Group specific optimisation of fMRI processing steps for child and adult data.

Motion is a major issue in functional magnetic resonance imaging (fMRI) dataseries and causes artifacts or increased overall noise obscuring signals of interest. It is particularly important to be able to control for and correct these artifacts when dealing with child data. We analysed the data from 35 children (4-8 years old) and 13 adults (18-30 years old) during an emotional face paradigm. The children were split into low and high motion groups on the basis of having less or more than an estimated maximal movement of one voxel (3.75 mm) and one degree of rotation in any motion direction between any pair of scans in the run. Several different preprocessing steps were evaluated for their ability to correct for the excess motion using agnostic canonical variates analysis (aCVA) in the NPAIRS (Nonparametric, Prediction, Activation, Influence, Reproducibility, re-Sampling) framework. The adult dataset was reasonably stable whereas the motion-prone child datasets benefited greatly from motion parameter regression (MPR). Motion parameter regression had a strong beneficial impact on all datasets, a result that was largely unaffected by other preprocessing choices; however, motion correction on its own did not have as much impact. The low motion child group subjected to MPR had reproducibility values at par with those of the adult group, but needed almost twice as many subjects to achieve this result, indicating weaker responses in young children. The aCVA showed greater sensitivity to the task response pattern than the mixed effects general linear model (mGLM) in the expected face processing regions, although the mGLM showed more responses in some other areas. This work illustrates that preprocessing choices must be made in a group-specific fashion to optimise fMRI results.

Functional connectivity metrics during stroke recovery.

We explore functional connectivity in nine subjects measured with 1.5T fMRI-BOLD in a longitudinal study of recovery from unilateral stroke affecting the motor area (Small et al., 2002). We found that several measures of complexity of covariance matrices show strong correlations with behavioral measures of recovery. In Schmah et al. (2010), we applied Linear and Quadratic Discriminants (LD and QD) computed on a principal components (PC) subspace to classify the fMRI volumes into "early" and "late" sessions. We demonstrated excellent classification accuracy with QD but not LD, indicating that potentially important differences in functional connectivity exist between the early and late sessions. Motivated by Mclntosh et al. (2008), who showed that EEG brain-signal variability and behavioral performance both increased with age during development, we investigated complexity of the covariance matrix for this longitudinal stroke recovery data set. We used three complexity measures: the sphericity index described by Abdi (2010); "unsupervised dimensionality", which is the number of PCs that minimizes unsupervised generalization error of a covariance matrix (Hansen et al., 1999); and "QD dimensionality", which is the number of PCs that minimizes the classification accuracy of QD. Although these approaches measure different kinds of complexity, all showed strong correlations with one or more behavioral tests: nine-hole peg test, hand grip test and pinch test. We could not demonstrate that either sphericity or unsupervised dimensionality were significantly different for the "early" and "late" sessions using a paired Wilcoxon test. However, the amount of relative behavioral improvement was correlated with sphericity of the overall covariance matrix (pooled across all sessions), as well as with the divergence of the eigenspectra between the "early" and "late" covariance matrices. Complexity measures that use the number of PCs (which optimize QD classification or unsupervised generalization) were correlated with the behavioral performance of the final session, but not with the relative improvement. These are suggestive, but limited, results given the sample size, restricted behavioral measurements and older 1.5T BOLD data sets. Nevertheless, they indicate one potentially fruitful direction for future data-driven fMRI studies of stroke recovery in larger, better-characterized longitudinal stroke data sets recorded at higher field strength. Finally, we produced sensitivity maps (Kjems et al., 2002) corresponding to both linear and quadratic discriminants for the "early" vs. "late" classification. These maps measure the influence of each voxel on the class assignments for a given classifier. Differences between the scaled sensitivity maps for the linear and quadratic discriminants indicate brain regions involved in changes in functional connectivity. These regions are highly variable across subjects, but include the cerebellum and the motor area contralateral to the lesion.

