SEARCH

SEARCH BY CITATION

Keywords:

  • SPECT;
  • Epilepsy;
  • Cerebral blood flow;
  • Sensitivity and specificity;
  • Statistical parametric mapping (SPM)

Abstract

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgments
  7. REFERENCES

Summary: Purpose: The goal of neuroimaging in epilepsy is to localize the region of seizure onset. Single-photon emission computed tomography with tracer injection during seizures (ictal SPECT) is a promising tool for localizing seizures. However, much uncertainty exists about how to interpret late injections, or injections done after seizure end (postictal SPECT). A widely available and objective method is needed to interpret ambiguous ictal and postictal scans, with changes in multiple brain regions.

Methods: Ictal or postictal SPECT scans were performed by using [99mTc]-labeled hexamethyl-propylene-amine-oxime (HMPAO), and images were analyzed by comparison with interictal scans for each patient. Forty-seven cases of localized epilepsy were studied. We used methods that can be implemented anywhere, based on freely downloadable software and normal SPECT databases (http://spect.yale.edu). Statistical parametric mapping (SPM) was used to localize a single region of seizure onset based on ictal (or postictal) versus interictal difference images for each patient. We refer to this method as ictal–interictal SPECT analyzed by SPM (ISAS).

Results: With this approach, ictal SPECT identified a single unambiguous region of seizure onset in 71% of mesial temporal and 83% of neocortical epilepsy cases, even with late injections, and the localization was correct in all (100%) cases. Postictal SPECT, conversely, with injections performed soon after seizures, was very poor at localizing a single region based on either perfusion increases or decreases, often because changes were similar in multiple brain regions. However, measuring which hemisphere overall had more decreased perfusion with postictal SPECT, lateralized seizure onset to the correct side in ∼80% of cases.

Conclusions: ISAS provides a validated and readily available method for epilepsy SPECT analysis and interpretation. The results also emphasize the need to obtain SPECT injections during seizures to achieve unambiguous localization.

Epilepsy is a devastating illness with a major economic and psychosocial impact (1). Although medications are often beneficial, millions have epileptic seizures that are refractory to medical therapy (2,3). Hope for a cure can be offered to these patients through selective surgery to remove the region of seizure onset. However, for this surgery to be performed, the region of seizure onset must be identified with a high level of certainty. The gold standard for seizure localization has been invasive intracranial EEG recording (4,5). However, this procedure carries some operative risk (6,7) and has limited spatial sampling because of the finite number of electrodes that can be inserted safely. Recent advances in neuroimaging methods offer the possibility of localizing seizures safely, noninvasively, and in a manner that maps seizure activity throughout the entire brain (8,9).

Although functional brain abnormalities in the interictal state (between seizures) can provide clues about the region of seizure onset, it is crucial to map brain activity in the ictal state (during seizures) to know the site of seizure initiation (10–12). Most neuroimaging methods such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET) are impractical here because images would have been acquired during seizures while patients are actively moving and medically unstable. Single-photon emission computed tomography (SPECT) has the unique advantage of mapping brain activity at the time of radiotracer injection (during seizures), whereas the actual imaging can be done 60 to 90 min later when the patient is fully stable. Thus after injection, the SPECT agent is rapidly taken up by the brain, based on cerebral blood flow (CBF) and does not redistribute (13,14). Because CBF is closely linked to neuronal activity (15), the SPECT injection provides a “snapshot” of brain activity at the time of injection, which can be viewed later by neuroimaging.

Ictal SPECT imaging has shown great promise in localizing seizure onset (12,13,16,17). Recent advances in quantitative analysis have further improved the diagnostic yield of ictal SPECT, including digital registration and subtraction of baseline (interictal) images taken from the same patient at a time when seizures are quiescent (12,16,18,19). However, the lack of a validated and widely available method for analyzing and interpreting ictal and interictal SPECT images has limited the use of this diagnostic tool (20). Interpretation of ictal SPECT can be ambiguous and subjective because multiple areas are often activated and may be considered positive, depending on analysis threshold (21). Preliminary studies using statistical parametric mapping (SPM) offer the possibility of objective interpretation (18,20,22,23); however, validation and threshold determination are needed, and freely available normal SPECT databases are necessary for this method to be widely applied. In addition, although prior studies suggest that early SPECT injection timing is critical for improving diagnostic yield (10,24–26), this has not been rigorously studied. The lack of this information has made it difficult for most centers to justify the considerable investment in personnel and technical resources needed to attain consistent early SPECT injections. Finally, it has long been known that focal CBF decreases occur in the postictal period (immediately after seizures) (27–30); and regional CBF decreases have also been reported during seizures both at the epileptic focus (31) and in surrounding regions (18,32–35). Prior work has not thoroughly investigated the possible role of these CBF decreases in seizure localization.

The goal of our present study is to provide a reliable method that can be used at any center to analyze and interpret epilepsy SPECT images. To accomplish this, we have studied [99mTc]-labeled hexamethyl-propylene-amine-oxime (HMPAO) SPECT in a group of patients with surgically confirmed seizure localization. We have included patients with both mesial temporal lobe epilepsy and neocortical epilepsy, because pathophysiology and surgical outcome may differ in these two groups (36). We then used ictal–interictal SPECT analyzed by SPM (ISAS). ISAS is similar to difference imaging (19), or SISCOM (16), in using differences between ictal and interictal images for each patient. However, ISAS has the added benefits of using a standard normal reference to determine statistical significance of any changes and providing more objective methods of interpretation. We determined appropriate statistical thresholds to optimize sensitivity and specificity of the method. We then analyzed the localizing value of CBF increases and decreases in the ictal and postictal periods in this population. Finally, we provide all details of the methods along with downloadable scans from our healthy normal SPECT database, so that this approach will be readily available to others.

