MRI simulation-based evaluation of image-processing and classification methods

REGION SIMILARITY RELATIONSHIP BETWEEN WATERSHED AND PENALIZED FUZZY HOPFIELD NEURAL NETWORK ALGORITHMS FOR BRAIN IMAGE SEGMENTATION

A robust image segmentation method that combines the watershed segmentation and penalized fuzzy Hopfield neural network (PFHNN) algorithms to minimize undesirable over-segmentation is described in this paper. This method incorporates spatial graph representation derived from the watershed segmented regions and cluster analysis information obtained from the PFHNN algorithm to achieve efficient image segmentation. The proposed scheme employs the Markov random field (MRF) model on the region adjacency graph (RAG) to improve the quality of watershed segmentation. Here, the fusion criterion is according to the correlation coefficient, which uses inter-region similarities to determine the merging of regions. Analysis of the performance of the proposed technique is presented through quantitative and qualitative validation experiments on benchmark images, and significant and promising segmentation results are presented using brain phantom simulated data.

WATERSHED SEGMENTATION WITH AUTOMATIC ALTITUDE SELECTION AND REGION MERGING BASED ON THE MARKOV RANDOM FIELD MODEL

Watershed transformation has proven to be an important tool in image analysis. However, the resulting image of watershed transformation is inevitably over-segmented due to the presence of noise or local irregularities in the input image. In this paper, the use of contour altitude at the immersion stage is proposed. Block gradient information computed from the input gradient image is defined and used to obtain a critical altitude of watershed flooding. This altitude is then refined based on entropy estimated from the intermediate segmentation result. Thereafter, an optimal altitude and its corresponding segmentation result can be obtained. Although this process can favorably reduce the number of regions, the quality of segmentation still requires further improvement. Hence, a Markov Random Field (MRF) model defined on a region adjacency graph (RAG) is adopted. Because the MRF model can merge neighboring regions that are similar in local statistic properties, it thus alleviates the over-segmentation problem and improves the quality of image segmentation. In the experimental studies, the proposed method has been tested using several benchmark images. It achieves improved appearance and energy indices in comparison with the results obtained by conventional methods.

Constrained reverse diffusion for thick slice interpolation of 3D volumetric MRI images

Due to physical limitations inherent in magnetic resonance imaging scanners, three dimensional volumetric scans are often acquired with anisotropic voxel resolution. We investigate several interpolation approaches to reduce the anisotropy and present a novel approach - constrained reverse diffusion for thick slice interpolation. This technique was compared to common methods: linear and cubic B-Spline interpolation and a technique based on non-rigid registration of neighboring slices. The methods were evaluated on artificial MR phantoms and real MR scans of human brain. The constrained reverse diffusion approach delivered promising results and provides an alternative for thick slice interpolation, especially for higher anisotropy factors.

A multiscale and multiblock fuzzy C-means classification method for brain MR images

PURPOSE

Classification of magnetic resonance (MR) images has many clinical and research applications. Because of multiple factors such as noise, intensity inhomogeneity, and partial volume effects, MR image classification can be challenging. Noise in MRI can cause the classified regions to become disconnected. Partial volume effects make the assignment of a single class to one region difficult. Because of intensity inhomogeneity, the intensity of the same tissue can vary with respect to the location of the tissue within the same image. The conventional "hard" classification method restricts each pixel exclusively to one class and often results in crisp results. Fuzzy C-mean (FCM) classification or "soft" segmentation has been extensively applied to MR images, in which pixels are partially classified into multiple classes using varying memberships to the classes. Standard FCM, however, is sensitive to noise and cannot effectively compensate for intensity inhomogeneities. This paper presents a method to obtain accurate MR brain classification using a modified multiscale and multiblock FCM.

METHODS

An automatic, multiscale and multiblock fuzzy C-means (MsbFCM) classification method with MR intensity correction is presented in this paper. We use a bilateral filter to process MR images and to build a multiscale image series by increasing the standard deviation of spatial function and by reducing the standard deviation of range function. At each scale, we separate the image into multiple blocks and for every block a multiscale fuzzy C-means classification method is applied along the scales from the coarse to fine levels in order to overcome the effect of intensity inhomogeneity. The result from a coarse scale supervises the classification in the next fine scale. The classification method is tested with noisy MR images with intensity inhomogeneity.

RESULTS

Our method was compared with the conventional FCM, a modified FCM (MFCM) and multiscale FCM (MsFCM) method. Validation studies were performed on synthesized images with various contrasts, on the simulated brain MR database, and on real MR images. Our MsbFCM method consistently performed better than the conventional FCM, MFCM, and MsFCM methods. The MsbFCM method achieved an overlap ratio of 91% or higher. Experimental results using real MR images demonstrate the effectiveness of the proposed method. Our MsbFCM classification method is accurate and robust for various MR images.

CONCLUSIONS

As our classification method did not assume a Gaussian distribution of tissue intensity, it could be used on other image data for tissue classification and quantification. The automatic classification method can provide a useful quantification tool in neuroimaging and other applications.

