Patrick S. Carmack

Hippocampal Dysfunction in Gulf War Veterans: Investigation with ASL Perfusion MR Imaging and Physostigmine Challenge

PURPOSE To determine, with arterial spin labeling (ASL) perfusion magnetic resonance (MR) imaging and physostigmine challenge, if abnormal hippocampal blood flow in ill Gulf War veterans persists 11 years after initial testing with single photon emission computed tomography and nearly 20 years after the 1991 Gulf War. MATERIALS AND METHODS The local institutional review board approved this HIPAA-compliant study. Veterans were screened for contraindications and gave written informed consent before the study. In a semiblinded retrospective protocol, veterans in three Gulf War illness groups-syndrome 1 (impaired cognition), syndrome 2 (confusion-ataxia), and syndrome 3 (central neuropathic pain)-and a control group received intravenous infusions of saline in an initial session and physostigmine in a second session, 48 hours later. Each infusion was followed by measurement of hippocampal regional cerebral blood flow (rCBF) with pulsed ASL. A mixed-effects linear model adjusted for age was used to test for differences in rCBF after the cholinergic challenge across the four groups. RESULTS Physostigmine significantly decreased hippocampal rCBF in control subjects (P < .0005) and veterans with syndrome 1 (P < .05) but significantly increased hippocampal rCBF in veterans with syndrome 2 (P < .005) and veterans with syndrome 3 (P < .002). The abnormal increase in rCBF was found to have progressed to the left hippocampus of the veterans with syndrome 2 and to both hippocampi of the veterans with syndrome 3. CONCLUSION Chronic hippocampal perfusion dysfunction persists or worsens in veterans with certain Gulf War syndromes. ASL MR imaging examination of hippocampal rCBF in a cholinergic challenge experiment may be useful as a diagnostic test for this condition.

Abnormal brain response to cholinergic challenge in chronic encephalopathy from the 1991 Gulf War

Several case definitions of chronic illness in veterans of the 1991 Persian Gulf War have been linked epidemiologically with environmental exposure to cholinesterase-inhibiting chemicals, which cause chronic changes in cholinergic receptors in animal models. Twenty-one chronically ill Gulf War veterans (5 with symptom complex 1, 11 with complex 2, and 5 with complex 3) and 17 age-, sex- and education-matched controls, underwent an 99mTc-HMPAO-SPECT brain scan following infusion of saline and >48 h later a second scan following infusion of physostigmine in saline. From each SPECT image mean normalized regional cerebral blood flow (nrCBF) from 39 small blocks of correlated voxels were extracted with geostatistical spatial modeling from eight deep gray matter structures in each hemisphere. Baseline nrCBF in symptom complex 2 was lower than controls throughout deep structures. The change in nrCBF after physostigmine (challenge minus baseline) was negative in complexes 1 and 3 and controls but positive in complex 2 in some structures. Since effects were opposite in different groups, no finding typified the entire patient sample. A hold-out discriminant model of nrCBF from 17 deep brain blocks predicted membership in the clinical groups with sensitivity of 0.95 and specificity of 0.82. Gulf War-associated chronic encephalopathy in a subset of veterans may be due to neuronal dysfunction, including abnormal cholinergic response, in deep brain structures.

Improved agreement between Talairach and MNI coordinate spaces in deep brain regions.

Disagreement between the Talairach atlas and the stereotaxic space commonly used in software like SPM is a widely recognized problem. Others have proposed affine transformations to improve agreement in surface areas such as Brodmanns areas. This article proposes a similar transformation with the goal of improving agreement specifically in the deep brain region. The task is accomplished by finding an affine transformation that minimizes the mean distance between the surface coordinates of the lateral ventricles in the Talairach atlas and the MNI templates. The result is a transformation that improves deep brain agreement over both the untransformed Talairach coordinates and the surface-oriented transformation. While the transformation improves deep brain agreement, surface agreement is generally made worse. For areas near the lateral ventricle, the transformation presented herein is valuable for applications such as region of interest (ROI) modeling.

