Longitudinal and cross-sectional analysis of atrophy in Alzheimer's disease: cross-validation of BSI, SIENA and SIENAX

Brain volume loss (atrophy) is widely used as a marker of disease progression. Atrophy has been measured with a variety of methods, some estimating atrophy rate from two temporally separated scans, and others estimating atrophy state from a single scan. Three popular tools for measuring brain atrophy are BSI and SIENA (rate) and SIENAX (state). Previous papers have shown BSI and SIENA to have similar accuracy, but no work has carefully compared both methods using the same data set. Here we compare these methods, using data from patients with Alzheimer's disease and age-matched controls. We also compare the SIENA longitudinal measure with atrophy state estimated by SIENAX using just the earliest scan taken from each subject. We show strong correspondence and similar sensitivity to atrophy between all 3 measures.     

Accurate,robust, and automated longitudinal and cross-sectional brain change analysis

Quantitative measurement of brain size, shape, and temporal change (for example, in order to estimate atrophy) is increasingly important in biomedical image analysis applications. New methods of structural analysis attempt to improve robustness, accuracy, and extent of automation. A fully automated method of longitudinal (temporal change) analysis, SIENA, was presented previously. In this paper, improvements to this method are described, and also an extension of SIENA to a new method for cross-sectional (single time point) analysis. The methods are fully automated, robust, and accurate: 0.15% brain volume change error (longitudinal): 0.5-1% brain volume accuracy for single-time point (cross-sectional). A particular advantage is the relative insensitivity to differences in scanning parameters. The methods provide easy manual review of their output by the automatic production of summary images which show the results of the brain extraction, registration, tissue segmentation, and final atrophy estimation.     

A comparison of MR based segmentation methods for measuring brain atrophy progression

Automated brain segmentation methods with a good precision and accuracy are required to detect subtle changes in brain volumes over time in clinical applications. However, the ability of established methods such as SIENA, US and kNN to estimate brain volume change have not been compared on the same data, nor been evaluated with ground-truth manual segmentations. We compared measurements of brain volume change between SIENA, US and kNN in terms of precision (repeatability) and accuracy (ground-truth) using one baseline and two repeated follow-up 1.5 T MRI scans after 4 years of 10 subjects. The coefficient of repeatability (brain volume/volume change) was larger for US (29.6 cc/2.84%) than for kNN (4.9 cc/0.31%) and SIENA (-/0.92%). In terms of absolute brain volume measurements US and kNN showed good correlation with the manual segmentations and with each other (all Spearman's correlation coefficients ρ≥0.96; all p<0.001). Concerning brain volume changes, SIENA showed a good (ρ=0.82; p=0.004), kNN a moderate (ρ=0.60; p=0.067) and US a weak (ρ=0.50; p=0.138) correlation with the manual segmentations. For measurements of volume change, SIENA-US (mean correlation coefficient and p-value: ρ=0.28; p=0.442) and US-kNN (ρ=0.17; p=0.641) showed a weak correlation, but correlation was fairly good for kNN-SIENA (ρ=0.65; p=0.048). In conclusion, US and kNN showed a good precision, accuracy and comparability for brain volume measurements. For measurements of volume change, SIENA showed the best performance. kNN is a good alternative if volume change measurements of other brain structures are required.     

Long-term global and regional brain volume changes following severe traumatic brain injury: a longitudinal study with clinical correlates

Traumatic brain injury (TBI) results in neurodegenerative changes that progress for months, perhaps even years post-injury. However, there is little information on the spatial distribution and the clinical significance of this late atrophy. In 24 patients who had sustained severe TBI we acquired 3D T1-weighted MRIs about 8 weeks and 12 months post-injury. For comparison, 14 healthy controls with similar distribution of age, gender and education were scanned with a similar time interval. For each subject, longitudinal atrophy was estimated using SIENA, and atrophy occurring before the first scan time point using SIENAX. Regional distribution of atrophy was evaluated using tensor-based morphometry (TBM). At the first scan time point, brain parenchymal volume was reduced by mean 8.4% in patients as compared to controls. During the scan interval, patients exhibited continued atrophy with percent brain volume change (%BVC) ranging between -0.6% and -9.4% (mean -4.0%). %BVC correlated significantly with injury severity, functional status at both scans, and with 1-year outcome. Moreover, %BVC improved prediction of long-term functional status over and above what could be predicted using functional status at approximately 8 weeks. In patients as compared to controls, TBM (permutation test, FDR 0.05) revealed a large coherent cluster of significant atrophy in the brain stem and cerebellar peduncles extending bilaterally through the thalamus, internal and external capsules, putamen, inferior and superior longitudinal fasciculus, corpus callosum and corona radiata. This indicates that the long-term atrophy is attributable to consequences of traumatic axonal injury. Despite progressive atrophy, remarkable clinical improvement occurred in most patients.     