Maternal-fetal blood incompatibility and neuromorphologic anomalies in schizophrenia: Preliminary findings

Prior research has shown that maternal-fetal Rhesus (Rh) and ABO blood incompatibility increase the risk for schizophrenia. In the present study, the relationship between blood incompatibility and volumes of brain structures previously implicated in schizophrenia was assessed in schizophrenia cases and controls from a large birth cohort. Rh/ABO incompatible cases had significantly reduced cortical gray matter volume compared to compatible cases, a finding which appears to be driven by significant volume reductions in the dorsolateral prefrontal cortex and inferior frontal cortex. Larger hippocampal and putamen volumes were also observed in exposed controls compared to unexposed controls. Although the sample size is small and replications are required, these data suggest that maternal-fetal blood incompatibility may increase the risk for altered brain morphology in both schizophrenia and in controls. The findings also suggest that the larger hippocampal volume in exposed controls may indicate a mechanism of adaptive resilience which diminishes the risk that controls will develop schizophrenia.

A 3D wavelet fusion approach for the reconstruction of isotropic-resolution MR images from orthogonal anisotropic-resolution scans

Hardware constraints, scanning time limitations, patient movement, and signal-to-noise ratio (SNR) considerations, restrict the slice-selection and the in-plane resolutions of MRI differently, generally resulting in anisotropic voxels. This nonuniform sampling can be problematic, especially in image segmentation and clinical examination. To alleviate this, the acquisition is divided into (two or) three separate scans, with higher in-plane resolutions and thick slices, yet orthogonal slice-selection directions. In this work, a noniterative wavelet-based approach for combining the three orthogonal scans is adopted, and its advantages compared with other existing methods, such as Fourier techniques, are discussed, including the consideration of the actual pulse response of the MRI scanner, and its lower computational complexity. Experimental results are shown on simulated and real 7 T MRI data.

Implementation of an absolute brain 1H-MRS quantification method to assess different tissue alterations in multiple sclerosis

Magnetic resonance spectroscopy has emerged as a sensitive modality to detect early and diffuse alterations in multiple sclerosis. Recently, the hypothesis of neurodegenerative pathogenesis has highlightened the interest for measurement of metabolites concentrations, to gain specificity, in a large brain volume encompassing different tissue alterations. Therefore, we proposed in this paper the implementation of an absolute quantification method based on localized spectroscopy at short (30 ms) and long (135 ms) echo time of a volume including normal appearing white matter, cortical gray matter, and lesions. First, methodological developments were implemented including external calibration, and corrections of phased-array coil sensitivity and cerebrospinal fluid volume contribution. Second, these improvements were validated and optimized using an original methodology based on simulations of brain images with lesions. Finally, metabolic alterations were assessed in 65 patients including 26 relapsing-remitting, 17 primary-progressive (PP), 22 secondary-progressive (SP) patients, and in 23 normal subjects. Results showed increases of choline, creatine, and myo-inositol concentrations in PP and SP patients compared to controls, whereas the concentration of N-acetyl compounds remained constant. The major finding of this study was the identification of Cho concentration and Cho/tNA ratio as putative markers of progressive onset, suggesting interesting perspectives in detection and followup of neurodegenerative processes.

On the convergence of EM-like algorithms for image segmentation using Markov random fields

Inference of Markov random field images segmentation models is usually performed using iterative methods which adapt the well-known expectation-maximization (EM) algorithm for independent mixture models. However, some of these adaptations are ad hoc and may turn out numerically unstable. In this paper, we review three EM-like variants for Markov random field segmentation and compare their convergence properties both at the theoretical and practical levels. We specifically advocate a numerical scheme involving asynchronous voxel updating, for which general convergence results can be established. Our experiments on brain tissue classification in magnetic resonance images provide evidence that this algorithm may achieve significantly faster convergence than its competitors while yielding at least as good segmentation results.

Applying a patient-specific bio-mathematical model of glioma growth to develop virtual [18F]-FMISO-PET images

Glioblastoma multiforme (GBM) is a class of primary brain tumours characterized by their ability to rapidly proliferate and diffusely infiltrate surrounding brain tissue. The aggressive growth of GBM leads to the development of regions of low oxygenation (hypoxia), which can be clinically assessed through [18F]-fluoromisonidazole (FMISO) positron emission tomography (PET) imaging. Building upon the success of our previous mathematical modelling efforts, we have expanded our model to include the tumour microenvironment, specifically incorporating hypoxia, necrosis and angiogenesis. A pharmacokinetic model for the FMISO-PET tracer is applied at each spatial location throughout the brain and an analytical simulator for the image acquisition and reconstruction methods is applied to the resultant tracer activity map. The combination of our anatomical model with one for FMISO tracer dynamics and PET image reconstruction is able to produce a patient-specific virtual PET image that reproduces the image characteristics of the clinical PET scan as well as shows no statistical difference in the distribution of hypoxia within the tumour. This work establishes proof of principle for a link between anatomical (magnetic resonance image [MRI]) and molecular (PET) imaging on a patient-specific basis as well as address otherwise untenable questions in molecular imaging, such as determining the effect on tracer activity from cellular density. Although further investigation is necessary to establish the predicitve value of this technique, this unique tool provides a better dynamic understanding of the biological connection between anatomical changes seen on MRI and biochemical activity seen on PET of GBM in vivo.

