SEARCH

SEARCH BY CITATION

Keywords:

  • Epilepsy;
  • Periictal SPECT;
  • CBF difference imaging;
  • Statistical parametric mapping (SPM)

Abstract

  1. Top of page
  2. Abstract
  3. METHODS
  4. Difference image analysis
  5. SPM image analysis
  6. RESULTS
  7. DISCUSSION
  8. REFERENCES

Summary:  Purpose: Statistical parametric mapping (SPM) is an image-analysis tool that assesses the statistical significance of cerebral blood flow (CBF) changes on a voxel-by-voxel basis, thereby removing the subjectivity inherent in conventional region-of-interest (ROI) analysis. Our platform of single-photon emission computed tomography (SPECT) ictal–interictal difference imaging in clinical epilepsy has been validated for localizing seizure onset. We extend the tools of SPM by further applying statistical measures for the significance of perfusion changes in individual patients to localize epileptogenic foci in patients with defined temporal lobe epilepsy by using paired scans in this preliminary study.

Methods: Twelve patients with pairs of periictal and interictal SPECT scans were analyzed in this comparison study between SPECT difference imaging and SPM difference analysis by using a reference database of paired normal healthy images. These 12 patients possessed seizure foci localized to the mesial temporal lobe as confirmed by surgical outcome and by hippocampal sclerosis on pathology. SPM was used to identify clusters of increased or decreased CBF in each patient in contrast to our control group.

Results: The regions having the most significant increased or decreased CBF by SPM analysis were in agreement with regions identified by conventional difference imaging and visual analysis by viewers blinded to the results of the SPM analysis. Differentiated further by time of radiopharmaceutical injection, six of seven patients injected within 100 s of seizure onset displayed hyperperfusion changes localized to the corresponding epileptogenic temporal lobe by both techniques. Among patients receiving injections after 100 s, both techniques showed primarily regions of hypoperfusion, which again were similar between these two methods.

Conclusions: The results provide strong evidence supporting SPM difference analysis in assessing regions of significant CBF change from baseline in concordance with our current clinically used technique of SPECT ictal–interictal difference imaging in epilepsy patients. Difference analysis using SPM could serve as a useful diagnostic tool in the evaluation of seizure focus in temporal lobe epilepsy.

Single-photon emission computed tomography (SPECT) provides important clinical information measuring regional cerebral blood flow changes in the evaluation of epileptic seizure foci. In general, SPECT has demonstrated hyperperfusion in the epileptogenic region periictally and hypoperfusion interictally. Zubal et al. (1) demonstrated improved localization using SPECT difference imaging, whereby interictal images are subtracted on a voxel-by-voxel basis following co-registration and normalization with ictal images. However, the limitation of this approach as with most others remains its degree of interobserver variability (1–3). Other methods of SPECT interpretation have increased diagnostic yield. Calculation of asymmetry indices between regions of interest (ROIs) and reference regions (semiquantitative analysis) has been performed by some investigators (4–6).

Statistical parametric mapping (SPM) is an increasingly established form of neuroimaging analysis to localize statistically significant changes in spatially normalized images on a voxel-by-voxel basis (7,8). Working within a standardized stereotactic space, SPM removes the subjectivity inherent in quantitation based on visual analysis. SPM has been applied to temporal lobe epilepsy including in the evaluation of SPECT (9). The purpose of this study was to determine the value of SPM in the analysis of ictal–interictal differences among a surgically defined group of temporal lobe epilepsy patients from our database in comparison with a technique our group established of SPECT ictal–interictal difference imaging.

METHODS

  1. Top of page
  2. Abstract
  3. METHODS
  4. Difference image analysis
  5. SPM image analysis
  6. RESULTS
  7. DISCUSSION
  8. REFERENCES

Patients

Twelve consecutive patients who met the following criteria were selected from the patients undergoing treatment for medically refractory epilepsy at the Yale Epilepsy program: All patients had a periictal and interictal SPECT scan performed during video- and scalp-EEG monitoring; all patients had a seizure focus localized to the left or right mesial temporal lobe based on a combination of scalp and/or intracranial EEG, magnetic resonance imaging (MRI), and positron emission tomography (PET); surgery was performed on all included patients with pathology demonstrating hippocampal sclerosis, and with a surgical outcome on medications of no seizures, or only auras for a minimum of 1 year follow-up. Six men and six women were represented in our group undergoing analysis. Mean patient age at time of SPECT imaging was 38 years (range, 17–65 years).