Brain MAPS: an automated, accurate and robust brain extraction technique using a template library

Whole brain extraction is an important pre-processing step in neuroimage analysis. Manual or semi-automated brain delineations are labour-intensive and thus not desirable in large studies, meaning that automated techniques are preferable. The accuracy and robustness of automated methods are crucial because human expertise may be required to correct any suboptimal results, which can be very time consuming. We compared the accuracy of four automated brain extraction methods: Brain Extraction Tool (BET), Brain Surface Extractor (BSE), Hybrid Watershed Algorithm (HWA) and a Multi-Atlas Propagation and Segmentation (MAPS) technique we have previously developed for hippocampal segmentation. The four methods were applied to extract whole brains from 682 1.5T and 157 3T T(1)-weighted MR baseline images from the Alzheimer's Disease Neuroimaging Initiative database. Semi-automated brain segmentations with manual editing and checking were used as the gold-standard to compare with the results. The median Jaccard index of MAPS was higher than HWA, BET and BSE in 1.5T and 3T scans (p<0.05, all tests), and the 1st to 99th centile range of the Jaccard index of MAPS was smaller than HWA, BET and BSE in 1.5T and 3T scans ( p<0.05, all tests). HWA and MAPS were found to be best at including all brain tissues (median false negative rate ≤0.010% for 1.5T scans and ≤0.019% for 3T scans, both methods). The median Jaccard index of MAPS were similar in both 1.5T and 3T scans, whereas those of BET, BSE and HWA were higher in 1.5T scans than 3T scans (p<0.05, all tests). We found that the diagnostic group had a small effect on the median Jaccard index of all four methods. In conclusion, MAPS had relatively high accuracy and low variability compared to HWA, BET and BSE in MR scans with and without atrophy.     

Measurement of brain atrophy in subcortical vascular disease: a comparison of different approaches and the impact of ischaemic lesions

Measurement of brain atrophy has been proposed as a surrogate marker in MS and degenerative dementias. Although cerebral small vessel disease predominantly affects white and subcortical grey matter, recent data suggest that whole brain atrophy is also a good indicator of clinical and cognitive status in this disease. Automated methods to measure atrophy are available that are accurate and reproducible in disease-free brains. However, optimal methods in small vessel disease have not been established and the impact of ischaemic lesions on different techniques has not been explored systematically. In this study, three contrasting techniques -- Statistical Parametric Mapping 5 (SPM5), SIENAX and BrainVisa -- were applied to measure cross-sectional atrophy (brain parenchymal fraction or BPF) in a large (n=143) two-centre cohort of patients with CADASIL, a genetic model of small vessel disease. All three techniques showed similar sensitivity to trends in BPF associated with age and lesion load. No single technique was particularly vulnerable to error as a result of lesions. Provided major errors in registration were excluded by visual inspection, manual correction of segmentations had a negligible impact with mean errors of 0.41% for SIENAX and 0.46% for BrainVisa. BPF correlated strongly with global cognitive function and physical disability, independent of the technique used. Correlation coefficients with the Minimental State Examination score were: BrainVisa 0.58, SIENAX 0.58, SPM5 0.60 (for all, p<0.001). These results suggest that all three methods can be applied reliably in patients with ischaemic lesions. Choice of analysis approach for this kind of clinical question will be determined by factors other than their robustness and precision, such as a desire to explore subtle localised changes using extensions of these processing tools.     

Accuracy assessment of global and local atrophy measurement techniques with realistic simulated longitudinal Alzheimer's disease images