A MR Brain Classification Method Based on Multiscale and Multiblock Fuzzy C-means

A fully automatic, multiscale and multiblock fuzzy C-means (MsbFCM) classification method with intensity correction for MR images is presented in this paper. We use a bilateral filter to process MR images and to build a multiscale image series by increasing the standard deviation of spatial function and reducing the standard deviation of range function. We separate every scale image into multiple blocks and for every block a multiscale fuzzy C-means classification method is applied along the scales from the coarse to fine levels to overcome the effect of intensity inhomogeneity. The method is robust for noise MR images with intensity inhomogeneity because of its multiscale and multiblock bilateral filtering scheme. Our method was compared with the conventional FCM, a modified FCM (MFCM) and multiscale FCM (MsFCM) method on synthesized images, simulated brain MR images, and real MR images. The MsbFCM method achieved an overlap ratio of greater than 91% as validated by the ground truth even if original images have 9% noise and 40% intensity inhomogeneity. Experimental results using real MR images demonstrate the effectiveness of the proposed method. Our MsbFCM classification method is accurate and robust for various MR images.

Realistic simulation of cardiac magnetic resonance studies modeling anatomical variability, trabeculae, and papillary muscles

Simulated magnetic resonance imaging brain studies have been generated for over a decade. Despite their useful potential, simulated cardiac studies are only emerging. This article focuses on the realistic simulation of cardiac magnetic resonance imaging datasets. The methodology is based on the XCAT phantom, which is modified to increase realism of the simulated images. Modifications include the modeling of trabeculae and papillary muscles based on clinical measurements and published data. To develop and evaluate our approach, the clinical database included 40 patients for anatomical measurements, 10 patients for papillary muscle modeling, and 10 patients for local gray value statistics. The virtual database consisted of 40 digital voxel phantoms. Histograms from different tissues were obtained from the real datasets and compared with histograms of the simulated datasets with the Chi-square dissimilarity metric (χ(2)) and Kullback-Leibler divergence. For the original phantom, χ(2) values averaged 0.65 ± 0.06 and Kullboek-Leibler values averaged 0.69 ± 0.38. For the modified phantom, χ(2) values averaged 0.34 ± 0.12 and Kullboek-Leibler values averaged 0.32 ± 0.15. The proposed approach demonstrated a noticeable improvement of the local appearance of the simulated images with respect to the ones obtained originally.

Simulation of High-Resolution Magnetic Resonance Images on the IBM Blue Gene/L Supercomputer Using SIMRI

Medical imaging system simulators are tools that provide a means to evaluate system architecture and create artificial image sets that are appropriate for specific applications. We have modified SIMRI, a Bloch equation-based magnetic resonance image simulator, in order to successfully generate high-resolution 3D MR images of the Montreal brain phantom using Blue Gene/L systems. Results show that redistribution of the workload allows an anatomically accurate 256³ voxel spin-echo simulation in less than 5 hours when executed on an 8192-node partition of a Blue Gene/L system.

Simulation of late gadolinium enhancement cardiac magnetic resonance studies

In this study we propose a pipeline for simulation of late gadolinium enhancement images. We used a modified version of the XCAT phantom to improve simulation realism. Modifications included the modeling of trabeculae and papillary muscles, and the increase of sublabels to resemble tissue intensity variability. Magnetic properties for each body tissue were sampled in three settings: from Gaussian distributions, combining Rayleigh-Gaussian distributions, and from Rayleigh distributions. Thirty-two simulated datasets were compared with 32 clinical datasets from infarcted patients. Histograms were obtained for five tissues: lung, pericardium, myocardium, blood and hyper-enhanced area. Real and simulated histograms were compared with the Chi-square dissimilarity metric (χ(2)) and Kullback-Leibler divergence (KL). The generated simulated images look similar to real images according to both metrics. Rayleigh and the Rayleigh-Gaussian models obtained comparable average results (respectively: χ(2)= 0.16 ± 0.12 and 0.18 ± 0.11; KL=0.15 ± 0.17 and 0.16 ± 0.18).

A wavelet multiscale denoising algorithm for magnetic resonance (MR) images

Based on the Radon transform, a wavelet multiscale denoising method is proposed for MR images. The approach explicitly accounts for the Rician nature of MR data. Based on noise statistics we apply the Radon transform to the original MR images and use the Gaussian noise model to process the MR sinogram image. A translation invariant wavelet transform is employed to decompose the MR 'sinogram' into multiscales in order to effectively denoise the images. Based on the nature of Rician noise we estimate noise variance in different scales. For the final denoised sinogram we apply the inverse Radon transform in order to reconstruct the original MR images. Phantom, simulation brain MR images, and human brain MR images were used to validate our method. The experiment results show the superiority of the proposed scheme over the traditional methods. Our method can reduce Rician noise while preserving the key image details and features. The wavelet denoising method can have wide applications in MRI as well as other imaging modalities.