Healthy normal control subjects were obtained from a database developed by one of us (C.G.), for which each patient had received two SPECT scans. These repeated scans from seven control patients (nonneurologic) were used in the design matrix for our SPM comparison technique.

Data acquisition

For periictal studies, patients received injections during clinical and EEG-documented seizures. On noting seizure onset, a technologist performed an intravenous injection of Tc-99m–labeled hexamethyl-propylene amine-oxime (HMPAO; Medi-Physics, Amersham Healthcare, Arlington Heights, IL,U.S.A.), which distributes within the brain and represents perfusion at the time of injection. The time of seizure onset, seizure termination, and Tc-99m–HMPAO injection was determined for each patient by viewing the videotape and EEG recordings obtained during inpatient continuous audiovisual EEG monitoring. Interictal injections of Tc-99m-HMPAO were performed in these same patients after ≥24 h of no seizure activity.

SPECT images were acquired within 90 min after injection. Projection data were acquired on a Picker PRISM 3000 (Marconi Medical Systems, Cleveland Heights, OH, U.S.A.) mounted with high-resolution fan-beam collimators. Data were acquired into 128 × 128 matrices over a 40-min period during which each of the three heads makes a 120-degree orbit. Transverse slices are reconstructed by using the routine clinical filtered backprojection algorithm with a Chang attenuation correction. The available reconstruction package allows prefiltering of the projection data by using a selectable apodization filter whose cut-off frequency can be set to the position at which the image power spectrum is equal to the noise level in the projection images. Subsequently, ∼30 to 40 transverse slices covering the whole brain were reconstructed by using the ramp filter. All images were attenuation corrected by using an attenuation coefficient that accounts for scattered events in the photopeak (0.11/cm).

All patients also received volumetric MRI scans with high-resolution T1-weighted coronal images used for co-registration of anatomic features with the SPECT scans. SPECT and MRI image data were transferred to personal computer with Linux operating system (5.1; Red Hat Software, Inc.) for manipulation and data analysis.

Difference image analysis

  1. Top of page
  2. Abstract
  3. METHODS
  4. Difference image analysis
  5. SPM image analysis
  6. RESULTS
  7. DISCUSSION
  8. REFERENCES

Our standard method of SPECT difference imaging analysis was used, as described previously (1,10). The periictal SPECT was aligned with the interictal SPECT with a rigid-body transformation by using multiresolution optimization of normalized mutual information (11,12). Corresponding values in the periictal and interictal scans were normalized to account for differences in tracer dose. These differences were modelled as a simple linear scaling between all corresponding brain image values in the two scans. Empirically this is seeking the optimal fit of a line to the two-dimensional histogram of corresponding periictal and interictal values in the registered images. Fitting was achieved by an initial least-squares estimation. This estimate was then further refined by minimization of difference histogram entropy to provide an estimate robust to local blood-flow changes. Using this linear intensity correction estimate, a voxel-by-voxel subtraction was performed between the interictal and normalized periictal values. Two sets of difference images containing positive differences [increased regional cerebral blood flow (rCBF)] and negative differences (decreased rCBF) were then created for clinical inspection.

MRI scans were aligned with the interictal SPECT images by multiresolution optimization of normalized mutual information. This technique has demonstrated excellent accuracy in human brain studies (11). To verify the alignment of all MRI-to-SPECT and SPECT-to-SPECT image registrations, an interactively selected isocontour was displayed in three orthogonal planes and visually inspected (12). An example of aligned and normalized interictal SPECT, ictal SPECT, and MRI images are shown in Figure 1. Positive and negative SPECT differences were then superimposed onto the patient's MRI scan to aid in anatomic localization.

imageimageimage

Figure 1. Example of aligned and normalized SPECT and MRI imagges. Axial images displayed at the same plane of section A. Interictal SPECT. B. Ictal SPECT. C. T1-weighted MRI.

Results were analyzed visually by consensus of three experienced readers. Readers were blinded to the patients' identities and clinical histories. Up to three regions each of ictal CBF increases and ictal CBF decreases were quantified, and the regions listed in order of importance perceived by the readers. Differences were quantified by report of maximal percentage change (percentage increase or decrease by using the interictal as reference) in the identified regions by using the described linear intensity correction estimate.

