with respect to the expert quantification and published data, and a high spatial similarity of the segmented and computed structures. This method is generic and applicable to any structure that can be defined by expert knowledge and morphological images.     

Three-dimensional segmentation and interpolation of magnetic resonance brain images

The authors propose a method for the 3-D reconstruction of the brain from anisotropic magnetic resonance imaging (MRI) brain data. The method essentially consists in two original algorithms both for segmentation and for interpolation of the MRI data. The segmentation process is performed in three steps. A gray level thresholding of the white and gray matter tissue is performed on the brain MR raw data. A global white matter segmentation is automatically performed with a global 3-D connectivity algorithm which takes into account the anisotropy of the MRI voxel. The gray matter is segmented with a local 3-D connectivity algorithm. Mathematical morphology tools are used to interpolate slices. The whole process gives an isotropic binary representation of both gray and white matter which are available for 3-D surface rendering. The power and practicality of this method have been tested on four brain datasets. The segmentation algorithm favorably compares to a manual one. The interpolation algorithm was compared to the shaped-based method both quantitatively and qualitatively.     

Adaptive fuzzy segmentation of magnetic resonance images

An algorithm is presented for the fuzzy segmentation of two-dimensional (2-D) and three-dimensional (3-D) multispectral magnetic resonance (MR) images that have been corrupted by intensity inhomogeneities, also known as shading artifacts. The algorithm is an extension of the 2-D adaptive fuzzy C-means algorithm (2-D AFCM) presented in previous work by the authors. This algorithm models the intensity inhomogeneities as a gain field that causes image intensities to smoothly and slowly vary through the image space. It iteratively adapts to the intensity inhomogeneities and is completely automated. In this paper, we fully generalize 2-D AFCM to three-dimensional (3-D) multispectral images. Because of the potential size of 3-D image data, we also describe a new faster multigrid-based algorithm for its implementation. We show, using simulated MR data, that 3-D AFCM yields lower error rates than both the standard fuzzy C-means (FCM) algorithm and two other competing methods, when segmenting corrupted images. Its efficacy is further demonstrated using real 3-D scalar and multispectral MR brain images.     

Model-based 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images

Human investigators instinctively segment medical images into their anatomical components, drawing upon prior knowledge of anatomy to overcome image artifacts, noise, and lack of tissue contrast. The authors describe: 1) the development and use of a brain tissue probability model for the segmentation of multiple sclerosis (MS) lesions in magnetic resonance (MR) brain images, and 2) an empirical comparison of the performance of statistical and decision tree classifiers, applied to MS lesion segmentation. Based on MR image data obtained from healthy volunteers, the model provides prior probabilities of brain tissue distribution per unit voxel in a standardized 3-D "brain space". In comparison to purely data-driven segmentation, the use of the model to guide the segmentation of MS lesions reduced the volume of false positive lesions by 50-80%     

Optimization neural networks for the segmentation of magnetic resonance images

The application of the Hopfield neural network for the multispectral unsupervised classification of MR images is reported. Winner-take-all neurons were used to obtain a crisp classification map using proton density-weighted and T(2)-weighted images in the head. The preliminary studies indicate that the number of iterations needed to reach ;good' solutions was nearly constant with the number of clusters chosen for the problem.     

Topology-preserving tissue classification of magnetic resonance brain images

This paper presents a new framework for multiple object segmentation in medical images that respects the topological properties and relationships of structures as given by a template. The technique, known as topology-preserving, anatomy-driven segmentation (TOADS), combines advantages of statistical tissue classification, topology-preserving fast marching methods, and image registration to enforce object-level relationships with little constraint over the geometry. When applied to the problem of brain segmentation, it directly provides a cortical surface with spherical topology while segmenting the main cerebral structures. Validation on simulated and real images characterises the performance of the algorithm with regard to noise, inhomogeneities, and anatomical variations.     