Data warehousing methods and processing infrastructure for brain recovery research.

In order to accelerate translational neuroscience with the goal of improving clinical care it has become important to support rapid accumulation and analysis of large, heterogeneous neuroimaging samples and their metadata from both normal control and patient groups. We propose a multi-centre, multinational approach to accelerate the data mining of large samples and facilitate data-led clinical translation of neuroimaging results in stroke. Such data-driven approaches are likely to have an early impact on clinically relevant brain recovery while we simultaneously pursue the much more challenging model-based approaches that depend on a deep understanding of the complex neural circuitry and physiological processes that support brain function and recovery. We present a brief overview of three (potentially converging) approaches to neuroimaging data warehousing and processing that aim to support these diverse methods for facilitating prediction of cognitive and behavioral recovery after stroke, or other types of brain injury or disease.

A regional covariance approach to the analysis of functional patterns in positron emission tomographic data.

This article provides a complete description of the subprofile scaling model (SSM) approach to the analysis of positron emission tomography (PET) data. The goals and assumptions underlying the development of SSM are outlined, and its strengths and weaknesses are discussed. It is demonstrated that all obtainable information about regional ratios can, in theory, be derived from the SSM regional covariance patterns. A general constraint on the ability to effectively remove global variation while identifying region-specific information about PET data sets is outlined and discussed within the SSM framework. Finally, an extension of the SSM technique to the generation of disease-specific covariance patterns is demonstrated for paraneoplastic cerebellar degeneration, the acquired immune deficiency syndrome dementia complex, and Parkinson's disease, and the importance of multidimensional descriptions of disease, such as may be obtained from PET data using SSM, is emphasized.

Commentary and opinion: I. Principal component analysis, variance partitioning, and "functional connectivity".

We briefly review the need for careful study of "variance partitioning" and "optimal model selection" in functional positron emission tomography (PET) data analysis, emphasizing the use of principal component analysis (PCA) and the importance of data analytic techniques that allow for heterogeneous spatial covariance structures. Using an [15O]water dataset, we demonstrate that--even after data processing--the intrasubject signal component of primary interest in baseline activation studies constitutes a very small fraction of the intersubject variance. This small intrasubject variance component is subtly but significantly changed by using analysis of covariance instead of scaled subprofile model processing before applying PCA. Finally, we argue that the concept of "functional connectivity" should be interpreted very generally until the relative roles of inter- and intrasubject variability in both disease and normal PET datasets are better understood.

Effects of percent thresholding on the extraction of [18F]fluorodeoxyglucose positron emission tomographic region-of-interest data.

Although we and others have employed a thresholding strategy to extract "peak" values from positron emission tomographic (PET) regions of interest (ROIs), the effects of peak picking on fitted fluorodeoxyglucose rate constants, regional metabolic rate for glucose (rCMRglc) profiles, patterns of regional metabolic covariation, and PET-neurobehavioral correlations have not been systematically investigated. Our results suggest that under some commonly encountered imaging conditions percent thresholding may increase sensitivity to regional activation; however, the effect of thresholding is determined by a number of factors, including the relative magnitude of regional activation, ROI size, and the specific threshold selected. The difference-annulus concept is proposed as a means to study the effects of different region drawing and thresholding strategies, and to determine if a given ROI contains one and only one source of covarying metabolic activity.

Scaled subprofile model: a statistical approach to the analysis of functional patterns in positron emission tomographic data.