SPM image analysis

  1. Top of page
  2. Abstract
  3. METHODS
  4. Difference image analysis
  5. SPM image analysis
  6. RESULTS
  7. DISCUSSION
  8. REFERENCES

Images were manipulated within MedX 3.28 (Sensor Systems, Inc., Sterling, VA, U.S.A.) containing SPM96 (Wellcome Department of Cognitive Neurology, Institute of Neurology, London, U.K.) software. Each subject's reconstructed SPECT images were realigned first with AIR (13) (rigid-body, six-parameter transformation) and then with the SPM-realign function. Volume images were made compatible for SPM through conversion into ANALYZE (AVW) format. Image pairs were co-registered by using each patient's interictal scan and each normal individual's initial scan as the reference image. SPM realigned each pair of scans creating a mean group image, which was derived from volume for subsequent normalization.

Images were spatially normalized in SPM to a standardized stereotactic space based on the Talairach and Tournoux atlas defined by automated definition of the anterior–posterior commissural (AC-PC) line (14). Spatial normalization was performed by using the PET template provided by SPM96 software. A nine-parameter linear transformation algorithm consistent with SPECT data was used to match each image to the template. Given the reduced spatial resolution of SPECT relative to PET, no affine parameters were used. Differences in global activity between scans were removed by proportional scaling with the default global scan value calculated in SPM96. To eliminate signals from highly variable nonbrain structures, all scans were cropped by using a 3-D contour that approximately defined the subdural space, drawn from a mean image of all normalized control scans. Images were smoothed by using gaussian kernel (10 × 10 × 12 mm) before statistical analysis.

SPM characterizes regionally specific differences within the adjusted data. Statistical parametric maps are spatially extended statistical processes used to distinguish these regional effects. The combination of the General Linear Model (GLM) and theory of gaussian random fields (RFT) allow statistical inferences to be made on region-specific effects in the brain (8). To examine images for differences in perfusion between scans obtained during periictal and interictal states, two fundamental comparisons were performed. SPM compared regions of both hyperfusion and hypoperfusion changes in our experimental group with healthy normal data pairs. A design matrix was created to assess regions of significant increase or decrease in blood flow between paired scans after subtraction to estimate variance at each voxel and the overall smoothness of the dataset.

The resulting set of values for each comparison generated a statistical parametric map of the t statistic (SPM{t}). These t values were then transformed by “gaussianization” to unit-normal distribution Z scores to apply the RFT. The resulting regions of difference were then examined at the cluster level: the significance of each cluster (a connected set of voxels, corresponding to a brain region of variable sizes) is estimated in SPM96 by using a bivariate distributional approximation of the number of voxels and peak Z-score in a region calculated with RFT. Analyses were performed with an uncorrected voxel-height threshold of p = 0.01 and an extent threshold (k) of 32 voxels. This cluster size was chosen to correspond to the level of resolution of our SPECT system. This threshold results in SPMs that contain only clusters above our system resolution, and so can be interpreted as physiologically meaningful. SPM image results were displayed in three orthogonal planes by using a “glass brain.” These generated regions of significance were then overlaid onto a normalized MR image within MedX to define the anatomic region(s) of perfusion alteration appropriately by using their x,y,z coordinates in Talairach space.

Up to three voxel clusters found by SPM were listed for CBF increases and decreases in order of Z-score significance as reported by SPM rather than by visual inspection. SPM analysis was performed independently by D.C., and anatomic locations of clusters were defined by H.B. Subcortical foci were excluded from both SPM analysis and difference image analysis.

RESULTS

  1. Top of page
  2. Abstract
  3. METHODS
  4. Difference image analysis
  5. SPM image analysis
  6. RESULTS
  7. DISCUSSION
  8. REFERENCES

Visual results from SPECT ictal–interictal difference imaging were compared with results obtained from SPM analysis of the same SPECT scans in 12 patients with surgically confirmed temporal lobe epilepsy. These studies were performed to assess the reliability of using a rigorous method of statistical analysis applied to SPECT neuroimaging in the localization of seizure foci. This technique is compared side to side with our prior method of difference imaging analysis.