Low-level segmentation of 3-D magnetic resonance brain images-a rule-based system

A rule-based, low-level segmentation system that can automatically identify the space occupied by different structures of the brain by magnetic resonance imaging (MRI) is described. Given three-dimensional image data as a stack of slices, it can extract brain parenchyma, cerebro-spinal fluid, and high-intensity abnormalities. The multiple feature environment of MR imaging is used to comput several low-level features to enhance the separability of voxels of different structures. The population distribution of each feature is considered and a confidence function is computed whose amplitude indicates the likelihood of a voxel, with a given feature value, being a member of a class of voxels. Confidence levels are divided into a set of ranges to define notions such as highly confident, moderately confident, and least confident. The rule-based system consists of a set of sequential stages in which partially segmented binary scenes of one stage guide the next stage. Some important low-level definitions and rules for a clinical imaging protocol are presented. The system is applied to several MR images.     

Segmentation of brain parenchyma and cerebrospinal fluid in multispectral magnetic resonance images

Presents a new method to segment brain parenchyma and cerebrospinal fluid spaces automatically in routine axial spin echo multispectral MR images. The algorithm simultaneously incorporates information about anatomical boundaries (shape) and tissue signature (grey scale) using a priori knowledge. The head and brain are divided into four regions and seven different tissue types. Each tissue type c is modeled by a multivariate Gaussian distribution N(mu(c),Sigma(c)). Each region is associated with a finite mixture density corresponding to its constituent tissue types. Initial estimates of tissue parameters {mu(c),Sigma(c )}(c=1,...,7) are obtained from k-means clustering of a single slice used for training. The first algorithmic step uses the EM-algorithm for adjusting the initial tissue parameter estimates to the MR data of new patients. The second step uses a recently developed model of dynamic contours to detect three simply closed nonintersecting curves in the plane, constituting the arachnoid/dura mater boundary of the brain, the border between the subarachnoid space and brain parenchyma, and the inner border of the parenchyma toward the lateral ventricles. The model, which is formulated by energy functions in a Bayesian framework, incorporates a priori knowledge, smoothness constraints, and updated tissue type parameters. Satisfactory maximum a posteriori probability estimates of the closed contour curves defined by the model were found using simulated annealing.     

Abnormality segmentation in brain images via distributed estimation

The aim of this paper is to introduce a novel semisupervised scheme for abnormality detection and segmentation in medical images. Semisupervised learning does not require pathology modeling and, thus, allows high degree of automation. In abnormality detection, a vector is characterized as anomalous if it does not comply with the probability distribution obtained from normal data. The estimation of the probability density function, however, is usually not feasible due to large data dimensionality. In order to overcome this challenge, we treat every image as a network of locally coherent image partitions (overlapping blocks). We formulate and maximize a strictly concave likelihood function estimating abnormality for each partition and fuse the local estimates into a globally optimal estimate that satisfies the consistency constraints, based on a distributed estimation algorithm. The likelihood function consists of a model and a data term and is formulated as a quadratic programming problem. The method is applied for automatically segmenting brain pathologies, such as simulated brain infarction and dysplasia, as well as real lesions in diabetes patients. The assessment of the method using receiver operating characteristic analysis demonstrates improvement in image segmentation over two-group analysis performed with Statistical Parametric Mapping (SPM).     

Interframe coding of magnetic resonance images

Presents a new interframe coding method for medical images, in particular magnetic resonance (MR) images. Until now, attempts in using interframe redundancies for coding MR images have been unsuccessful. The authors believe that the main reason for this is twofold: unsuitable interframe estimation models and the thermal noise inherent in magnetic resonance imaging (MRI). The interframe model used here is a continuous affine mapping based on (and optimized by) deforming triangles. The inherent noise of MRI is dealt with by using a median filter within the estimation loop. The residue frames are quantized with a zero-tree wavelet coder, which includes arithmetic entropy coding. This particular method of quantization allows for progressive transmission, which aside from avoiding buffer control problems is very attractive in medical imaging applications.     