Decoupled active contour (DAC) for boundary detection

The accurate detection of object boundaries via active contours is an ongoing research topic in computer vision. Most active contours converge toward some desired contour by minimizing a sum of internal (prior) and external (image measurement) energy terms. Such an approach is elegant, but suffers from a slow convergence rate and frequently misconverges in the presence of noise or complex contours. To address these limitations, a decoupled active contour (DAC) is developed which applies the two energy terms separately. Essentially, the DAC consists of a measurement update step, employing a Hidden Markov Model (HMM) and Viterbi search, and then a separate prior step, which modifies the updated curve based on the relative strengths of the measurement uncertainty and the nonstationary prior. By separating the measurement and prior steps, the algorithm is less likely to misconverge; furthermore, the use of a Viterbi optimizer allows the method to converge far more rapidly than energy-based iterative solvers. The results clearly demonstrate that the proposed approach is robust to noise, can capture regions of very high curvature, and exhibits limited dependence on contour initialization or parameter settings. Compared to five other published methods and across many image sets, the DAC is found to be faster with better or comparable segmentation accuracy.

Design and construction of a brain phantom to simulate neonatal MR images

This paper presents the design and construction of a 3D digital neonatal neurocranial phantom and its application for the simulation of brain magnetic resonance (MR) images. Commonly used digital brain phantoms (e.g. BrainWeb) are based on the adult brain. With the growing interest in computer-aided methods for neonatal MR image processing, there is a growing demand a digital phantom and brain MR image simulator especially for the neonatal brains. This is due to the pronounced differences between adult and neonatal brains not only in terms of size but also, more importantly, in terms of geometrical proportions and the need to subdivide white matter into two different tissue types in neonates. Therefore the neonatal brain phantom created in the here presented work consists of 9 different tissue types: skin, fat, muscle, skull, dura mater, gray matter, myelinated white matter, nonmyelinated white matter and cerebrospinal fluid. Each voxel has a vector consisting of 9 components, one for each of these nine tissue types. This digital phantom can be used to map simulated magnetic resonance signal intensities resulting in simulated MR images of the newborns head. These images with controlled degradation of the image data present a representative, reproducible data set ideal for development and evaluation of neonatal MRI analysis methods, e.g. segmentation and registration algorithms.

Segmentation priors from local image properties: without using bias field correction, location-based templates, or registration

We present a novel approach for generating information about a voxel's tissue class membership based on its signature--a collection of local image textures estimated over a range of neighborhood sizes. The approach produces a form of tissue class priors that can be used to initialize and regularize image segmentation. The signature-based approach is a departure from current location-based methods, which derive tissue class likelihoods based on a voxel's location in standard template space. To use location-based priors, one needs to register the volume in question to the template space, and estimate the image intensity bias field. Two optimizations, over more than a thousand parameters, are needed when high order nonlinear registration is employed. In contrast, the signature-based approach is independent of volume orientation, voxel position, and largely insensitive to bias fields. For these reasons, the approach does not require the use of population derived templates. The prior information is generated from variations in image texture statistics as a function of spatial scale, and an SVM approach is used to associate signatures with tissue types. With the signature-based approach, optimization is needed only during the training phase for the parameter estimation stages of the SVM hyperplanes, and associated PDFs; a training process separate from the segmentation step. We found that signature-based priors were superior to location-based ones aligned under favorable conditions, and that signature-based priors result in improved segmentation when replacing location-based ones in FAST (Zhang et al., 2001), a widely used segmentation program. The software implementation of this work is freely available as part of AFNI http://afni.nimh.nih.gov.

Fast algorithm for calculation of inhomogeneity gradient in magnetic resonance imaging data

PURPOSE

To develop and implement a new approach for correcting the intensity inhomogeneity in magnetic resonance imaging (MRI) data.

MATERIALS AND METHODS

The algorithm is based on the assumption that intensity inhomogeneity in MR data is multiplicative and smoothly varying. Using a statistically stable method, the algorithm first calculates the partial derivative of the inhomogeneity gradient across the data. The algorithm then solves for the gradient field and fits it to a parametric surface. It was tested on both simulated and real human and animal MRI data.

RESULTS

The algorithm is shown to restore the homogeneity in all images that were tested. On real human brain images the algorithm demonstrated superior or comparable performance relative to some of the commonly used intensity inhomogeneity correction methods such as SPM, BrainSuite, and N3.

CONCLUSION

The proposed algorithm provides an alternative method for correcting the intensity inhomogeneity in MR images. It is shown to be fast and its performance is superior or comparable to algorithms described in the published literature. Due to its generality, this algorithm is applicable to MR images of both humans and animals.