Using a white matter reference to remove the dependency of global signal on experimental conditions in SPECT analyses

Proportional scaling models are often used in functional imaging studies to remove confounding of local signals by global effects. It is generally assumed that global effects are uncorrelated with experimental conditions. However, when the global effect is estimated by the global signal, defined as the intracerebral average, incorrect inference may result from the dependency of the global signal on preexisting conditions or experimental manipulations. In this paper, we propose a simple alternative method of estimating the global effect to be used in a proportional scaling model. Specifically, by defining the global signal with reference strictly to a white matter region within the centrum semiovale, the dependency is removed in experiments where white matter is unaffected by the disease effect or experimental treatments. The increase in the ability to detect changes in regional blood flow is demonstrated in a SPECT study of healthy and ill Gulf War veterans in whom it is suspected that brain abnormalities influence the traditional calculation of the global signal. Controlling for the global effect, ill veterans have significantly lower intracerebral averages than healthy controls (P = 0.0038), evidence that choice of global signal has an impact on inference. Scaling by the modified global signal proposed here results in an increase in sensitivity leading to the identification of several regions in the insula and frontal cortex where ill veterans have significantly lower SPECT emissions. Scaling by the traditional global signal results in the loss of sensitivity to detect these regional differences. Advantages of this alternative method are its computational simplicity and its ability to be easily integrated into existing analysis frameworks such as SPM.

Accounting for Spatial Dependence in the Analysis of SPECT Brain Imaging Data

The size and complexity of brain imaging databases confront statistical analysts with a variety of issues when assessing brain activation differences between groups of subjects. Detecting small group differences in activation is compounded by the need to analyze hundreds of thousands of spatially correlated measurements per image. These analyses are especially problematic when, as is typical, the number of subjects in each group is small. In this article a comprehensive analysis of single-photon emission computed tomography (SPECT) brain images demonstrates that spatial modeling can increase the sensitivity of group comparisons. The key statistical approach for increasing the sensitivity of group comparisons is the spatial modeling of intervoxel correlations. Correlations among normalized SPECT counts in 2 × 2 × 2 mm3 voxels are shown to be very large in neighboring voxels and to decrease in magnitude until they become negligible among those approximately 5–7 voxels (10–14 mm) apart. Exploiting this correlation structure, blocks of contiguous voxels are defined within each of several structures within the deep brain so that the geometric centers of the blocks are no closer than the range at which voxel counts can be considered uncorrelated. Using kriging methods, block averages and their prediction variances are calculated. For each structure of interest, the block averages within the structure are weighted by their prediction variances, producing a structure average for each subject. The subject averages and their prediction variances are used in a linear model to compare group effects. This analysis is shown to be more sensitive to group mean differences than the voxel-by-voxel analysis commonly used by medical researchers. The procedures are applied to comparisons of SPECT brain imaging data from four groups of subjects, three of which have variants of the 1991 Gulf War syndrome and one of which is a control group. Commonly used voxel-by-voxel group comparisons do not identify any brain structures that are significantly different for the syndrome and control groups in the analysis of cholinergic response to a physostigmine drug challenge. Spatial modeling and analyses of these data do identify regions of the deep brain that exhibit statistically significant group differences. These results are consistent with medical evidence that these structures might have been affected by Gulf War chemical exposures.

Key properties of D-optimal designs for event-related functional MRI experiments with application to nonlinear models

To properly formulate functional magnetic resonance imaging (fMRI) experiments with complex mental activity, it is advantageous to permit great flexibility in the statistical components of the design of these studies. The length of an experiment, the placement of various stimuli and the modeling approach used all affect the ability to detect mental activity. Major advances in understanding the implications of various designs of fMRI experiments have taken place over the last decade. Nevertheless, new and increasingly difficult issues relating to the modeling of hemodynamic responses and the detection of activated brain regions continue to arise because of the increasing complexity of the experiments. In this article, the D-optimality criterion is used in conjunction with a genetic algorithm to create probability-based design generators for the selection of designs in event-related fMRI experiments where the hemodynamic response function is modeled with a function that is nonlinear in the parameters. The designs produced by these generators are shown to perform well compared with locally D-optimal designs and provide insight into optimal design characteristics that investigators can utilize in the selection of interstimulus intervals. Designs with these characteristics are shown to be applicable to fMRI studies involving one or two stimulus types. The designs are also shown to be robust with respect to misspecification of an AR(1) error autocorrelation and compare favorably with a maximin procedure.