The data obtained from measurements of regional rCMRglu using [18F]fluorodeoxyglucose (FDG)/positron emission tomographic (PET) data contain more structure than can be identified with group mean rCMRglu profiles or regional correlation coefficients. This additional structure is revealed by a novel mathematical-statistical model of regional metabolic interactions that explicitly represents rCMRglu profiles as a combination of region-independent global effects, a group mean pattern and a mosaic of interacting networks. In its application to FDG/PET data, this model removes global subject effects [global scaling factors (GSFs)] and a group mean pattern (profile) so as to maximize statistical power for the detection and simultaneous discovery of all networks of two or more regions that form a significant and consistent linearly covarying pattern. The model approach presented here was applied to the combined rCMRglu data from 12 demented AIDS patients and 18 normal controls: Two significant metabolic covariance pattern descriptors that together accounted for 71 to 96% of the rCMRglu/GSF variation across subjects for 22/28 regions in the AIDS group were extracted. Each descriptor was found to be highly correlated with performance on several neuropsychological tests, providing independent validation of the analysis technique as a means of discovering and describing behaviorally related components of group rCMRglu profiles.

An improved approach for measurement of regional cerebral rate constants in the deoxyglucose method with positron emission tomography.

A new experimental approach was designed to measure regional rate constants for [18F]2-fluoro-2-D-deoxyglucose in the three-compartment model. A programmed infusion was used to keep the plasma tracer concentration constant for the first 20 min. Positron emission tomography images of the brain were taken every minute for 20 min in a low-resolution mode, then every 15-20 min for 2-3 h in a medium-resolution mode. Two simplified operational equations for the calculation of the regional rate constants were derived that incorporated the contribution of the vascular compartment to tissue activity. The first equation was applied to data collected during the initial 20 min (when the concentration of plasma tracer was constant) to estimate the values of k1*, k2*, k3*, and the fraction of the vascular compartment. The second equation was applied to data collected during the whole experimental period to find the value of k4* and to provide a better estimate of k3*. The regional rate constants measured experimentally in three dogs and a brain tumor patient agreed well with those in the literature. This method permits estimation of the local CMRglu under pathological conditions using regionally measured rate constants and provides new information on the pathophysiological meaning of the rate constants.

Absolute quantitation in neurological PET: do we need it?

This article addresses the question posed in the title by examining the effects of parameters traditionally associated with improved absolute quantitation, on the analysis of 12 acquired immune deficiency syndrome dementia complex (ADC) patients compared to a normal control group. Results are discussed within the framework of the subprofile scaling model (SSM) for analyzing patterns of regional covariation. It is demonstrated that the ability to extract measures of group discrimination and disease progression are unaffected by (1) limited improvements in image resolution, (2) the use of transmission scan smoothing, (3) the application of a scatter deconvolution correction, and (4) converting region-of-interest measurements of counts per voxel to measurements of regional CMRglc. This "robustness" of the SSM approach is partly due to the extraction of disease-related subject weights, independent of any subject's global scaling effects. It is argued that other analysis techniques that initially reduce intersubject variation (e.g., using regional ratios or normalizing by global metabolic rates before applying traditional multivariate procedures) lack analytic features that may be important to identify multidimensional, disease-related image patterns. Based on the ADC patient data, it is concluded that measures of group discrimination and disease progression will not necessarily benefit from the organization of parameters traditionally associated with improved absolute quantitation.

Principal component analysis and the scaled subprofile model compared to intersubject averaging and statistical parametric mapping: I. "Functional connectivity" of the human motor system studied with [15O]water PET.

Using [15O]water PET and a previously well studied motor activation task, repetitive finger-to-thumb opposition, we compared the spatial activation patterns produced by (1) global normalization and intersubject averaging of paired-image subtractions, (2) the mean differences of ANCOVA-adjusted voxels in Statistical Parametric Mapping, (3) ANCOVA-adjusted voxels followed by principal component analysis (PCA), (4) ANCOVA-adjustment of mean image volumes (mean over subjects at each time point) followed by F-masking and PCA, and (5) PCA with Scaled Subprofile Model pre- and postprocessing. All data analysis techniques identified large positive focal activations in the contralateral sensorimotor cortex and ipsilateral cerebellar cortex, with varying levels of activation in other parts of the motor system, e.g., supplementary motor area, thalamus, putamen; techniques 1-4 also produced extensive negative areas. The activation signal of interest constitutes a very small fraction of the total nonrandom signal in the original dataset, and the exact choice of data preprocessing steps together with a particular analysis procedure have a significant impact on the identification and relative levels of activated regions. The challenge for the future is to identify those preprocessing algorithms and data analysis models that reproducibly optimize the identification and quantification of higher-order sensorimotor and cognitive responses.