Structural brain alterations in schizophrenia following fetal exposure to the inflammatory cytokine interleukin-8

BACKGROUND

Maternal infection during pregnancy has been repeatedly associated with increased risk for schizophrenia. Nevertheless, most viruses do not cross the placenta; therefore, the damaging effects to the fetus appear to be related to maternal antiviral responses to infection (e.g. proinflammatory cytokines). Fetal exposure to the proinflammatory cytokine interleukin-8 (IL-8) has been significantly associated with risk of schizophrenia in offspring. This study sought to determine the association between fetal exposure to IL-8 and structural brain changes among schizophrenia cases and controls.

METHODS

Subjects were 17 cases diagnosed with schizophrenia from the Developmental Insult and Brain Anomaly in Schizophrenia (DIBS) study. Psychiatric diagnoses were determined among offspring with semi-structured interviews and medical records review. IL-8 was determined from assays in archived prenatal sera and volumetric analyses of neuroanatomical regions were obtained from T1-weighted magnetic resonance imaging in adulthood. Eight controls were included for exploratory purposes.

RESULTS

Among cases, fetal exposure to increases in IL-8 was associated with significant increases in ventricular cerebrospinal fluid, significant decreases in left entorhinal cortex volumes and significant decreases in right posterior cingulate volumes. Decreases that approached significance also were found in volumes of the right caudate, the putamen (bilaterally), and the right superior temporal gyrus. No significant associations were observed among controls.

CONCLUSION

Fetal exposure to elevations in maternal IL-8 led to structural neuroanatomic alterations among cases in regions of the brain consistently implicated in schizophrenia research. In utero exposure to elevations in IL-8 may partially account for brain disturbances commonly found in schizophrenia.

Hybrid spline-based multimodal registration using local measures for joint entropy and mutual information

We introduce a new hybrid approach for spline-based elastic registration of multimodal medical images. The approach uses point landmarks as well as intensity information based on local analytic measures for joint entropy and mutual information. The information-theoretic similarity measures are computationally efficient and can be optimized independently for each voxel. We have applied our approach to synthetic images, brain phantom images, as well as clinically relevant multimodal medical images. We also compared our measures with previous measures.

Bloch simulations with intra-voxel spin dephasing

A common problem in simulations of MRI-experiments based on the numerical solution of the Bloch equations is the finite number of isochromats used in the calculations. This usually results in false or spurious signals and is a source of various differences between calculated and experimentally obtained data. In this paper, we are proposing a technique representing each sample voxel by a central and three additional isochromats, slightly shifted in orthogonal directions from center, thus providing a linear approximation of intra-voxel dephasing. This approach allows for further improvement and precision of the calculated NMR signal and virtually avoids the problem related to an finite set of isochromats. Here we provide details of the algorithm together with examples of simulations which prove the efficiency of this approach.

Automatic cerebral and cerebellar hemisphere segmentation in 3D MRI: adaptive disconnection algorithm

This paper describes the automatic Adaptive Disconnection method to segment cerebral and cerebellar hemispheres of human brain in three-dimensional magnetic resonance imaging (MRI). Using the partial differential equations based shape bottlenecks algorithm cooperating with an information potential value clustering process, it detects and cuts, first, the compartmental connections between the cerebrum, the cerebellum and the brainstem in the white matter domain, and then, the interhemispheric connections of the extracted cerebrum and cerebellum volumes. As long as the subject orientation in the scanner is given, the variations in subject location and normal brain morphology in different images are accommodated automatically, thus no stereotaxic image registration is required. The modeling of partial volume effect is used to locate cerebrum, cerebellum and brainstem boundaries, and make the interhemispheric connections detectable. The Adaptive Disconnection method was tested with 10 simulated images from the BrainWeb database and 39 clinical images from the LONI Probabilistic Brain Atlas database. It obtained lower error rates than a traditional shape bottlenecks algorithm based segmentation technique (BrainVisa) and linear and nonlinear registration based brain hemisphere segmentation methods. Segmentation accuracies were evaluated against manual segmentations. The Adaptive Disconnection method was also confirmed not to be sensitive to the noise and intensity non-uniformity in the images. We also applied the Adaptive Disconnection method to clinical images of 22 healthy controls and 18 patients with schizophrenia. A preliminary cerebral volumetric asymmetry analysis based on these images demonstrated that the Adaptive Disconnection method is applicable to study abnormal brain asymmetry in schizophrenia.

3-D brain MRI tissue classification on FPGAs

Many automatic algorithms have been proposed for analyzing magnetic resonance imaging (MRI) data sets. With the increasingly large data sets being used in brain mapping, there has been a significant rise in the need for accelerating these algorithms. Partial volume estimation (PVE), a brain tissue classification algorithm for MRI, was implemented on a field-programmable gate array (FPGA)-based high performance reconfigurable computer using the Mitrion-C high-level language (HLL). This work develops on prior work in which we conducted initial studies on accelerating the prior information estimation algorithm. In this paper, we extend the work to include probability density estimation and present new results and additional analysis. We used several simulated and real human brain MR images to evaluate the accuracy and performance improvement of the proposed algorithm. The FPGA-based probability density estimation and prior information estimation implementation achieved an average speedup over an Itanium 2 CPU of 2.5 x and 9.4 x , respectively. The overall performance improvement of the FPGA-based PVE algorithm was 5.1 x with four FPGAs.