METHODS

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgments
  7. REFERENCES

We have posted additional details of the methods and downloads at the following URL (http://spect.yale.edu/), which should allow any center to replicate this analysis.

Patients

All procedures were in compliance with Yale Human Investigations Committee and National Institutes of Health (NIH) guidelines. Informed consent was obtained from all subjects. Inclusion and exclusion criteria were chosen to obtain a group of patients who had well-localized epilepsy and had undergone SPECT studies. Consecutive patients were chosen from 1994 through 2003 who met the following inclusion criteria: ictal or postictal SPECT performed during video and scalp EEG monitoring; interictal SPECT performed after a seizure-free period of ≥24 h; combination of tests [MRI, fluorodeoxyglucose (FDG)-PET, neuropsychology testing, angiogram Wada test, scalp EEG, and intracranial EEG] supporting localization of seizure onset. Mesial temporal lobe epilepsy patients all had successful surgery with no seizures during a minimum follow-up period of 1 year after anteromedial temporal lobe resection, and all had hippocampal sclerosis on pathology except one who had a mesial temporal ganglioglioma; neocortical epilepsy patients all had combination of tests (listed earlier) supporting localization of seizure onset, and all had successful surgery with no seizures during a minimum follow-up period of 1 year after focal resection, except for four patients who could not undergo focal resection because seizure onset was in primary sensorimotor or language cortex. Patients were excluded based on the following: unclear localization or localization in multiple brain regions; seizures that secondarily generalized; SPECT images acquired during intracranial EEG monitoring (37); patients with prior resective surgery or other major anatomic brain abnormalities; and poor scan quality or unknown SPECT injection timing. Secondarily generalized seizures were excluded because a preliminary analysis suggests poor SPECT localization in this group, likely due to seizure propagation (38). Patients with major anatomic brain abnormalities (i.e., prior surgery) were excluded because this complicates the spatial warping in SPM (although it is feasible to use the same analysis methods in this group). Based on these criteria, we included a total of 47 ictal–interictal scan pairs. These consisted of 28 ictal–interictal scan pairs from 26 patients diagnosed with right or left mesial temporal lobe epilepsy (17 female and nine male; mean age, 38 years; age range, 14–57 years) and 19 scan pairs from 16 patients with neocortical localization (11 female and five male; mean age, 30 years; age range, 9–60 years).

Healthy normal subjects

Fourteen healthy normal control subjects were obtained from a database of repeated SPECT scans acquired on different days, as described previously (18,39). Thus in the normal subjects, as in the patients, two scans were obtained for each individual. This was necessary, because our analysis looked at differences between ictal and interictal scans in each patient and then determined whether these differences were statistically significant in comparison to the normal variation seen from one scan to the next in healthy controls. Subjects were neurologically normal and were screened by using the Structured Clinical Interview for DSM-IV (SCID). Scan pairs from eight females and six males were included, 12 right-handed and two left-handed; mean age, 33 years; age range, 24–48 years. Downloadable scans for these normal subjects (28 scans total) are available in Analyze format at http://spect.yale.edu/, and will also be published in the Society for Nuclear Medicine database (http://brainscans.indd.org/bcnd_search_form.htm).

SPECT imaging

Ictal SPECT studies were performed during continuous scalp EEG and video monitoring. After noting seizure onset, a technologist injected patients intravenously with [99mTc]-labeled HMPAO (Medi-Physics; Amersham Healthcare, Arlington Heights, IL, U.S.A.), as soon as possible. Patients were asked to close their eyes during the injection. Interictal images were obtained in these same patients after ≥24 h of no seizure activity, in a quiet room, while patients were at rest, awake, but with their eyes closed. Healthy normal subjects were injected under these same conditions on 2 separate days.

SPECT images were acquired within 90 min after injection. Projection data were acquired for 40 min on a Picker IRIX (Philips Medical Systems, Best, Netherlands), and transverse slices were reconstructed as described previously (18,19). SPECT image data were transferred to personal computer and saved in Analyze format by using ImageJ (http://rsb.info.nih.gov/ij/).

Image analysis

SPECT ictal–interictal difference image pairs were compared with the repeated healthy normal scan pairs as described by Chang et al. (18). Differences between ictal and interictal SPECT images were calculated for each patient. Statistical parametric mapping (SPM) (40,41) was then used to calculate the significance of increases and decreases between interictal and ictal SPECT scans. This was done by using normal scan pairs to obtain an estimate of expected random variations from one scan to the next for each location in the brain. Major changes in the method since the work of Chang et al. (18) include a larger healthy normal database, improved SPM software to version SPM2 (Wellcome Department of Cognitive Neurology, London, U.K. http://www.fil.ion.ucl.ac.uk/spm/), and code modifications allowing semiautomated image entry, drastically reducing processing time (see http://spect.yale.edu/ for details and more objective methods for interpreting results, as described in the next section). We refer to the method used here as Ictal–interictal SPECT Analyzed by SPM (ISAS).

SPM analyses were run on a MATLAB 6.1 (The MathWorks, Inc., Natick, MA, U.S.A.) platform. Default SPM2 parameters for analysis of SPECT images were used, except where noted. More detailed information and downloads for implementing this analysis can be found at http://spect.yale.edu/. Reconstructed images, in Analyze format, were reoriented, and cross-hairs were set to the anterior commissure. Realignment was performed for each ictal–interictal scan pair and for each normal scan pair, and a mean image was created for each pair. Spatial normalization (warping) was performed by warping each mean image to the SPM SPECT template, and the same parameters were then applied to warp each individual scan to the template. Trilinear interpolation was used, and spatially normalized images were written with a bounding box equal to the template, with voxel dimensions of 2 × 2 × 2 mm. To remove extraneous signal from outside the brain, spatially normalized images were then masked with the SPM binary brain-mask image template (brainmask.img). A gaussian kernel of 16 × 16 × 16 mm was then used to smooth images before beginning the statistical analysis. Global Intensity normalization to correct for differences in total brain counts between scan pairs was performed by using proportional scaling with an analysis threshold of 0.8 (40,41). As a double check for problems in the spatial warping, smoothing, or masking steps, especially due to extraneous signal from outside the brain, we also analyzed images without these preprocessing steps by using Rview freeware (http://www.colin-studholme.net/software/software.html).