Table 1 lists the anatomic region(s) along with both quantitative percentage changes seen on difference analysis and SPM results associated with the highest Z-scores. Difference image results are shown in the left column, and the SPM analysis is shown in the right column. Patients are listed in order of time of SPECT injection after seizure onset. Overall, a high degree of correspondence is evident between the anatomic regions of CBF increases and decreases identified by difference imaging and SPM. SPM results are reported by cluster p value and cluster size (k = number of voxels), and voxel Z-score. Difference analysis results are listed in the order reported by the readers, with the region thought most important listed first. Corresponding regions found by both techniques are listed in the same row, and are indicated by gray shading. Cluster-level significance of <0.05 is indicated by boldface. We limited ourselves, as in prior articles, to reporting only up to the three areas of maximal blood-flow change by both approaches.

Patients 1–7 were injected within 100 s of seizure onset. In six of these seven patients, both difference analysis and SPM localized regions of hyperperfusion to the correct temporal lobe. It is generally agreed that early rather than late injections and hyperperfusion rather than hypoperfusion changes are best localized by SPECT. The SPECT scans for patient 4 were of technically inferior quality, and neither method correctly identified the region of seizure onset. Patients 8–12 were all injected at a time >100 s after seizure onset with mostly CBF decreases reported by both analysis methods. If all patients are included, seven of 12 patients localized to the same region, based on CBF increases. Although it appears that CBF decreases are not reliable for identifying the region of seizure onset, the correspondence between these two methods is further strengthened by the fact that they generally identified the same regions even for late injections. Also in general, more regions of CBF changes are identified in Table 1 by SPM because up to three clusters were listed as long as they contained significant voxels. Application of more stringent criteria for voxel-level significance in SPM may be necessary to better parallel the established difference-analysis method.

In patients 2, 7, and 8, very large confluent foci were evident in the SPM results among CBF decreases, extending from one hemisphere into the other. For these patients, the analysis was repeated with more stringent thresholds (uncorrected voxel-height, p = 0.001 and k = 125), which produced more discrete foci (reported as such in Table 1). Distinct foci could not be resolved even with repeated thresholding in those patients with confluent foci extending only from one medial frontal lobe contralaterally. In these patients, results are instead reported as bifrontal. Such thresholds are not necessary to demonstrate our main results. If the original less stringent thresholds are applied to these three patients, our results are not altered significantly. Localization of positive and negative differences is unchanged, although negative changes do merge into larger clusters.

The side of surgery was the same as the side of significant SPECT increases (when present) and the side of EEG onset for all patients except patient 3, who had documented bilateral (right or left) onset seizures. For this patient, although the seizure recorded for SPECT imaging came from the left side based on EEG, the right side had greater seizure frequency and was eventually resected.

Figure 2 illustrates the temporal region of the seizure focus in patient 7. The difference images show a large perfusion increase in the right temporal lobe (64%) extending into the right inferior parietal lobe. In this figure, SPM results are displayed as perfusion changes overlayed onto the MRI image of only the reference brain disregarding skull, skin, and eyes. The SPM results show very similar perfusion increase (Z score = 5.94). For CBF decreases, both difference and SPM analysis identified corresponding regions in the left whole temporal (52%, Z score = 5.27) and right whole frontal lobes (53%, Z score = 4.97 but not statistically significant at the cluster level). The corresponding findings in this patient are representative for our study in general as indicated earlier.

image

Figure 2. Comparison of difference imaging and SPM. Images are from patient 7 in Table 1 (same patient as in Figure 1). Red represents CBF increases and blue CBF decreases. Left images were obtained with difference imaging. Right images were obtained with SPM, showing similar regions of hyper- and hypo-fusion to difference imaging.

Download figure to PowerPoint

DISCUSSION

  1. Top of page
  2. Abstract
  3. METHODS
  4. Difference image analysis
  5. SPM image analysis
  6. RESULTS
  7. DISCUSSION
  8. REFERENCES

A variety of tools are used in combination by clinicians to enhance their understanding of a patient's epileptogenic pathology including EEG, Wada testing, neuropsychiatric batteries, and functional neuroimaging modalities such as PET and SPECT. Localization of seizure foci remains primarily a qualitative approach relying on interpretation through visual analysis. We provide support here for the use of the models implemented in SPM for the objective analysis of SPECT images used in the management of epilepsy patients. We extend this approach by applying it within the framework of our successful and clinically used method of SPECT periictal–interictal difference imaging analysis. SPM provides a potentially significant clinical alternative to conventional ROI analysis.