MCAT to XCAT: The Evolution of 4-D Computerized Phantoms for Imaging Research: Computer models that take account of body movements promise to provide evaluation and improvement of medical imaging devices and technology

Recent work in the development of computerized phantoms has focused on the creation of ideal "hybrid" models that seek to combine the realism of a patient-based voxelized phantom with the flexibility of a mathematical or stylized phantom. We have been leading the development of such computerized phantoms for use in medical imaging research. This paper will summarize our developments dating from the original four-dimensional (4-D) Mathematical Cardiac-Torso (MCAT) phantom, a stylized model based on geometric primitives, to the current 4-D extended Cardiac-Torso (XCAT) and Mouse Whole-Body (MOBY) phantoms, hybrid models of the human and laboratory mouse based on state-of-the-art computer graphics techniques. This paper illustrates the evolution of computerized phantoms toward more accurate models of anatomy and physiology. This evolution was catalyzed through the introduction of nonuniform rational b-spline (NURBS) and subdivision (SD) surfaces, tools widely used in computer graphics, as modeling primitives to define a more ideal hybrid phantom. With NURBS and SD surfaces as a basis, we progressed from a simple geometrically based model of the male torso (MCAT) containing only a handful of structures to detailed, whole-body models of the male and female (XCAT) anatomies (at different ages from newborn to adult), each containing more than 9000 structures. The techniques we applied for modeling the human body were similarly used in the creation of the 4-D MOBY phantom, a whole-body model for the mouse designed for small animal imaging research. From our work, we have found the NURBS and SD surface modeling techniques to be an efficient and flexible way to describe the anatomy and physiology for realistic phantoms. Based on imaging data, the surfaces can accurately model the complex organs and structures in the body, providing a level of realism comparable to that of a voxelized phantom. In addition, they are very flexible. Like stylized models, they can easily be manipulated to model anatomical variations and patient motion. With the vast improvement in realism, the phantoms developed in our lab can be combined with accurate models of the imaging process (SPECT, PET, CT, magnetic resonance imaging, and ultrasound) to generate simulated imaging data close to that from actual human or animal subjects. As such, they can provide vital tools to generate predictive imaging data from many different subjects under various scanning parameters from which to quantitatively evaluate and improve imaging devices and techniques. From the MCAT to XCAT, we will demonstrate how NURBS and SD surface modeling have resulted in a major evolutionary advance in the development of computerized phantoms for imaging research.

Grid infrastructures for computational neuroscience: the neuGRID example

Neuroscience is increasingly making use of statistical and mathematical tools to extract information from images of biological tissues. Computational neuroimaging tools require substantial computational resources and the increasing availability of large image datasets will further enhance this need. Many efforts have been directed towards creating brain image repositories including the recent US Alzheimer Disease Neuroimaging Initiative. Multisite-distributed computing infrastructures have been launched with the goal of fostering shared resources and facilitating data analysis in the study of neurodegenerative diseases. Currently, some Grid- and non-Grid-based projects are aiming to establish distributed e-infrastructures, interconnecting compatible imaging datasets and to supply neuroscientists with the most advanced information and communication technologies tools to study markers of Alzheimer’s and other brain diseases, but they have so far failed to make a difference in the larger neuroscience community. NeuGRID is an Europeon comission-funded effort arising from the needs of the Alzheimer’s disease imaging community, which will allow the collection and archiving of large amounts of imaging data coupled with Grid-based algorithms and sufficiently powered computational resources. The major benefit will be the faster discovery of new disease markers that will be valuable for earlier diagnosis and development of innovative drugs. The initial setup of neuGRID will feature three nodes equipped with supercomputer capabilities and resources of more than 300 processor cores, 300 GB of RAM memory and approximately 20 TB of physical space. The scope of this article is highlights the new perspectives and potential for the study of the neurodegenerative disorders using the emerging Grid technology.

Virtual glioblastoma: growth, migration and treatment in a three-dimensional mathematical model

OBJECTIVES

Glioblastomas are aggressive primary brain cancers that are characterized by extensive infiltration into the brain and are highly resistant to treatment. Through mathematical modelling, we model the process of invasion and predict the relative importance of mechanisms contributing to malignant invasion. Clinically, we predict patterns of tumour recurrence following various modes of therapeutic intervention.

MATERIALS AND METHODS

Our mathematical model uses a realistic three-dimensional brain geometry and considers migrating and proliferating cells as separate classes. Several mechanisms for infiltrative migration are considered. Methods are developed for simulating surgical resection, radiotherapy and chemotherapy.