For statistical analysis, each patient's ictal–interictal difference scan pair was then compared with our database of repeated scans from the 14 healthy normal patients by using the multigroups conditions and covariates model. To save time and to reduce data-entry errors, we used an SPM code modification to enter automatically the preprocessed normal scans when setting up the model. For detailed instructions on code modifications and downloadable healthy normal scans, see http://spect.yale.edu/. Contrasts were set up to look at relative perfusion increases (ictal–interical) and decreases (interictal–ictal). SPM allows separate thresholds to be set for the minimal spatial extent and amplitude (height) of changes that will be included in the analysis. We have used an extent threshold (k) of 125 voxels (35,42), because at a voxel size of 2 × 2 × 2 mm, 125 voxels corresponds to 1 cc, which is approximately equivalent to spatial resolution of SPECT in tissue (intrinsic spatial and energy resolutions of our camera system were 4.5 mm and 18 keV, with a measured system spatial resolution of 9 mm for brain scans). The choice of height threshold was determined by receiver operating characteristic (ROC) analysis, as discussed later. Based on the ROC analysis, a height threshold (individual voxel-level significance) of p = 0.01 (Z-score, 2.33) was used for subsequent patient analyses.

SPM identifies clusters of voxels with changes that exceed these thresholds. We further considered only clusters of voxels with significance level of p < 0.05 (corrected cluster-level significance), as discussed in the next section. This procedure effectively accounts for family-wise error, correcting for multiple comparisons for the entire brain based on gaussian random field theory, as implemented by SPM (41).

Total time for analysis of a single case, with an ictal and interictal SPECT, including preprocessing steps, was 13 min per analysis, with a PC with Windows XP operating system, 3.0 GHz Pentium processor, and 1 GB RAM.

“Reading rules” for interpretation of ISAS

Criteria for interpreting ISAS results were established aimed at identifying a single positive region whenever possible. The interpretation consisted of first identifying the most significant cluster of contiguous voxels, referred to as a “positive cluster.” Next, the anatomic region (lobe) containing this cluster was identified and referred to as a “positive region.” The following criteria were used:

  • 1
    A cluster of voxels was considered positive if it was
    • a. 
      Significant at the cluster level (corrected significance level, p < 0.05)
    • b. 
      Most significant at the cluster level (lowest corrected p value of all clusters)
    • c. 
      Rules for ties: If the corrected cluster level significance, p, is <0.001 for more than one cluster, this can lead to apparent ties for the most significant cluster. This is because the SPM glass brain window for volume statistics rounds p values to 3 significant digits by default, yielding p values of “0.000.” The p values without rounding can be obtained by clicking on the “0.000” for each cluster. The p values without rounding will then appear in the Matlab command window and can be used to determine the most significant cluster. In the case of truly tied p values, then both clusters were considered positive.
  • 2
    A region (lobe) was considered positive if
    • a. 
      It contained the majority of the voxels from a positive cluster.
    • b. 
      Rules for ties: If a positive cluster involved two or more regions equally, then all involved regions were considered positive.

Determination of SPECT injection timing

Timing data were collected from retrospective video and EEG review by readers blinded to the imaging results, as described previously (35). Seizure-onset time was defined as the earliest, and seizure end as the latest EEG or clinical evidence of seizure activity. SPECT injection time was defined as the time when the syringe plunger was fully depressed. Ictal SPECT scans were defined as those injected between seizure onset and end, whereas postictal SPECT scans were defined as those injected after seizure end.

Receiver operating characteristic (ROC) analysis

ROC curves were estimated to determine the optimal p-value height threshold for ictal–interictal SPM SPECT analysis. Each of the cerebral hemispheres was divided into frontal, temporal, perirolandic, parietal, and occipital lobes for a total of 10 regions per patient, by using anatomic criteria as described previously (35). One region in each patient corresponded to the correct localization of seizure onset, based on surgical outcome and other clinical data as described earlier (Patients section). The height threshold for SPECT image analysis was varied from very stringent (p = 10−11) to very lax (p = 0.3) so that sensitivity and specificity could be determined at each height threshold. A true-positive was defined as a region that was both positive by SPECT and was the correct clinical localization of seizure onset (based on surgical outcome and other data as described earlier). A false-positive region was positive by SPECT but not correct clinically. A true-negative region was negative by SPECT and clinically. A false-negative region was negative by SPECT but was the clinical region of seizure onset. Sensitivity and specificity were determined at each height threshold across patients and regions (470 regions total), as has been previously described (21). It should be noted that when the same analysis was applied to each of the healthy normal scan pairs, this yielded no significant clusters.

ROC curves were generated with ROCKIT, version 0.9b (Charles E. Metz, University of Chicago, Chicago, IL, U.S.A.) for Windows, by using a maximum-likelihood estimate of the data. A z-score test was then performed based on differences between the areas under different ROC curves (43). In this manner, a probability value was computed to indicate the statistical significance of the difference in area under the curves generated for true ictal injections, as compared with postictal injections in both the mesial temporal and neocortical groups.