Generalised correlated cross-validation

Since its introduction by [Stone, M. (1974), ‘Cross-validatory Choice and the Assessment of Statistical Predictions (with discussion)’, Journal of the Royal Statistical Society, B36, 111–133] and [Geisser, S. (1975), ‘The Predictive Sample Reuse Method with Applications’, Journal of the American Statistical Association, 70, 320–328], cross-validation has been studied and improved by several authors including [Burman, P., Chow, E., and Nolan, D. (1994), ‘A Cross-validatory Method for Dependent Data’, Biometrika, 81(2), 351–358], [Hart, J. and Yi, S. (1998), ‘One-sided Cross-validation’, Journal of the American Statistical Association, 93(442), 620–630], [Racine, J. (2000), ‘Consistent Cross-validatory Model-selection for Dependent Data: hv-block Cross-validation’, Journal of Econometrics, 99, 39–61], [Hart, J. and Lee, C. (2005), ‘Robustness of One-sided Cross-validation to Autocorrelation’, Journal of Multivariate Analysis, 92(1), 77–96], and [Carmack, P., Spence, J., Schucany, W., Gunst, R., Lin, Q., and Haley, R. (2009), ‘Far Casting Cross Validation’, Journal of Computational and Graphical Statistics, 18(4), 879–893]. Perhaps the most widely used and best known is generalised cross-validation (GCV) [Craven, P. and Wahba, G. (1979), ‘Smoothing Noisy Data with Spline Functions’, Numerical Mathematics, 31, 377–403], which establishes a single-pass method that penalises the fit by the trace of the smoother matrix assuming independent errors. We propose an extension to GCV in the context of correlated errors, which is motivated by a natural definition for residual degrees of freedom. The efficacy of the new method is investigated with a simulation experiment on a kernel smoother with bandwidth selection in local linear regression. Next, the winning methodology is illustrated by application to spatial modelling of fMRI data using a nonparametric semivariogram. We conclude with remarks about the heteroscedastic case and a potential maximum likelihood framework for Gaussian random processes.

Far Casting Cross-Validation

Cross-validation has long been used for choosing tuning parameters and other model selection tasks. It generally performs well provided the data are independent, or nearly so. Improvements have been suggested which address ordinary cross-validation’s (OCV) shortcomings in correlated data. Whereas these techniques have merit, they can still lead to poor model selection in correlated data or are not readily generalizable to high-dimensional data. The proposed solution, far casting cross-validation (FCCV), addresses these problems. FCCV withholds correlated neighbors in every aspect of the cross-validation procedure. The result is a technique that stresses a fitted model’s ability to extrapolate rather than interpolate. This generally leads to better model selection in correlated datasets. Whereas FCCV is less than optimal in the independence case, our improvement of OCV applies more generally to higher dimensional error processes and to both parametric and nonparametric model selection problems. To facilitate introduction, we consider only one application, namely estimating global bandwidths for curve estimation with local linear regression. We provide theoretical motivation and report some comparative results from a simulation experiment and on a time series of annual global temperature deviations. For such data, FCCV generally has lower average squared error when disturbances are correlated. Supplementary materials are available online.

A new class of semiparametric semivariogram and nugget estimators

Several authors have proposed nonparametric semivariogram estimators. Shapiro and Botha (1991) did so by application of Bochners theorem and Cherry et al. (1996) further investigated this technique where it performed favorably against parametric estimators even when data were generated under the parametric model. While the former makes allowances for a prescribed nugget and the latter outlines a possible approach, neither of these demonstrate nugget estimation in practice, which is essential to spatial modeling and proper statistical inference. We propose a modified form of this method, which admits practical nugget estimation and broadens the basis. This is achieved by a simple change to the basis and an appropriate restriction of the node space as dictated by the first root of the Bessel function of the first kind of order @n. The efficacy of this new unsupervised semiparametric method is demonstrated via application and simulation, where it is shown to be comparable with correctly specified parametric models while outperforming misspecified ones. We conclude with remarks about selecting the appropriate basis and node space definition.

Predictive Power of Standard Variables on Postsecondary Achievement in Arkansas.

Today, as many as 25 to 40 percent of students who attend college qualify for some form of remedial education program provided by postsecondary institutions (Kaye, Lord, Bottoms, Presson, & Cornet, 2006). Many colleges and universities view the inclusion of remediation as an integral part of their educational mission. However, the costs of such remedial programs are high. While estimates of actual costs vary greatly, most assume a national remediation expense of over one billion dollars annually. In Arkansas, the cost of remediation is over