Improved resolution for PET volume imaging through three-dimensional iterative reconstruction.

UNLABELLED: It has been demonstrated that in two-dimensional iterative reconstruction, a resolution model can improve image resolution while controlling noise. With the lower noise levels of three-dimensional PET volume imaging, these iterative reconstruction advantages may be extended to three dimensions to further improve the reconstructed image resolution. METHODS: We have implemented three-dimensional versions of iterative filtered backprojection (IFBP) and the maximum likelihood by expectation maximization (ML-EM) reconstruction algorithms and applied them to three-dimensional PET volume datasets. The results were compared to images obtained using the standard three-dimensional reprojection reconstruction (3DRP) algorithm. RESULTS: For IFBP with 15 iterations and no regularization compared to 3DRP, both using a ramp filter, the transaxial resolution improved 52%, and the axial resolution improved 39%. With a strong regularization, the transaxial and axial resolution improvements were reduced to 6% and 5%, respectively. If a Hanning roll-off is applied to the ramp filter in the transaxial direction, the transaxial resolution for IFBP without regularization improved 35% compared to 3DRP; with regularization the improvement dropped to 19%. The axial resolution for IFBP and 3DRP was unaffected by this transaxial smoothing in the reconstruction filter. With the same Hanning roll-off, the noise for IFBP without regularization increased by a factor of 6 compared to 3DRP; with regularization the noise was increased only by a factor of 3. Compared to IFBP, the three-dimensional ML-EM reconstruction produced similar resolution improvements with a much smaller increase in noise and slower convergence. Resolution improvements from both IFBP and ML-EM reconstructions are visually apparent in three-dimensional FDG brain images and result in increased activation signals in a three-dimensional [15O]water functional activation study. CONCLUSION: Our results demonstrate that resolution improvement is possible for IFBP and ML-EM compared to 3DRP with or without noise increase.

Display of merged multimodality brain images using interleaved pixels with independent color scales.

UNLABELLED: This article reviews common methods for two-dimensional display of registered multimodality brain images and describes a software package for presentation of merged MRI and PET images that runs on a workstation with an eight-bit color display. The software package displays merged brain images from multiple modalities in a way that is readily manipulated, visually pleasing and easy to interpret. The display method used, i.e., interleaving of alternate pixels with independent color scales, is effective in producing merged images with high contrast-detail for each volume. Interleaving images from different volumes creates unusual perceptual effects, one of which is the apparent camouflage of low-contrast signals by high values in the paired volume. METHODS: The camouflage effect was thought to arise from perceptual merging of adjacent pixels. An observer experiment was conducted to investigate this tendency of high-activity PET data to obscure low-contrast detail in interleaved MRI data in spite of the digital independence of neighboring pixels. Four observers were presented with 20 combinations of signal plus background targets with uniform mask images, using a two-alternative forced-choice experimental design with 50 trials per combination. RESULTS: The psychophysical evaluation of the ability of human observers to detect the simple test objects in an interleaved image presentation indicated a statistically significant camouflage effect of one volume on the other for some combinations of target and mask. The concept of perceptual merging of adjacent pixels was able to predict which combinations caused the greatest degradations in performance. CONCLUSIONS: The image interleaving approach to the display of two-dimensional slices from registered image volumes makes efficient use of an eight-bit color display. Contrast resolution of both individual volumes is high compared with that in other techniques and the volumes are presented in familiar color scales. However, the method yields an unexpected camouflage effect that tends to obscure low-contrast signals. The practical effect of such camouflage on the interpretation of clinical images remains to be investigated.