RESULTS

The model provides clinically realistic predictions of tumour growth and recurrence following therapeutic intervention. Specific results include (i) invasiveness is governed largely by the ability of glioblastoma cells to degrade and migrate through the extracellular matrix and the ability of single migrating cells to form colonies; (ii) tumours originating deeper in the brain generally grow more quickly than those of superficial origin; (iii) upon surgery, the margins and geometry of resection significantly determine the extent and pattern of postoperative recurrence; (iv) radiotherapy works synergistically with greater resection margins to reduce recurrence; (v) simulations in both two- and three-dimensional geometries give qualitatively similar results; and (vi) in an actual clinical case comprising several surgical interventions, the model provides good qualitative agreement between the simulated and observed course of the disease.

CONCLUSIONS

The model provides a useful initial framework by which biological mechanisms of invasion and efficacy of potential treatment regimens may be assessed.

Multispectral analysis of multimodal images

INTRODUCTION

An increasing number of multimodal images represent a valuable increase in available image information, but at the same time it complicates the extraction of diagnostic information across the images. Multispectral analysis (MSA) has the potential to simplify this problem substantially as unlimited number of images can be combined, and tissue properties across the images can be extracted automatically.

MATERIALS AND METHODS

We have developed a software solution for MSA containing two algorithms for unsupervised classification, an EM-algorithm finding multinormal class descriptions and the k-means clustering algorithm, and two for supervised classification, a Bayesian classifier using multinormal class descriptions and a kNN-algorithm. The software has an efficient user interface for the creation and manipulation of class descriptions, and it has proper tools for displaying the results.

RESULTS

The software has been tested on different sets of images. One application is to segment cross-sectional images of brain tissue (T1- and T2-weighted MR images) into its main normal tissues and brain tumors. Another interesting set of images are the perfusion maps and diffusion maps, derived images from raw MR images. The software returns segmentations that seem to be sensible.

DISCUSSION

The MSA software appears to be a valuable tool for image analysis with multimodal images at hand. It readily gives a segmentation of image volumes that visually seems to be sensible. However, to really learn how to use MSA, it will be necessary to gain more insight into what tissues the different segments contain, and the upcoming work will therefore be focused on examining the tissues through for example histological sections.

Evaluation of automated brain MR image segmentation and volumetry methods

We compare three widely used brain volumetry methods available in the software packages FSL, SPM5, and FreeSurfer and evaluate their performance using simulated and real MR brain data sets. We analyze the accuracy of gray and white matter volume measurements and their robustness against changes of image quality using the BrainWeb MRI database. These images are based on "gold-standard" reference brain templates. This allows us to assess between- (same data set, different method) and also within-segmenter (same method, variation of image quality) comparability, for both of which we find pronounced variations in segmentation results for gray and white matter volumes. The calculated volumes deviate up to >10% from the reference values for gray and white matter depending on method and image quality. Sensitivity is best for SPM5, volumetric accuracy for gray and white matter was similar in SPM5 and FSL and better than in FreeSurfer. FSL showed the highest stability for white (<5%), FreeSurfer (6.2%) for gray matter for constant image quality BrainWeb data. Between-segmenter comparisons show discrepancies of up to >20% for the simulated data and 24% on average for the real data sets, whereas within-method performance analysis uncovered volume differences of up to >15%. Since the discrepancies between results reach the same order of magnitude as volume changes observed in disease, these effects limit the usability of the segmentation methods for following volume changes in individual patients over time and should be taken into account during the planning and analysis of brain volume studies.

Partial volume effects in dynamic contrast magnetic resonance renal studies

This is the first study of partial volume effect in quantifying renal function on dynamic contrast enhanced magnetic resonance imaging. Dynamic image data were acquired for a cohort of 10 healthy volunteers. Following respiratory motion correction, each voxel location was assigned a mixing vector representing the 'overspilling' contributions of each tissue due to the convolution action of the imaging system's point spread function. This was used to recover the true intensities associated with each constituent tissue. Thus, non-renal contributions from liver, spleen and other surrounding tissues could be eliminated from the observed time-intensity curves derived from a typical renal cortical region of interest. This analysis produced a change in the early slope of the renal curve, which subsequently resulted in an enhanced glomerular filtration rate estimate. This effect was consistently observed in a Rutland-Patlak analysis of the time-intensity data: the volunteer cohort produced a partial volume effect corrected mean enhancement of 36% in relative glomerular filtration rate with a mean improvement of 7% in r(2) fitting of the Rutland-Patlak model compared to the same analysis undertaken without partial volume effect correction. This analysis strongly supports the notion that dynamic contrast enhanced magnetic resonance imaging of kidneys is substantially affected by the partial volume effect, and that this is a significant obfuscating factor in subsequent glomerular filtration rate estimation.

Comparison of tissue segmentation algorithms in neuroimage analysis software tools

Accurate segmentation of different brain tissues is of much importance in magnetic resonance imaging. This paper presents a comparison of the existing segmentation algorithms that are deployed in the neuroimaging community as part of two widely used software packages. The results obtained in this comparison can be used to select the appropriate segmentation algorithm for the neuroimaging application of interest. In addition to the entire brain area, a comparison is carried out for the subcortical region of the brain in terms of its gray matter composition.