Analysis of dynamic magnetic resonance images

Dynamic magnetic resonance (MR) imaging with contrast agents is a very promising technique for studying tissue perfusion in vivo. A temporal series of magnetic resonance images of the same slice are acquired following the injection of a contrast agent into the blood stream. The image intensity depends on the local concentration of the contrast agent, so that tissue perfusion can be studied by the image series. A new method of analyzing such series is described here. Nonparametric linear regression is used for modeling the image intensity along the series on a pixel by pixel basis. After modeling, some relevant quantities describing the time series are obtained and displayed as images. Due to its flexibility, this approach is preferred to parametric modeling when pathology is present since this can induce a wide spread of patterns for the pixel image intensity along time. Results of the application of the method to series of dynamic magnetic resonance images from ischaemic rat brains after the injection of the susceptibility agent Sprodiamide Inj. (Dy-DTPA-BMA) are shown and compared to results from a related known method.     

Markov random field segmentation of brain MR images

Describes a fully-automatic three-dimensional (3-D)-segmentation technique for brain magnetic resonance (MR) images. By means of Markov random fields (MRF's) the segmentation algorithm captures three features that are of special importance for MR images, i.e., nonparametric distributions of tissue intensities, neighborhood correlations, and signal inhomogeneities. Detailed simulations and real MR images demonstrate the performance of the segmentation algorithm. In particular, the impact of noise, inhomogeneity, smoothing, and structure thickness are analyzed quantitatively. Even single-echo MR images are well classified into gray matter, white matter, cerebrospinal fluid, scalp-bone, and background. A simulated annealing and an iterated conditional modes implementation are presented     

Neuromagnetic localization using magnetic resonance images

Ionic flow associated with neural activation of the brain produces a magnetic field, called the neuromagnetic field, that can be measured outside the head using a highly sensitive superconducting quantum interference device (SQUID)-based neuromagnetometer. Under certain conditions, the sources producing the neuromagnetic field can be localized from a sampling of the neuromagnetic field. Neuromagnetic measurements alone, however, do not contain sufficient information to visualize brain structure. Thus, it is necessary to combine neuromagnetic localization with an anatomical imaging technique such as magnetic resonance imaging (MRI) to visualize both function and anatomy in vivo. Using experimentally measured human neuromagnetic fields and magnetic resonance images, the authors have developed a technique to register accurately these two modalities and have applied the registration procedure to portray the spatiotemporal distribution of neural activity evoked by auditory stimulation.     

Feature-directed spiral coding of magnetic resonance images

The authors investigate the encoding of magnetic resonance (MR) images of the human body using various lossless techniques, and presents a new form of spiral encoding. The algorithm used relies partially on the overall shape of the bounding contour of the image in achieving the compression and uses a traditional run-based technique combined with an adaptive Huffman coder to encode the complete image. Comparisons are made between the feature-directed spiral encoding and the traditional paths; the latter include the scanning pattern associated with the normal raster scanned display and the path for a display that could be used in following a linearised quadtree encoding. The new method tracks the `greater' contour of the overall image and, once the path has been established and tuples recorded, the inner contours are automatically generated. The process is repeated for each of the inner contours with a reducing radius towards the centre. The results are given for the various techniques in terms of compression ratios. The new spiralling method achieves an approximate 5.29% saving over the traditional techniques and also gives structure to the compressed image     

Precise segmentation of 3-D magnetic resonance angiography

Accurate automatic extraction of a 3-D cerebrovascular system from images obtained by time-of-flight (TOF) or phase contrast (PC) magnetic resonance angiography (MRA) is a challenging segmentation problem due to the small size objects of interest (blood vessels) in each 2-D MRA slice and complex surrounding anatomical structures (e.g., fat, bones, or gray and white brain matter). We show that due to the multimodal nature of MRA data, blood vessels can be accurately separated from the background in each slice using a voxel-wise classification based on precisely identified probability models of voxel intensities. To identify the models, an empirical marginal probability distribution of intensities is closely approximated with a linear combination of discrete Gaussians (LCDG) with alternate signs, using our previous EM-based techniques for precise linear combination of Gaussian-approximation adapted to deal with the LCDGs. The high accuracy of the proposed approach is experimentally validated on 85 real MRA datasets (50 TOF and 35 PC) as well as on synthetic MRA data for special 3-D geometrical phantoms of known shapes.     