Effect of selecting a fixed dephosphorylation rate on the estimation of rate constants and rCMRGlu from dynamic [18F] fluorodeoxyglucose/PET data.

Several publications have discussed the estimation and physiologic significance of regional [18F]fluorodeoxyglucose (FDG) rate constants and metabolic rates. Most of these studies analyzed dynamic data collected over 45-60 min; three rate constants (k1-k3) and blood volume (Vb) were estimated and the regional cerebral metabolic rate for glucose (rCMRGlu) was subsequently derived using the measured blood glucose value and a regionally invariant value of the lumped constant (LC). The dephosphorylation rate constant (k4) was either neglected, or a fixed value was used in the estimation procedure to obtain the remaining parameters. To compare the rate constants obtained by different authors using different values of k4 is impossible without knowledge of the effect of selecting different fixed values of k4 (including zero) on the estimated rate constants and rCMRGlu. Based on our analysis of FDG/PET data from nine normal volunteer subjects, we conclude that inclusion of a fixed value for k4, in spite of a scaling effect on the absolute values of model parameters, has no effect on the coefficient of variation (CV) of within- and between-subject parameter estimates and glucose metabolic rates.

Stability of regional cerebral glucose metabolism in the normal brain measured by positron emission tomography.

Cerebral glucose utilization (LCMRGI) was measured using the [18F]fluorodeoxyglucose method with PET in two groups of ten healthy young volunteers, each scanned in a resting state under different methodological conditions. In addition, five subjects had a second scan within 48 hr. Mean hemispheric values averaged 45.8 +/- 3.3 mumol/100 g/min in the right cerebral hemisphere and 47.0 +/- 3.7 mumol/100 g/min in the left hemisphere. A four-way analysis of variance (group, sex, region, hemisphere) was carried out on the results using three different methods of data manipulation: (a) the raw values of glucose utilization, (b) LCMRGI values "normalized" by the mean hemispheric gray matter LCMRGI value, and (c) log transformed LCMRGI values. For all analysis techniques, significantly higher LCMRGI values were consistently seen in the left mid and posterior temporal area and caudate nucleus relative to the right, and in the right occipital region relative to the left. The coefficient of variation of intrasubject regional differences (9.9%) was significantly smaller than the coefficient of variation for regions between subjects (16.5%). No differences were noted between the sexes and no effect of repeat procedures was seen in subjects having multiple scans. In addition, inter-regional LCMRGI correlations were examined both in values from the 20 normal subjects, as well as in a set of hypothetical "abnormal" values. Results were compared with those reported from other PET centers; despite certain methodological differences, the intersubject and inter-regional variation of LCMRGI is fairly constant.

[13N]cisplatin PET to assess pharmacokinetics of intra-arterial versus intravenous chemotherapy for malignant brain tumors.

The biodistribution, blood clearance, and in vivo transformation of cisplatin (cisdiaminedichloroplatinum, DDP) were studied in rats using 13N-labeled and unlabeled DDP. Following the i.v. injection of [13N]DDP, virtually no 13N activity was detected in brain tissue, and no measurable amount of the 13N label was displaced from [13N]DDP. Based on these results, [13N]DDP/positron emission tomographic (PET) scans were performed in two glioblastoma patients undergoing Phase II intra-arterial (i.a.) DDP chemotherapy: [13N]DDP was infused i.v. over 13-15 min, during which time serial PET scans were obtained. One hour later, [13N]DDP mixed with cold DDP (100 mg/m2 therapeutic dose) was infused at the same rate i.a., and a second sequence of PET scans was acquired. The pharmacologic advantage of i.a. administration was calculated as the ratio of integrated tumor/brain count ratios for the i.a. and i.v. studies. Our preliminary results demonstrate the feasibility of quantifying the pharmacologic advantage of i.a. DDP chemotherapy in individual brain tumor patients using [13N]DDP and PET.

Abnormal cerebral glucose metabolism in HIV-1 seropositive subjects with and without dementia.