The evaluation of atrophy quantification methods based on magnetic resonance imaging have been usually hindered by the lack of realistic gold standard data against which to judge these methods or to help refine them. Recently [Camara, O., Schweiger, M., Scahill, R., Crum, W., Sneller, B., Schnabel, J., Ridgway, G., Cash, D., Hill, D., Fox, N., 2006. Phenomenological model of diffuse global and regional atrophy using finite-element methods. IEEE Trans. Med.l Imaging 25, 1417-1430], we presented a technique in which atrophy is realistically simulated in different tissue compartments or neuroanatomical structures with a phenomenological model. In this study, we have generated a cohort of realistic simulated Alzheimer's disease (AD) images with known amounts of atrophy, mimicking a set of 19 real controls and 27 probable AD subjects, with an improved version of our atrophy simulation methodology. This database was then used to assess the accuracy of several well-known computational anatomy methods which provide global (BSI and SIENA) or local (Jacobian integration) estimates of longitudinal atrophy in brain structures using MR images. SIENA and BSI results correlated very well with gold standard data (Pearson coefficient of 0.962 and 0.969 respectively), achieving small mean absolute differences with respect to the gold standard (percentage change from baseline volume): BSI of 0.23%+/-0.26%; SIENA of 0.22%+/-0.28%. Jacobian integration was guided by both fluid and FFD-based registration techniques and resulting deformation fields and associated Jacobians were compared, region by region, with gold standard ones. The FFD-based technique outperformed the fluid one in all evaluated structures (mean absolute differences from the gold standard in percentage change from baseline volume): whole brain, FFD=0.31%, fluid=0.58%; lateral ventricles, FFD=0.79%; fluid=1.45%; left hippocampus, FFD=0.82%; fluid=1.42%; right hippocampus, FFD=0.95%; fluid=1.62%. The largest errors for both local techniques occurred in the sulcal CSF (FFD=2.27%; fluid=3.55%) regions. For large structures such as the whole brain, these mean absolute differences, relative to the applied atrophy, represented similar percentages for the BSI, SIENA and FFD techniques (controls/patients): BSI, 51.99%/16.36%; SIENA, 62.34%/21.59%; FFD, 41.02%/24.95%. For small structures such as the hippocampi, these percentages were larger, especially for controls where errors were approximately equal to the small applied changes (controls/patients): FFD, 92.82%/43.61%. However, these apparently large relative errors have not prevented the global or hippocampal measures from finding significant group separation in our study. The evaluation framework presented here will help in quantifying whether the accuracy of future methodological developments is sufficient for analysing change in smaller or less atrophied local brain regions. Results obtained in our experiments with realistic simulated data confirm previously published estimates of accuracy for both evaluated global techniques. Regarding Jacobian Integration methods, the FFD-based one demonstrated promising results and potential for being used in clinical studies alongside (or in place of) the more common global methods. The generated gold standard data has also allowed us to identify some stages and sets of parameters in the evaluated techniques--the brain extraction step in the global techniques and the number of multi-resolution levels and the stopping criteria in the registration-based methods--that are critical for their accuracy.     

Multi-spectral brain tissue segmentation using automatically trained k-Nearest-Neighbor classification

Conventional k-Nearest-Neighbor (kNN) classification, which has been successfully applied to classify brain tissue in MR data, requires training on manually labeled subjects. This manual labeling is a laborious and time-consuming procedure. In this work, a new fully automated brain tissue classification procedure is presented, in which kNN training is automated. This is achieved by non-rigidly registering the MR data with a tissue probability atlas to automatically select training samples, followed by a post-processing step to keep the most reliable samples. The accuracy of the new method was compared to rigid registration-based training and to conventional kNN-based segmentation using training on manually labeled subjects for segmenting gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) in 12 data sets. Furthermore, for all classification methods, the performance was assessed when varying the free parameters. Finally, the robustness of the fully automated procedure was evaluated on 59 subjects. The automated training method using non-rigid registration with a tissue probability atlas was significantly more accurate than rigid registration. For both automated training using non-rigid registration and for the manually trained kNN classifier, the difference with the manual labeling by observers was not significantly larger than inter-observer variability for all tissue types. From the robustness study, it was clear that, given an appropriate brain atlas and optimal parameters, our new fully automated, non-rigid registration-based method gives accurate and robust segmentation results. A similarity index was used for comparison with manually trained kNN. The similarity indices were 0.93, 0.92 and 0.92, for CSF, GM and WM, respectively. It can be concluded that our fully automated method using non-rigid registration may replace manual segmentation, and thus that automated brain tissue segmentation without laborious manual training is feasible.     

Automatic segmentation of left and right cerebral hemispheres from MRI brain volumes using the graph cuts algorithm

An automated algorithm has been developed to segment stripped (non-brain tissue excluded) T1-weighted MRI brain volumes into left and right cerebral hemispheres and cerebellum+brainstem. The algorithm, which uses the Graph Cuts technique, performs a fully automated segmentation in approximately 30 s following pre-processing. It is robust and accurate and has been tested on datasets from two scanners using different field strengths and pulse sequences. We describe the Graph Cuts algorithm and compare the results of Graph Cuts segmentations against "gold standard" manual segmentations and segmentations produced by three popular software packages used by neuroimagers: BrainVisa, CLASP, and SurfRelax.     