Hypoperfusion asymmetry index

Because perfusion increases were not always useful for detecting the region of seizure onset, particularly in postictal studies, we also evaluated the clinical utility of perfusion decreases. Cluster localizations for perfusion decreases were determined as already described for perfusion increases. In addition, we calculated a hypoperfusion asymmetry index to determine which hemisphere had more extensive CBF decreases. The hypoperfusion asymmetry index was calculated based on the volume of significant voxels (at a voxel-level significance threshold p = 0.01, and cluster-level significance k = 125, as before) showing decreased perfusion in each hemisphere. Thus we took the number of voxels in the left hemisphere minus number of voxels in the right hemisphere divided by the total number of hypoperfusion voxels [(kleftkright) / (kleft+kright)]. Binary images corresponding to each hemisphere were created in MRIcro (for detailed methods and downloadable binary masks, see http://spect.yale.edu/) and were used to determine kleft and kright by using the small volume-correction function in SPM.

RESULTS

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgments
  7. REFERENCES

We analyzed SPECT scan pairs in four different groups based on injection timing and seizure localization: mesial temporal ictal and postictal; and neocortical ictal and postictal (Tables 1 and 2). Although injections were done as soon as possible after noting seizure onset, many patients were injected after the seizure had ended. True-ictal SPECT injections were obtained in seven of the mesial temporal lobe cases, and 21 were injected postictally. In patients with neocortical onset, six were injected during ictus, and 13 postictal injections were obtained.

Table 1. Localization of CBF increases in mesial temporal seizures
Group Pt no. a Localizationb Seizure duration (s)SPECT injection time (s)cLocation of most significant clusterdCluster significancee (p) Cluster volumef (k)Location of most significant voxelgCoordinates of most significant voxelh
  1. Clusters localized to the correct lobe are shown in bold.

  2. R, right; L, left; CBF, cerebral blood flow.

  3. aPatients are listed in order of SPECT injection time.

  4. bLocalization based on surgery and other clinical data (see Methods).

  5. cSPECT injection times are relative to seizure end.

  6. dLocation is lobe where the majority of voxels are located (see Methods).

  7. ep, cluster level significance corrected for multiple comparisons for the entire brain.

  8. fk, cluster size in voxels (voxel size, 2 ×2 × 2 mm).

  9. gLocation of the most significant voxel within the most significant cluster.

  10. hx, y, z coordinates are in MNI space.

Ictal 1L temporal152 −84 L temporal<0.00122,717L medial occipital−14 −78 0
 2R temporal153 −69 R temporal<0.00110,429R posterior lateral temporal66 −28 −4
 3R temporal104 −25 R temporal<0.00128,444R middle temporal gyrus38 4 −28
 4R temporal 95 −23 R temporal<0.00117,538R frontal (white matter)28 34 12
 5R temporal111 −18 R temporal<0.00131,298R lateral temporal64 −26 −10
 6L temporal 79 −16 L temporal and L occipital<0.00119,246R lateral occipital42 −82 12
 7L temporal100 −9L temporal, L parietal, bilateral occipital<0.00122,799L lateral parietal−50 −50 22
Postictal 8L temporal 48  8R occipital and R temporal<0.001 5,888R occiptotemporal50 −52 −10
 9L temporal 68 12R parietal 0.001 3,031R medial parietal (precuneus)8 −66 42
10L temporal 61 15Bilateral occipital<0.00115,664L medial occipital−12 −88 20
11R temporal103 20R temporal<0.00126,322R lateral temporooccipital48 −52 −10
12R temporal 62 24R temporal<0.00113,896R lateral temporal66 −28 −6
13R temporal 79 26L frontal<0.001 4,423L frontal (white matter)−24 34 22
14R temporal 45 29R temporal<0.001 4,107R medial temporal26 −16 −8
15L temporal120 30Bilateral occipital<0.00110,608Left medial occipital−2 −84 −4
16R temporal 36 33No significant clusters 0.952   262R lateral occipital40 −82 12
17R temporal123 35R temporal<0.001 6,486R medial temporal24 −12 −22
18R temporal 46 35L frontal<0.0018,975L middle frontal gyrus−26 −16 62
19L temporal 55 36Bilateral precentral<0.001 5,900R precentral gyrus20 −20 54
20R temporal 79 39L parietal<0.00116,863L precentral gyrus−28 −20 62
21R temporal 50 42R temporal<0.001 4,555R medial temporal18 −2 −28
22L temporal118 44Bilateral occipital<0.00121,203R medial occipital6 −86 18
23L temporal 50 48Bilateral occipital<0.001 4,837Midline occipital (cuneus)6 −88 18
24R temporal 55 52L parietal<0.00125,756L medial parietooccipital−10 −80 48
25L temporal 26 53L temporoparietal 0.001 2,294L temporoparietal−36 −44 20
26R temporal125 55Bilateral occipital<0.00131,331L medial occipital−14 −78 2
27L temporal 36 72R occipital and L temporal<0.00136,732R lateral temporooccipital44 −54 −8
28L temporal 66118Bilateral occipital 0.013 2,009L medial occipital−8 −86 24
Table 2. Localization of CBF increases in neocortical seizures
Group Pt no. a Localizationb Seizure duration (s)SPECT injection time (s) cLocation of most significant clusterdCluster significance (p) e Cluster volume (k) fLocation of most significant voxelgCoordinates of most significant voxelh
  1. Clusters localized to the correct lobe are shown in bold.

  2. R, right; L, left; CBF, cerebral blood flow.

  3. aPatients are listed in order of SPECT injection time.

  4. bLocalization based on surgery and other clinical data (see Methods).

  5. cSPECT injection times are relative to seizure end.

  6. dLocation is lobe where the majority of voxels are located (see Methods).

  7. ep, cluster level significance corrected for multiple comparisons for the entire brain.

  8. fk, cluster size in voxels (voxel size, 2 × 2 × 2 mm).