Performance measure characterization for evaluating neuroimage segmentation algorithms

Characterizing the performance of segmentation algorithms in brain images has been a persistent challenge due to the complexity of neuroanatomical structures, the quality of imagery and the requirement of accurate segmentation. There has been much interest in using the Jaccard and Dice similarity coefficients associated with Sensitivity and Specificity for evaluating the performance of segmentation algorithms. This paper addresses the essential characteristics of the fundamental performance measure coefficients adopted in evaluation frameworks. While exploring the properties of the Jaccard, Dice and Specificity coefficients, we propose new measure coefficients Conformity and Sensibility for evaluating image segmentation techniques. It is indicated that Conformity is more sensitive and rigorous than Jaccard and Dice in that it has better discrimination capabilities in detecting small variations in segmented images. Comparing to Specificity, Sensibility provides consistent and reliable evaluation scores without the incorporation of image background properties. The merits of the proposed coefficients are illustrated by extracting neuroanatomical structures in a wide variety of brain images using various segmentation techniques.

Modeling interaction for segmentation of neighboring structures

This paper presents a new method for segmenting medical images by modeling interaction between neighboring structures. Compared to previously reported methods, the proposed approach enables simultaneous segmentation of multiple neighboring structures for improved robustness. During the segmentation process, the object contour evolution and shape prior estimates are influenced by the interactions between neighboring shapes consisting of attraction, repulsion, and competition. Instead of estimating the a priori shape of each structure independently, an interactive maximum a posteriori shape estimation method is used for estimating the shape priors by considering shape prior distribution, neighboring shapes, and image features. Energy functionals are then formulated to model the interaction and segmentation. With the proposed method, neighboring structures with similar intensities and/or textures, and blurred boundaries can be extracted simultaneously. Experimental results obtained on both synthetic data and medical images demonstrate that the introduced interaction between neighboring structures improves segmentation performance compared with other existing approaches.

MRI segmentation using dialectical optimization

Biology, Psychology and Social Sciences are intrinsically connected to the very roots of the development of algorithms and methods in Computational Intelligence, as it is easily seen in approaches like genetic algorithms, evolutionary programming and particle swarm optimization. In this work we propose a new optimization method based on dialectics using fuzzy membership functions to model the influence of interactions between integrating poles in the status of each pole. Poles are the basic units composing dialectical systems. In order to validate our proposal we designed a segmentation method based on the optimization of k-means using dialectics for the segmentation of MR images. As a case study we used 181 MR synthetic multispectral images composed by proton density, T(1)- and T(2)-weighted synthetic brain images of 181 slices with 1 mm, resolution of 1 mm(3), for a normal brain and a noiseless MR tomographic system without field inhomogeneities, amounting a total of 543 images, generated by the simulator BrainWeb [2]. Our principal target here is comparing our proposal to k-means, fuzzy c-means, and Kohonen's self-organized maps, concerning the quantization error, we proved that our method can improved results obtained using k-means.

Automatic construction of 3D-ASM intensity models by simulating image acquisition: application to myocardial gated SPECT studies

Active shape models bear a great promise for model-based medical image analysis. Their practical use, though, is undermined due to the need to train such models on large image databases. Automatic building of point distribution models (PDMs) has been successfully addressed and a number of autolandmarking techniques are currently available. However, the need for strategies to automatically build intensity models around each landmark has been largely overlooked in the literature. This work demonstrates the potential of creating intensity models automatically by simulating image generation. We show that it is possible to reuse a 3D PDM built from computed tomography (CT) to segment gated single photon emission computed tomography (gSPECT) studies. Training is performed on a realistic virtual population where image acquisition and formation have been modeled using the SIMIND Monte Carlo simulator and ASPIRE image reconstruction software, respectively. The dataset comprised 208 digital phantoms (4D-NCAT) and 20 clinical studies. The evaluation is accomplished by comparing point-to-surface and volume errors against a proper gold standard. Results show that gSPECT studies can be successfully segmented by models trained under this scheme with subvoxel accuracy. The accuracy in estimated LV function parameters, such as end diastolic volume, end systolic volume, and ejection fraction, ranged from 90.0% to 94.5% for the virtual population and from 87.0% to 89.5% for the clinical population.

Regularized field map estimation in MRI

In fast magnetic resonance (MR) imaging with long readout times, such as echo-planar imaging (EPI) and spiral scans, it is important to correct for the effects of field inhomogeneity to reduce image distortion and blurring. Such corrections require an accurate field map, a map of the off-resonance frequency at each voxel. Standard field map estimation methods yield noisy field maps, particularly in image regions with low spin density. This paper describes regularized methods for field map estimation from two or more MR scans having different echo times. These methods exploit the fact that field maps are generally smooth functions. The methods use algorithms that decrease monotonically a regularized least-squares cost function, even though the problem is highly nonlinear. Results show that the proposed regularized methods significantly improve the quality of field map estimates relative to conventional unregularized methods.