Quantitative comparison of four brain extraction algorithms

In a companion paper (Rehm et al., 2004), we introduced Minneapolis Consensus Strip (McStrip), a hybrid algorithm for brain/non-brain segmentation. In this paper, we compare the performance of McStrip and three brain extraction algorithms (BEAs) in widespread use within the neuroimaging community--Statistical Parametric Mapping v.2 (SPM2), Brain Extraction Tool (BET), and Brain Surface Extractor (BSE)--to the "gold standard" of manually stripped T1-weighted MRI brain volumes. Our comparison was based on quantitative boundary and volume metrics, reproducibility across repeat scans of a single subject, and assessments of performance consistency across datasets acquired on different scanners at different institutions. McStrip, a hybrid method incorporating warping to a template, intensity thresholding, and edge detection, consistently outperformed SPM2, BET, and BSE, all of which rely on a single algorithmic strategy.     

Anatomically constrained region deformation for the automated segmentation of the hippocampus and the amygdala: Method and validation on controls and patients with Alzheimer's disease

We describe a new algorithm for the automated segmentation of the hippocampus (Hc) and the amygdala (Am) in clinical Magnetic Resonance Imaging (MRI) scans. Based on homotopically deforming regions, our iterative approach allows the simultaneous extraction of both structures, by means of dual competitive growth. One of the most original features of our approach is the deformation constraint based on prior knowledge of anatomical features that are automatically retrieved from the MRI data. The only manual intervention consists of the definition of a bounding box and positioning of two seeds; total execution time for the two structures is between 5 and 7 min including initialisation. The method is evaluated on 16 young healthy subjects and 8 patients with Alzheimer's disease (AD) for whom the atrophy ranged from limited to severe. Three aspects of the performances are characterised for validating the method: accuracy (automated vs. manual segmentations), reproducibility of the automated segmentation and reproducibility of the manual segmentation. For 16 young healthy subjects, accuracy is characterised by mean relative volume error/overlap/maximal boundary distance of 7%/84%/4.5 mm for Hc and 12%/81%/3.9 mm for Am; for 8 Alzheimer's disease patients, it is 9%/84%/6.5 mm for Hc and 15%/76%/4.5 mm for Am. We conclude that the performance of this new approach in data from healthy and diseased subjects in terms of segmentation quality, reproducibility and time efficiency compares favourably with that of previously published manual and automated segmentation methods. The proposed approach provides a new framework for further developments in quantitative analyses of the pathological hippocampus and amygdala in MRI scans.     

Skull stripping based on region growing for magnetic resonance brain images

In this paper, we propose a new skull stripping method for T1-weighted magnetic resonance (MR) brain images. Skull stripping has played an important role in neuroimage research because it is a basic preliminary step in many clinical applications. The process of skull stripping can be challenging due to the complexity of the human brain, variable parameters of MR scanners, individual characteristics, etc. In this paper, we aim to develop a computationally efficient and robust method. In the proposed algorithm, after eliminating the background voxels with histogram analysis, two seed regions of the brain and non-brain regions were automatically identified using a mask produced by morphological operations. Then we expanded these seed regions with a 2D region growing algorithm based on general brain anatomy information. The proposed algorithm was validated using 56 volumes of human brain data and simulated phantom data with manually segmented masks. It was compared with two popular automated skull stripping methods: the brain surface extractor (BSE) and the brain extraction tool (BET). The experimental results showed that the proposed algorithm produced accurate and stable results against data sets acquired from various MR scanners and effectively addressed difficult problems such as low contrast and large anatomical connections between the brain and surrounding tissues. The proposed method was also robust against noise, RF, and intensity inhomogeneities.     