  9. gLocation of the most significant voxel within the most significant cluster.

  10. hx, y, z coordinates are in MNI space.

Ictal29R temporal neocortex168−149R temporal<0.00119,963R lateral temporal50 2 −12
30L temporal neocortex252−141L temporal<0.00130,233L lateral parietotemporal−52 −40 −12
31R superior temp/inferior parietal138 −87No significant clusters 0.117 1,013R superior temp/inferior parietal46 −32 18
32L rolandic 43 −21L rolandic<0.001 2,961L precentral gyrus−26 −18 62
33R superior frontal 21  −6R superior frontal<0.001 4,203R medial superior frontal−2 4 58
34R frontal 31  −5R frontal<0.001 7,726L frontal polar−26 38 22
Postictal35R temporal neocortex 85   7R temporal<0.001 8,262R lateral temporal50 2 −14
36L temporal neocortex 74   8No significant clusters0.272  850Midline frontal polar2 58 6
37R superior frontal 74  13R frontal<0.001 3,858R superior frontal18 −16 48
38L temporal neocortex 59  14R occipital<0.00112,878R occipital pole6 −86 16
39R rolandic  5  19L frontal<0.001 4,815L frontal polar−16 44 26
40R rolandic  5  20L frontal<0.001 4,055L frontal polar−14 40 15
41R frontal 66  20R temporal<0.00119,779R lateral temporal64 −26 −8
42R temporal neocortex 34  39L frontal<0.001 6,747L frontal polar−22 38 20
43R frontal 36  43Bilateral occipital<0.00114,408L medial occipital−6 −84 32
44R temporal neocortex 75  45L occipital<0.001 3,927L occipital−14 −78 0
45L rolandic 32  60Bilateral parietal and bilateral occipital<0.00110,996Midline Parietoccipital sulcus0 −74 40
46L frontal 11  60Bilateral frontal and bilateral rolandic<0.00117,907L prefrontal−20 30 38
47L temporal neocortex 31 166Bilateral occipital<0.00126,769R occipital24 −78 12

Effects of injection timing and analysis threshold

Ictal–interictal SPECT analysis by SPM (ISAS) was performed by calculating the difference between ictal minus interictal SPECT for each patient, and assessing the statistical significance of these differences using our database of repeated normal scans. We sought (a) to determine the optimal threshold for the SPM analysis and (b) to investigate the effects of injection time on the localizing value of this analysis method. ROC analysis was done to address both of these questions. Fig. 1 illustrates the ROC curves for ictal and postictal studies in both mesial temporal and neocortical epilepsy. These curves demonstrate far better sensitivity and specificity for ictal compared with postictal SPECT. This is true in both mesial temporal and neocortical epilepsy over a range of height thresholds. Comparison of the areas under the curves confirms that ictal injections are significantly more accurate at predicting the region of seizure onset than are postictal injections (p = 0.0001 for mesial temporal cases, and p = 0.027 for neocortical cases).

image

Figure 1. Receiver operating characteristic (ROC) curves for ictal–interictal SPECT analyzed by SPM (ISAS). A: Mesial temporal epilepsy (n = 28). B: Neocortical epilepsy (n = 19). Curves represent sensitivity and specificity of regional SPECT increases (hyperperfusion vs. interictal SPECT) in predicting seizure focus at varying SPM height threshold. Ictal SPECT is significantly more sensitive and specific than postictal SPECT for both (A) mesial temporal (p = 0.0001) and (B) neocortical (p = 0.027) epilepsy. Arrows, The closest point on the curves corresponding to an SPM height threshold of p = 0.01. Moving along curves to the right corresponds to less stringent SPM thresholds (higher sensitivity), whereas moving to the left corresponds to more stringent thresholds (higher specificity). A threshold value of p = 0.01 (arrows) is in a range on the ROC curves that optimizes both sensitivity and specificity (43,44). Curves were generated from results of regional SPECT analysis by using ROCKIT, as described in the Methods.

Download figure to PowerPoint

The ROC analysis also demonstrated that an appropriate threshold at which to operate on the curve is p = 0.01. As shown in Fig. 1 (arrows), operating at a threshold of p = 0.01 provides a relatively low false-positive fraction with a high true-positive fraction in both mesial temporal and neocortical patients and is in a range on the ROC curves that optimizes both sensitivity and specificity (43,44). Thus, for ictal injections, we found a sensitivity of 93% and specificity 87% for mesial temporal epilepsy, and a sensitivity of 77% and specificity of 93% for neocortical epilepsy.

Localizing value of ictal SPECT

The goal of ictal SPECT is to obtain a single unambiguous localization for surgical resection. It is important to note that the ROC analysis will consider a particular region to be true positive even when another region in the same scan is false positive. We therefore sought to validate the threshold selection and ISAS method in general under more stringent and clinically useful conditions for localization. To do this, we defined an analysis as correctly localizing if the positive cluster (as defined in Methods) for the hyperperfusion contrast was localized to the correct lobe, and only to the correct lobe. In other words, if a cluster of voxels with significant perfusion changes was distributed equally over more than one region, then it was considered wrong because it would not provide an unambiguous localization. Figure 2 shows the results of this analysis, with the percentage of unambiguous correct localizations over a range of p-value thresholds. Based on this analysis, p = 0.01 maximizes the chances of getting a correct, unambiguous localization and is, therefore, an appropriate choice for the height threshold, in agreement with the ROC analysis. In addition, Fig. 2 provides further evidence that ictal SPECT is far superior to postictal SPECT for localizing seizure onset based on regional hyperperfusion; this is true for both mesial temporal and neocortical epilepsy.

image

Figure 2. Percentage unambiguous correct localization based on single-photon emission computed tomography hyperperfusion. A: Mesial temporal epilepsy (n = 28). B: Neocortical epilepsy (n = 19). Scan results were considered to be correct if the region of clinical onset and only the region of clinical onset was found to be positive, based on the location of a significant positive cluster. Correctly localizing scans as a percentage of total scans are indicated at varying Statistical parametric mapping height thresholds, P.