UNLABELLED: This study was undertaken in order to extend our previous finding of relative basal ganglia hypermetabolism in AIDS dementia complex (ADC) and to develop clinically useful metabolic indices of CNS involvement in HIV-seropositive (HIV+) subjects. METHODS: Twenty-one HIV+ subjects (11 with AIDS) underwent FDG-PET scanning; 12 had a follow-up scan at 6 mo and 4 had a third scan at 12 mo. Forty-three age-matched heterosexual volunteers served as controls. FDG-PET scanning was performed with arterial blood sampling, and scan data were analyzed using the Scaled Subprofile Model (SSM) with principal component analysis. RESULTS: SSM/principal component analysis of the combined (HIV+ and controls) FDG-PET dataset extracted two major disease-related metabolic components: (a) a nonspecific indicator of cerebral dysfunction, which was significantly correlated with age, cerebral atrophy and ADC stage and (b) the striatum, which was heavily weighted (relatively hypermetabolic) and appeared to provide a disease-specific measure of early CNS involvement. CONCLUSION: FDG-PET scans provide quantitative measures of abnormal functional connectivity in HIV-seropositives-with or without AIDS or ADC. These measures, which are robust across centers with respect to instrumentation, scanning technique and disease severity, appear to track the progression of CNS involvement in patients with subclinical neurologic or neuropsychologic dysfunction.

Comparison of voxel- and volume-of-interest-based analyses in FDG PET scans of HIV positive and healthy individuals.

UNLABELLED: Abnormal glucose metabolic patterns in the brain have been reported for HIV-1 seropositive (HIV+) patients when compared with seronegative healthy individuals. The metabolic covariance pattern obtained from voxel- or volume-of-interest (VOI)-based multivariate data analysis techniques can be used to characterize disease and potentially to detect and monitor disease progression in the early stage of AIDS dementia complex. However, the arbitrary smoothing typically applied to PET data during reconstruction and processing to reduce noise has an unknown effect on the data, especially for the voxel-based analysis. METHODS: To investigate the impact of smoothing on a discrimination task, we applied principal component analysis with scaled subprofile-model preprocessing (SSM/PCA) followed by Fisher discriminant analysis to FDG PET data that were reconstructed and processed with different degrees of smoothing. Receiver operating characteristic curves were used to compare the ability of derived metabolic covariance patterns to discriminate HIV+ patients from healthy volunteers. RESULTS: For the voxel-based analysis, we found that the spatial distribution of voxel weights from the SSM/PCA analysis suggested edge effects along major tissue and cerebrospinal fluid boundaries, indicative of a disease-specific pattern of cerebral atrophy for the HIV+ patients. In terms of its discrimination performance, this metabolic covariance pattern is stable and insensitive to a wide range of smoothing kernels, except for ramp reconstruction and Hanning reconstruction with 7 x 7 x 7 block smoothing. In these reconstructions, the discrimination performance decreased as a result of high image noise and excessive smoothing, respectively. Our results also indicated that if sufficient variance from the VOI measurements is included, the overall performance of a conventional VOI-based analysis can be similar to that of the voxel-based analysis for the same discrimination task. However, the VOI-based analysis performed poorly at low false-positive fraction and is less tolerant to noise in the metabolic covariance pattern than the voxel-based analysis. CONCLUSION: We have obtained a unique covariance pattern of brain glucose metabolism for HIV+ patients compared with healthy volunteers. Discrimination based on this covariance pattern was found to be insensitive to a wide range of image smoothness.

Iterative algebraic reconstruction algorithms for emission computed tomography: a unified framework and its application to positron emission tomography.