The value of SPM in epilepsy is gaining acceptance in a variety of applications; however, most reported studies have focused on the identification of abnormalities in single patient images, in the periictal or interictal state. Groups have examined the utility of SPM in 18FDG PET and 11C-flumazenil PET in similar mesial temporal lobe epilepsy patients, examining grouped patient data against controls as well as for individual patient scans (15,16). The sensitivity of SPM for detection of regional abnormalities in single SPECT scans compared with a reference group has been the subject of recent reports of simulations (17). Lee et al. (9) also recently examined the utility of the subtraction method, by using a normal brain database. In our present study, to a high degree, our SPM results reproduce our findings of subjectively drawing ROIs and measuring the quantitative percentage changes between paired ictal–interictal scans. In doing so, SPM eliminates the interrater variability component of imaging analysis.

We reported the three most significant changes found in an SPM in order of voxel-level significance, even if the cluster-level significance was low. In actuality, the number of regions reported by SPM that have cluster level significance <0.05 listed in the table is similar to the number of regions that we reported by visual analysis. Our results may further suggest that, with appropriate spatial thresholding relative to the known resolution of the imaging system, those clusters with significance <0.05 may be the most physiologically relevant. Furthermore, our SPM results reaffirm the prior findings based on difference analysis reporting the transition from hyper- to hypoperfusion after radiopharmaceutical injection at ∼100 s after seizure onset (10).

Traditional forms of seizure focus analysis using SPECT rely on the advantage gained from acquiring an image of rCBF that is representative of the brain's state at the time of radiopharmaceutical injection. Our difference analysis technique allowed variance estimates to be pooled across the brain when quantifying differences between periictal and interictal scans. Statistical parametric maps incorporate error variance or the reliability of measuring a change effect. They further remove the confounding influences of intersubject global perfusion differences. Using our reference set of paired control images introduces scan-to-scan variability at each voxel averaged across all subjects, and a smoothness measure for the types of scans analyzed. In our present application of SPM, the within-subject, scan-to-scan variability is estimated from the healthy normal control group; however, the subject-to-subject variability of these differences at each specific voxel is assumed to be low. This may be a limitation in the model implemented in SPM, as it is a fixed-effects model. Although there has been progress in the development of random-effects analyses for functional MRI data with this platform, those techniques are not readily applicable to SPECT and PET data. A different approach, such as the use of nonparametric models, may circumvent this problem and will be the subject of future studies.

Our increasing understanding of SPM applications may yield a more direct comparison of the effect size measured by percentage change by our difference imaging technique. Interestingly, in some patients, extratemporal as well as subcortical structures displayed regions of significant perfusion change, which we will continue to explore. We plan to extend our future analyses by increasing our database of normal healthy paired scans. A primary limitation of this study was our small healthy normal sample size. Our reduced degrees of freedom introduced type II errors, diminishing the sensitivity of our analysis and the more conservative inferences from generated t statistic to normalized Z unit score. Although we are confident of our results, increasing sample size in future analyses would further remove any probability of missing any true regions of perfusion difference.

An attractive feature about our approach is that it requires only the two scans in question, making it clinically feasible and practical. The study demonstrates, through a very different method and model, that the localization and direction of change are consistent. Traditional qualitative analysis measured an index of change in percentage form, but it could not enhance the spatial extent or degree of change as provided by the significance of a p value for an isolated focus. Many possible extensions with SPM are possible now that we have convergent information that localization is consistent. For example, in a group of patients with similar foci, one can address the regions or networks that change in activity as a function of the activation of the epileptic focus, along with information about the time course of that relation. Our preliminary trials with SPM99 software with a small sampling of patients also verified the clusters identifying both hyper- and hypoperfusion changes reported here.

The objective advantages of SPM to ictal–interictal difference imaging deserve further evaluation for application in a clinical setting. Its high degree of correspondence within our present framework of SPECT difference imaging provides increasing support that once further refined, similar SPM analyses can provide a relatively quick and reliable means for determining epileptogenic foci in individual patients as well as through cross-institutional comparisons of patient data.