Download figure to PowerPoint

Having established the analysis threshold (p = 0.01), we now discuss the results of ISAS for localizing seizures in our patient group (see Tables 1 and 2). SPECT hyperperfusion correctly and unambiguously localized the region of seizure onset in five (71%) of seven mesial temporal patients and five (83%) of six neocortical patients with true ictal injections. In the two temporal lobe patients that were considered incorrect (patients 6 and 7; Table 1), the localization included the correct lobe; however, adjacent regions were considered positive as well, making the localization ambiguous. For the one true-ictal neocortical patient where hyperperfusion was incorrect (patient 31, Table 2), the most significant voxel was located in the correct lobe; however, the cluster was not significant (p < 0.05) at the cluster level after correction for multiple comparisons. Note that in all cases in which ISAS gave a single unambiguous localization based on ictal SPECT, the localization was correct (100%).

Figure 3A shows typical results from a mesial temporal patient injected ictally. This patient (patient 2), had significant perfusion increases localized to the right temporal lobe, consistent with the region of seizure onset. Similarly, results typical for neocortical patients injected during the seizure are illustrated in Fig. 4A. This patient (patient 32), was diagnosed with epilepsy originating from left peri-rolandic cortex, and consistent with this, CBF increases were localized to that region.

image

Figure 3. Ictal and postictal single-photon emission computed tomography (SPECT) in mesial temporal lobe epilepsy analyzed by statistical parametric mapping (SPM). A: Ictal SPECT from patient with right mesial temporal lobe onset (patient 2). Seizure duration was 153 s, and [99mTc]- hexamethyl-propylene-amine-oxime (HMPAO) was injected 69 s before seizure end. The most significant hyperperfusion cluster was localized to the right temporal lobe (cluster-level significance, p < 0.001, corrected for multiple comparisons; z-score of most significant voxel, 5.06; cluster size, k = 10,429 voxels). The most significant region of hypoperfusion was localized to the left occipital lobe (p = 0.035; Z = 4.04; k = 1,625). B: Postictal SPECT from patient with left temporal lobe onset (patient 8). Seizure duration, 48 s; injection time, 8 s after seizure end. Hyperperfusion was localized to right occipital and right temporal lobes (p < 0.001; z = 4.72; k = 5,888). The most significant decreases were localized to left frontal and parietal lobes (p < 0.001; Z = 6.13; k = 18,433). For A and B, each SPECT scan was background subtracted by using the patient's corresponding interictal SPECT, and the difference was then compared with the database of normal SPECT pairs by using SPM (see Methods). Cerebral blood flow increases are shown as red-white, and decreases are shown as blue-green; color bars indicate t values. Coronal views are shown superimposed on the SPM magnetic resonance imaging template. Extent threshold, k = 125 voxels; voxel-level height threshold, p = 0.01.

Download figure to PowerPoint

image

Figure 4. Ictal and postictal single-photon emission computed tomography (SPECT) in neocortical epilepsy analyzed by statistical parametric mapping (SPM). A: Ictal SPECT from patient with left rolandic onset (patient 32). Seizure duration was 43 s, and injection time was 21 s before seizure end. The most significant hyperperfusion cluster was localized to the left rolandic cortex (cluster-level significance, p < 0.0001, corrected for multiple comparisons; z-score of most significant voxel = 5.98; cluster size, k = 13,872 voxels). The most significant hypoperfusion was localized to right occipital lobe (p < 0.001; z = 5.44; k = 14,228). B: Postictal SPECT from patient with left lateral temporal neocortical onset (patient 38). Seizure duration, 59 s; injection time, 14 s after seizure end. Hyperperfusion was localized to right occipital lobe (p < 0.0001; z = 5.14; k = 12,878). The most significant decreases were localized to left frontal and temporal lobes (p < 0.001; z = 6.74; k = 18,928). For A and B, each SPECT scan was background subtracted by using the patient's corresponding interictal SPECT, and the difference was then compared with the database of normal SPECT pairs by using SPM (see Methods). Cerebral blood flow increases are shown as red-white, and decreases are shown as blue-green; color bars indicate t values. Coronal views are shown superimposed on the SPM magnetic resonance imaging template. Extent threshold, k = 125 voxels; voxel-level height threshold, p = 0.01.

Download figure to PowerPoint

Although this method of analysis was highly accurate with true-ictal injections, the accuracy rate was much lower for patients who were injected with the SPECT agent after the seizure had ended. For mesial temporal patients, only five (24%) of 21, and for neocortical patients, only two (15%) of 13 were correctly localized by using hyperperfusion (see Tables 1 and 2). Of note, the patients that did have correct localizations tended to be injected in the early postictal period (within ∼40 s after seizure end), rather than later (see Tables 1 and 2), although even in this early postictal period, fewer than half of the localizations were correct. Even if only patients with a single unambiguous localization were considered, the localization remained poor at five of 11 mesial temporal and two of eight neocortical patients injected postictally (see Tables 1 and 2).

Although patients injected postictally usually had incorrect localization, we wondered whether the regions of hyperperfusion at these late times could, at least, help lateralize seizure onset to the correct hemisphere. However, in patients who were injected postictally and had incorrect localization, all temporal lobe patients had perfusion increases that were bilateral or in the contralateral hemisphere (see Table 1), and for extratemporal patients, only one (patient 41) had increases confined to the hemisphere of seizure onset (see Table 2). Thus with postictal SPECT injections, CBF increases were not generally useful in either localizing or lateralizing the epileptogenic region.