探索高原登山的脑结构改变

目的 探讨一次短暂高原登山引起的脑结构改变.方法 对15名厦门大学学生登山队员[男9名,女6名,19~23岁,平均(21.0±1.1)岁]分别于攀登珠穆朗玛峰前、后进行常规T2W及高分辨率全脑3DT1W结构成像;应用SIENA软件分别对登山前、后高分辨率3DT1W结构像进行全脑灰质、白质分割,计算体积萎缩百分率,并进行统计分析;对全脑进行基于体素的纵向脑萎缩评价,获取显著萎缩脑区.结果 视觉观察,登山前、后所有登山队员常规T2WI均未发现异常,但脑灰质及白质体积均有明显减少,脑灰质萎缩百分率为(2.70±1.43)%,白质萎缩百分率为(1.43±1.36)%,差异有统计学意义(P<0.01);基于体素的全脑统计分析发现,萎缩脑区包括左侧额叶、胼胝体压部、双侧颞极、双侧枕叶距状沟周围及双侧小脑半球,以优势半球受损明显.结论 高原登山运动可引起脑白质和灰质萎缩,且灰质萎缩更明显.     

Comparisons between Alzheimer disease, frontotemporal lobar degeneration, and normal aging with brain mapping

Alzheimer disease (AD) and frontotemporal lobar degeneration (FTLD) are both common degenerative dementias in the under 65 age group. Although clinical criteria have been defined for both diseases, there is considerable overlap in clinical features, and hence, diagnosis still can be very difficult particularly in the early stages of the disease. As a result, there has been increasing interest in using magnetic resonance imaging to better characterize these diseases and to aid in diagnosis. Voxel-based morphometry (VBM) is an automated technique that assesses patterns of regional gray matter atrophy on magnetic resonance imaging between 2 groups of subjects. It is unbiased in that it looks throughout the whole brain and does not require any a priori assumptions concerning which structures to assess, giving it a significant advantage over traditional region of interest-based methods. Voxel-based morphometry has been widely used to assess patterns of regional atrophy in subjects with AD and FTLD. These studies have demonstrated specific patterns of regional loss in both diseases, compared the 2 diseases to look for differences that could be diagnostically useful, and have correlated regions of gray matter loss to cognitive and behavioral deficits in these subjects. This article will review the findings of these studies and discuss the role of VBM in these neurodegenerative diseases.     

Longitudinal characterization of brain atrophy of a Huntington's disease mouse model by automated morphological analyses of magnetic resonance images

Mouse models of human diseases play crucial roles in understanding disease mechanisms and developing therapeutic measures. Huntington's disease (HD) is characterized by striatal atrophy that begins long before the onset of motor symptoms. In symptomatic HD, striatal volumes decline predictably with disease course. Thus, imaging based volumetric measures have been proposed as outcomes for presymptomatic as well as symptomatic clinical trials of HD. Magnetic resonance imaging of the mouse brain structures is becoming widely available and has been proposed as one of the biomarkers of disease progression and drug efficacy testing. However, three-dimensional and quantitative morphological analyses of the brains are not straightforward. In this paper, we describe a tool for automated segmentation and voxel-based morphological analyses of the mouse brains. This tool was applied to a well-established mouse model of Huntington's disease, the R6/2 transgenic mouse strain. Comparison between the automated and manual segmentation results showed excellent agreement in most brain regions. The automated method was able to sensitively detect atrophy as early as 4 weeks of age and accurately follow disease progression. Comparison between ex vivo and in vivo MRI suggests that the ex vivo end-point measurement of brain morphology is also a valid approach except for the morphology of the ventricles. This is the first report of longitudinal characterization of brain atrophy in a mouse model of Huntington's disease by using automatic morphological analysis.     

A new improved version of the realistic digital brain phantom

Image analysis methods must be tested and evaluated within a controlled environment. Simulations can be an extremely helpful tool for validation because ground truth is known. We created the digital brain phantom that is at the heart of our publicly available database of realistic simulated magnetic resonance image (MRI) volumes known as BrainWeb. Even though the digital phantom had l mm(3) isotropic voxel size and a small number of tissue classes, the BrainWeb database has been used in more than one hundred peer-reviewed publications validating different image processing methods. In this paper, we describe the next step in the natural evolution of BrainWeb: the creation of digital brain phantom II that includes three major improvements over the original phantom. First, the realism of the phantom, and the resulting simulations, was improved by modeling more tissue classes to include blood vessels, bone marrow and dura mater classes. In addition. a more realistic skull class was created. The latter is particularly useful for SPECT, PET and CT simulations for which bone attenuation has an important effect. Second, the phantom was improved by an eight-fold reduction in voxel volume to 0.125 mm(3). Third, the method used to create the new phantom was modified not only to take into account the segmentation of these new structures, but also to take advantage of many more automated procedures now available. The overall process has reduced subjectivity and manual intervention when compared to the original phantom, and the process may be easily applied to create phantoms from other subjects. MRI simulations are shown to illustrate the difference between the previous and the new improved digital brain phantom II. Example PET and SPECT simulations are also presented.     