Acknowledgment: This study was supported by NIH grant NS35674 (RO1) and Dana Foundation Clinical Hypotheses in Neuroscience Award.

REFERENCES

  1. Top of page
  2. Abstract
  3. METHODS
  4. Difference image analysis
  5. SPM image analysis
  6. RESULTS
  7. DISCUSSION
  8. REFERENCES
  • 1
    Zubal IG, Spencer SS, Imam K, et al. Difference images calculated from ictal and interictal technetium-99m-HMPAO SPECT scans of epilepsy. J Nucl Med 1995;36:6849.
  • 2
    Spanaki MV, Spencer SS, Wisniewski G, et al. Evolution and localization of postictal blood flow changes in partial seizures demonstrated by SPECT: use of quantitative difference images. J Epilepsy 1998;11:2533.
  • 3
    O'Brien TJ, So EL, Mullan BP, et al. Subtraction ictal SPECT co-registered to MRI improves clinical usefulness of SPECT in localizing the surgical seizure focus. Neurology 1998;50:44554.
  • 4
    Stefan H, Bauer J, Feistel H, et al. Regional cerebral blood flow during focal seizures of temporal and frontocentral onset. Ann Neurol 1990;27:1626.
  • 5
    Rowe CC, Berkovic SF, Austin MC, et al. Patterns of postictal cerebral blood flow in temporal lobe epilepsy. Neurology 1991;41:1096103.
  • 6
    Duncan R, Patterson J, Roberts R, et al. Ictal/postictal SPECT in the pre-surgical localization of complex partial seizures. J Neurol Neurosurg Psychiatry 1993;56:1418.
  • 7
    Friston KJ, Holmes A, Poline JB, et al. Detecting activations in PET and fMRI: levels of inference and power. Neuroimage 1996;40:22335.
  • 8
    Friston KJ, Holmes AP, Worsley KJ, et al. Statistical parametric maps in functional imaging: a general linear approach. Hum Brain Mapp 1995;2:89210.
  • 9
    Lee JD, Kim HJ, Lee BI, et al. Evaluation of ictal brain SPECT using statistical parametric mapping in temporal lobe epilepsy. J Nucl Med 2000;27:165865.
  • 10
    Avery RA, Spencer SS, Spanaki MV, et al. Effect of injection time on postictal SPECT perfusion changes in medically refractory epilepsy. Eur J Nucl Med 1999;26:8306.
  • 11
    West J, Fitzpatrick JM, Wang MY, et al. Comparison and evaluation of retrospective intermodality image registration techniques. J Comput Assist Tomogr 1997;21:55466.
  • 12
    Fitzpatrick JM, Hill DLG, Shyr Y, et al. Visual assessment of the accuracy of retrospective registration of MR and CT images of the brain. IEEE Trans Med Imaging 1998;17:57185.DOI: 10.1109/42.730402
  • 13
    Imran MB, Kawashima R, Sato K, et al. Mean regional cerebral blood flow images of normal subjects using technetium-99m-HMPAO by automated image registration. J Nucl Med 1998;39:2037.
  • 14
    Talairach J, Tournoux P. Co-planar stereotaxic atlas of the human brain. Stuttgart: Thieme, 1988.
  • 15
    Van Boagert P, Massager N, Tugendhaft P, et al. Statistical parametric mapping of regional glucose metabolism in mesial temporal lobe epilepsy. Neuroimage 2000;12:12938.DOI: 10.1006/nimg.2000.0606
  • 16
    Koepp MJ, Hammers A, Labbe C, et al. 11C-Flumazenil PET in patients with refractory temporal lobe epilepsy and normal MRI. Neurology 2000;54:3329.
  • 17
    Stamatakis EA, Glabus MF, Wyper DJ, et al. Validation of statistical parametric mapping (SPM) in assessing cerebral lesions: a simulation study. Neuroimage 1999;10:397407.DOI: 10.1006/nimg.1999.0477
  • 18
    Studholme C, Hill DLG, Hawkes DJ. An overlap invariant entropy measure of 3D medical image alignment. Pattern Recogn 1999;32:7186.
  • 19
    Studholme C, Hill DLG, Hawkes DJ. Automated three-dimensional registration of magnetic resonance and positron emission tomography brain images by multi-resolution optimisation of voxel similarity measures. Med Phys 1997;24:2535.DOI: 10.1118/1.598130