Could localization in the ictal or postictal periods be improved by using different “reading rules” (see Methods) for interpreting SPECT hyperperfusion? We repeated the interpretation of the SPECT images by using several variant methods and found that the criteria used here work best. For example, if we performed the same analysis but then localized the seizure focus by the most significant voxel instead of by the location of the most significant cluster, we obtained far worse results. Thus the location of the most significant hyperperfusion voxel (see Tables 1 and 2) correctly localized only three (43%) of seven true-ictal mesial temporal patients; four (67%) of six true-ictal neocortical patients; four (19%) of 21 postictal mesial temporal patients; and three (23%) of 13 postictal neocortical patients. Similarly, use of different rules to decide between tied clusters or regions (see Methods) did not improve the localization.

Lateralizing value of postictal SPECT

Because perfusion increases were not helpful in predicting the region of seizure onset in patients with late injections, we also investigated the localizing value of CBF decreases. By using the same rules for cluster localization as for hyperperfusion, and height threshold p = 0.01, hypoperfusion correctly and unambiguously localized the correct lobe in only two (9.5%) of 21 mesial temporal patients injected postictally (Table 3) and one (7.7%) of 13 neocortical patients (Table 4). In some patients [five (24%) of 21] for mesial temporal and [seven (54%) of 13] for neocortical, the hypoperfusion was sensitive for detecting the region of seizure onset, but not specific (the localization included the correct lobe, but was not limited to that region). Use of a more stringent height threshold did not greatly improve on the localizing value of CBF decreases. Similarly, use of the most significant voxel instead of the most significant cluster did not improve localization, yielding the correct localization in only eight (38%) of 21 postictal mesial temporal cases and only five (38%) of 13 neocortical cases. Thus in patients injected after seizure end, CBF decreases did not provide reliable information as to the lobe of seizure onset.

Table 3. Localization and asymmetry index of CBF decreases in mesial temporal lobe seizures
GroupPt. no. aLocalizationbLocation of most significant clustercVoxels leftd (k)Voxels right (k)Asymmetry Indexe
  1. Clusters localized unambiguously to the correct lobe are shown in bold (patients 13 and 28). In addition, asymmetry indices lateralizing to the correct hemisphere are shown in bold.

  2. R, right; L, left; CBF, cerebral blood flow.

  3. aPatients are listed in order of SPECT injection time.

  4. bLocalization based on surgery and other clinical data (see Methods).

  5. cLobe where the majority of voxels are located.

  6. dTotal number of significant voxels in each hemisphere (voxel size, 2 × 2 × 2 mm).

  7. eHypoperfusion asymmetry index =[(kleftkright) / (kleft+kright)].

Ictal1L temporalL frontal10,219 6,666  0.21
2R temporalL occipital 5,747 1,400  0.61
3R temporalBilateral frontal and L temporal21,78416,514  0.14
4R temporalL frontal and L occipital 7,330 1,526  0.66
5R temporalBilateral frontal20,97620,626  0.01
6L temporalL frontal20,990 4,690  0.63
7L temporalL frontal28,45913,045  0.37
Postictal8L temporalL frontal and L parietal17,622   758  0.92
9L temporalL frontal 5,772 3,006  0.32
10L temporalL lateral temporal and occipital12,53913,247−0.03
11R temporalR frontal 8,31525,543−0.51
12R temporalR occipital 3,19215,114−0.65
13R temporalR temporal 1,912 4,989−0.45
14R temporalNo significant clusters 2,051 1,806  0.06
15L temporalL lateral temporal and L occipital13,616 2,444  0.70
16R temporalNo significant clusters    52 1,050−0.91
17R temporalBilateral frontal 7,698 2,968  0.44
18R temporalR parietal 2,39219,744−0.78
19L temporalL frontal17,056 5,992  0.48
20R temporalR parietal and R frontal 1,10432,711−0.93
21R temporalR occipital 1,037 8,047−0.77
22L temporalBilateral frontal and L temporal20,66315,670  0.14
23L temporalL parietal16,328 1,835  0.80
24R temporalR frontal and R temporal 4,95336,021−0.76
25L temporalR frontal 1,136 4,1730.57
26R temporalR temporal, R parietal, and R frontal10,47426,903−0.44
27L temporalL parietal and L frontal30,645 7,886  0.59
28L temporalL temporal 8,965 2,736  0.53
Table 4. Localization and asymmetry index of CBF decreases in neocortical seizures
Group Pt no. a LocalizationbLocation of most significant clustercVoxels leftd (k)Voxels right (k)Asymmetry indexe
  1. Clusters localized unambiguously to the correct lobe are shown in bold (patient 37). In addition, asymmetry indices lateralizing to the correct hemisphere are shown in bold.

  2. R, right; L, left; CBF, cerebral blood flow.

  3. aPatients are listed in order of SPECT injection time.

  4. bLocalization based on surgery and other clinical data (see Methods).

  5. cLobe where the majority of voxels are located.

  6. dTotal number of significant voxels in each hemisphere (voxel size, 2 × 2 × 2 mm).

  7. eHypoperfusion asymmetry index =[(kleftkright) /(kleft+kright)].