Improved model-based magnetic resonance spectroscopic imaging

Model-based techniques have the potential to reduce the artifacts and improve resolution in magnetic resonance spectroscopic imaging, without sacrificing the signal-to-noise ratio. However, the current approaches have a few drawbacks that limit their performance in practical applications. Specifically, the classical schemes use less flexible image models that lead to model misfit, thus resulting in artifacts. Moreover, the performance of the current approaches is negatively affected by the magnetic field inhomogeneity and spatial mismatch between the anatomical references and spectroscopic imaging data. In this paper, we propose efficient solutions to overcome these problems. We introduce a more flexible image model that represents the signal as a linear combination of compartmental and local basis functions. The former set represents the signal variations within the compartments, while the latter captures the local perturbations resulting from lesions or segmentation errors. Since the combined set is redundant, we obtain the reconstructions using sparsity penalized optimization. To compensate for the artifacts resulting from field inhomogeneity, we estimate the field map using alternate scans and use it in the reconstruction. We model the spatial mismatch as an affine transformation, whose parameters are estimated from the spectroscopy data.     

Reconstruction of magnetic resonance images using one-dimensional techniques

Whenever DFT (discrete Fourier transform) processing of a multidimensional discrete signal is required, one can apply either a multidimensional FFT (fast Fourier transform) algorithm, or a single-dimension FFT algorithm, both using the same number of points. That is, the dimensions of a "multidimensional" signal, and of its spectrum, are a matter of choice. Every multidimensional sequence is completely equivalent to a one-dimensional function in both "time" and "frequency" domains. This statement applied to MRI (magnetic resonance imaging) explains why one can reconstruct the slice by using either one-dimensional or two-dimensional methods, as it is already done in echo planar methods. In the commonly used spin warp methods, the image can be also reconstructed by either one- or two-dimensional processing. However, some artifacts in the images reconstructed from the original "zig-zag" echo planar trajectory, are shown to be due to the wrong dimensionality of the FFT applied.     

Performance-based classifier combination in atlas-based image segmentation using expectation-maximization parameter estimation

It is well known in the pattern recognition community that the accuracy of classifications obtained by combining decisions made by independent classifiers can be substantially higher than the accuracy of the individual classifiers. We have previously shown this to be true for atlas-based segmentation of biomedical images. The conventional method for combining individual classifiers weights each classifier equally (vote or sum rule fusion). In this paper, we propose two methods that estimate the performances of the individual classifiers and combine the individual classifiers by weighting them according to their estimated performance. The two methods are multiclass extensions of an expectation-maximization (EM) algorithm for ground truth estimation of binary classification based on decisions of multiple experts (Warfield et al., 2004). The first method performs parameter estimation independently for each class with a subsequent integration step. The second method considers all classes simultaneously. We demonstrate the efficacy of these performance-based fusion methods by applying them to atlas-based segmentations of three-dimensional confocal microscopy images of bee brains. In atlas-based image segmentation, multiple classifiers arise naturally by applying different registration methods to the same atlas, or the same registration method to different atlases, or both. We perform a validation study designed to quantify the success of classifier combination methods in atlas-based segmentation. By applying random deformations, a given ground truth atlas is transformed into multiple segmentations that could result from imperfect registrations of an image to multiple atlas images. In a second evaluation study, multiple actual atlas-based segmentations are combined and their accuracies computed by comparing them to a manual segmentation. We demonstrate in both evaluation studies that segmentations produced by combining multiple individual registration-based segmentations are more accurate for the two classifier fusion methods we propose, which weight the individual classifiers according to their EM-based performance estimates, than for simple sum rule fusion, which weights each classifier equally.     

Anatomic segmentation and volumetric calculations in nuclear magnetic resonance imaging

A comprehensive methodology for image segmentation is presented. Tools for differential and intensity contouring, and outline optimization are discussed, as well as the methods for automating such procedures. After segmentation, regional volumes and image intensity distributions can be determined. The methodology is applied to nuclear magnetic resonance images of the brain. Examples of the results of volumetric calculations for the cerebral cortex, white matter, cerebellum, ventricular system, and caudate nucleus are presented. An image intensity distribution is demonstrated for the cerebral cortex.