In this paper, a unified framework of iterative algebraic reconstruction for emission computed tomography (ECT) and its application to positron emission tomography (PET) is presented. The unified framework is based on an algebraic image restoration model and contains conventional iterative algebraic reconstruction algorithms: ART, SIRT, Landweber iteration (LWB), the generalized Landweber iteration (GLWB), the steepest descent method (STP), as well as iterative filtered backprojection (IFBP) reconstruction algorithms: Chang's method, Walters' method, and a modified iterative MAP. The framework provides an effective tool to systematically study conventional iterative algebraic algorithms and IFBP algorithms. Based on this framework, conventional iterative algebraic algorithms and IFBP algorithms are generalized. It is shown from the algebraic point of view that IFBP algorithms are not only excellent methods for correction of attenuation (either uniform or nonuniform) but are also good general iterative reconstruction algorithms (they can be applied to either attenuated or attenuation-free projections and converge very fast). The convergence behavior of iterative algebraic algorithms is discussed and insight is drawn into the fast convergence property of IFBP algorithms. A simulated PET system is used to evaluate IFBP algorithms and LWB in comparison with the maximum likelihood estimation via expectation maximization algorithm (MLE-EM) and the filtered backprojection (FBP) algorithm. The simulation results indicate that for both attenuation-free projection and attenuated projection cases IFBP algorithms have a significant computational advantage over LWB and MLE-EM, and have performance advantages over FBP in terms of contrast recovery and/or noise-to-signal ratios (NSRs) in regions of interest.

Comparison of matched BOLD and FAIR 4.0T-fMRI with [15O]water PET brain volumes.

Valid comparisons of functional activation volumes from fMRI and PET require accurate registration, matched spatial resolution, and if possible matched noise. We coregistered 4.0T-fMRI and PET volumes, using a series of linear and nonlinear transformations applied to the PET volumes. Because of the limited number of fMRI slices that were available, PET volumes were transformed to the fMRI space. Since 4.0T-fMRI and 4.0T-MRI volumes have significant spatial distortion due to magnet inhomogeneities, high resolution 1.5T-MRI volumes were nonlinearly transformed to 4.0T-MRI volumes as part of the transformation chain. The smoothing effects of these registration transformations were measured, in order to match the spatial resolution of the coregistered fMRI and PET volumes. Spatial resolution of the transformed PET volumes in the fMRI space was degraded by up to 60% due to the transformation process. Due to both the image acquisition characteristics and the coregistration process, the transformed PET volumes had a spatial resolution that was lower than that of tMRI. Therefore, significant smoothing of fMRI volumes was necessary to match their spatial resolution with that of the transformed PET volumes. Matching the spatial resolution of the fMRI volumes to those of the transformed PET volumes was achieved by matching the shape of their point spread functions. In order to do this, Gaussian kernels were employed to smooth the fMRI volumes. We were unable to simultaneously match the resolution and noise of fMRI and PET signals in the motor cortex. Activation maps derived from transformed PET and smoothed fMRI volumes were compared. Contralateral motor cortex was active in all modalities but there were large variations in the size of the activated region and its signal to noise ratio across BOLD, FAIR, and PET images within each subject. Nevertheless, the relative CBF changes measured by FAIR were consistent with those determined by PET.

Practical tradeoffs between noise, quantitation, and number of iterations for maximum likelihood-based reconstructions.

Emission computerised tomography images reconstructed using a maximum likelihood-expectation maximization (ML)-based method with different reconstruction kernels and 1-200 iterations are compared to images reconstructed using filtered backprojection (FBP). ML-based reconstructions using a single pixel (SP) kernel with or without a sieve filter show no quantitative advantage over FBP except in the background where a reduction of noise is possible if the number of iterations is kept small (<50). ML-based reconstructions using a Gaussian kernel with a multipixel full-width-at-half-maximum (FWHM) and a large number of iterations (200) require a sieve filtering step to reduce the noise and contrast overshoot in the final images. These images have some small quantitative advantages over FBP depending on the structures being imaged. It is demonstrated that a feasibility stopping criterion controls the noise in a reconstructed image, but is insensitive to quantitation errors, and that the use of an appropriate overrelaxation parameter can accelerate the convergence of the ML-based method during the iterative process without quantitative instabilities.