Registration and machine learning-based automated segmentation of subcortical and cerebellar brain structures

The large amount of imaging data collected in several ongoing multi-center studies requires automated methods to delineate brain structures of interest. We have previously reported on using artificial neural networks (ANN) to define subcortical brain structures. Here we present several automated segmentation methods using multidimensional registration. A direct comparison between template, probability, artificial neural network (ANN) and support vector machine (SVM)-based automated segmentation methods is presented. Three metrics for each segmentation method are reported in the delineation of subcortical and cerebellar brain regions. Results show that the machine learning methods outperform the template and probability-based methods. Utilization of these automated segmentation methods may be as reliable as manual raters and require no rater intervention.     

An automated algorithm for the computation of brain volume change from sequential MRIs using an iterative principal component analysis and its evaluation for the assessment of whole-brain atrophy rates in patients with probable Alzheimer's disease

This article introduces an automated method for the computation of changes in brain volume from sequential magnetic resonance images (MRIs) using an iterative principal component analysis (IPCA) and demonstrates its ability to characterize whole-brain atrophy rates in patients with Alzheimer's disease (AD). The IPCA considers the voxel intensity pairs from coregistered MRIs and identifies those pairs a sufficiently large distance away from the iteratively determined PCA major axis. Analyses of simulated and real MRI data support the underlying assumption of a linear relationship in paired voxel intensities, identify an outlier distance threshold that optimizes the trade-off between sensitivity and specificity in the detection of small volume changes while accounting for global intensity changes, and demonstrate an ability to detect changes as small as 0.04% of brain volume without confounding effects of between-scan shifts in voxel intensity. In eight patients with probable AD and eight age-matched normal control subjects, the IPCA was comparable to the established but partly manual digital subtraction (DS) method in characterizing annual rates of whole-brain atrophy: resulting rates were correlated (Spearman rank correlation = 0.94, P < 0.0005) and comparable in distinguishing probable AD from normal aging (IPCA-detected atrophy rates: 2.17 +/- 0.52% per year in the patients vs. 0.41 +/- 0.22% per year in the controls [Wilcoxon-Mann-Whitney test P = 7.8 x 10(-4)]; DS-detected atrophy rates: 3.51 +/- 1.31% per year in the patients vs. 0.48 +/- 0.29% per year in the controls [P = 7.8 x 10(-4)]). The IPCA could be used in tracking the progression of AD, evaluating the disease-modifying effects of putative treatments, and investigating the course of other normal and pathological changes in brain morphology.     

The relationship between whole brain volume and disability in multiple sclerosis: a comparison of normalized gray vs. white matter with misclassification correction

We used SPM99 to obtain normalized whole brain volumes of gray matter, white matter, and total parenchyma in patients with multiple sclerosis (MS) (n = 41) and age-/sex-matched normal controls (n = 18). As SPM99's automated gray/white matter volumes were significantly influenced by tissue compartment misclassification due to the effect of MS-related brain lesions, we corrected these automated volumes for misclassification before performing our primary analyses. For MS patients (disease duration = 9.5 +/- 6.3 years; EDSS score = 3.2 +/- 1.8; 25FTW = 6.6 +/- 3.1 s), we also measured lesion load (total T1 hypointense [T1LV] and FLAIR hyperintense lesion volume [FLLV]), central brain atrophy (third ventricular width [TVW] and bicaudate ratio [BCR]), and clinical status (Expanded Disability Status Scale [EDSS] and 25-ft timed walk [25FTW]). Patients with MS had lower gray matter (707 +/- 33 cm(3) [-3.9%], P = 0.003) and total parenchymal volume (1088 +/- 48 cm(3) [-3.8%], P = 0.003), but only a trend for lower white matter volume (381 +/- 25 cm(3) [-3.7%], P = 0.052) relative to normal controls (gray matter: 736 +/- 33 cm(3); total parenchyma: 1132 +/- 49 cm(3); white matter: 396 +/- 26 cm(3)). Gray matter atrophy was related to clinical status (EDSS, 25FTW, and disease duration), lesion load (T1LV and FLLV), and central brain atrophy (TVW and BCR), whereas white matter atrophy was related to only central brain atrophy. These findings suggest that gray matter loss is related to other aspects of brain pathology and has more clinical relevance than white matter atrophy in MS.