True ictal29R temporal neocortexR parietal 4,04810,444−0.44
30L temporal neocortexBilateral occipital13,27013,315  0.00
31R superior temporoinferior parietalR temporal 2,219 3,457−0.22
32L rolandicR occipital 8,546 9,7290.06
33R superior frontalL parietal   883 1,404−0.23
34R frontalR parietal 1,726 4,797−0.47
Postictal35R temporal neocortexR Frontal and R parietal 1,55322,734−0.87
36L temporal neocortexL temporal and L occipital 1,853 2,5210.15
37R superior frontalR frontal 3,703 6,352−0.26
38L temporal neocortexL temporal and L frontal16,805 4,061  0.61
39R rolandicR occipital 2,583 6,096−0.40
40R rolandicNo significant clusters   494 1,428−0.49
41R frontalR frontal and R parietal12,69717,314−0.15
42R temporal neocortexR occipital and R temporal   12319,271−0.99
43R frontalBilateral frontal 9,41813,239−0.17
44R temporal neocortexR temporal and R frontal 5,03224,024−0.65
45L rolandicBilateral frontal10,77112,2620.06
46L frontalR frontal and R occipital10,03512,737−0.12
47L temporal neocortexL frontal, L temporal and L parietal19,059 7,970  0.41

Although hypoperfusion of a specific lobe was not helpful for localization, we noticed that the hemisphere of seizure onset tended to exhibit more regions of hypoperfusion than the contralateral hemisphere (see Tables 3 and 4). These regions often extended beyond the lobe of seizure onset but tended to be greater on the side of seizure onset, especially in the postictal period. Therefore we calculated a hypoperfusion asymmetry index in the hope that postictal scans, which could not correctly identify the lobe of seizure onset, could at least identify the correct hemisphere. The hypoperfusion asymmetry index correctly lateralized the hemisphere of onset in 17 (81%) of 21 postictal mesial temporal patients and 10 (77%) of 13 patients with neocortical localization, injected postictally (see Tables 3 and 4). In reviewing these results, it is apparent that very small values for the hypoperfusion asymmetry index may tend to give the incorrect lateralization (e.g., patients 10, 14, and 45). As a rule of thumb, it can be seen in these data, for example, that if only hypoperfusion asymmetry with a magnitude >0.12 is considered, then the index correctly lateralizes seizures in 17 (89%) of 19 mesial temporal postictal injections, and 10 (91%) of 11 neocortical postictal injections.

Figures 3B and 4B illustrate results typical of late injections obtained in mesial temporal and neocortical patients, respectively. In Fig. 3B, a postictal injection after a left temporal seizure (patient 8) exhibits false localization of CBF increases to the right temporal lobe. Conversely, CBF decreases in this case were lateralized to the correct (left) hemisphere but were not confined to the correct lobe. Figure 4B shows the results of a postictal injection after a neocortical seizure originating from the left lateral temporal region (patient 38). In this case, CBF increases are falsely localizing to the right occipital lobe. However, CBF decreases are, again, lateralized to the correct hemisphere but not localized to the correct lobe.

DISCUSSION

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgments
  7. REFERENCES

The goal of SPECT studies in epilepsy patients is to localize a single region of seizure onset objectively and unambiguously for epilepsy surgery. The results reported here emphasize the importance of true-ictal SPECT for achieving this goal. In addition, whereas we found that postictal SPECT does not unambiguously localize the lobe of seizure onset, postictal perfusion decreases can be useful for lateralizing the correct hemisphere. These results were obtained by using ISAS, an objective analysis method based on SPM, which we have validated and made available for public use along with our healthy normal SPECT database (http://spect.yale.edu/).

Our results demonstrate that ictal SPECT is far superior to postictal SPECT for seizure localization. These results are in agreement with prior work suggesting that ictal SPECT is superior (10,17,24–26), but our results are more conclusive and statistically rigorous. We report that CBF increases on ictal SPECT provided correct unambiguous localization in 70–80% of cases, whereas this was true in <10% of postictal injections. The clear superiority of ictal over postictal scans also was supported by ROC analysis in both mesial temporal and extratemporal epilepsy. We found that some clinically useful information can be salvaged from postictal SPECT scans, because postictal CBF decreases can at least lateralize seizure onset to the correct hemisphere in ∼80% of postictal scans. This hypoperfusion in the hemisphere of seizure onset, particularly in the postictal period, was in agreement with prior studies (18,32–35). Thus whereas hypoperfusion was not good for unambiguously localizing the lobe of seizure onset, it was quite successful at lateralizing the afflicted hemisphere by using the hypoperfusion asymmetry index. Postictal scans may be helpful in cases in which other localizing information is scarce and can help tailor the placement of intracranial EEG electrodes to a more limited region, but they do not meet the goal of localizing seizures sufficiently to prevent the need for intracranial monitoring. These results demonstrate the critical need for true ictal SPECT injections to localize seizure onset. Ictal SPECT scans can be attained anywhere—it is simply a matter of personnel and logistics to inject patients immediately on seizure onset and then to acquire SPECT images in a timely manner. We hope that these results will help justify the investment of crucial resources needed at all medical centers performing SPECT imaging, so that patients can receive studies that are diagnostically useful.

This study also demonstrates that ISAS is a relatively easy, quick, and objective method for analyzing ictal–interical SPECT differences. In a recent review by Knowlton et al. (20), the advantages of SPM analysis for providing objective interpretation of epilepsy SPECT results were discussed, and the authors identified the need for these methods to be widely available, not just in specialized centers. We hope that this work will begin to fill that need by providing a freely available method that can be implemented anywhere. All that is required is a computer running MATLAB and an operator with sufficient imaging experience to download and implement the SPM analysis (see http://spect.yale.edu/). As in difference imaging (19) or SISCOM (16), ISAS is based on calculating the difference between ictal and interictal SPECT for each patient. Added advantages of ISAS are (a) the statistical significance of these differences is calculated by comparison with the normal variation from one scan to the next in healthy normal subjects, and (b) more-objective methods are used for interpretation of the results. We have validated ISAS by using our population of mesial temporal and neocortical epilepsy patients and our database of healthy normal SPECT pairs. Ideally, each center would establish its own database of normal SPECT pairs acquired under the same conditions used for their clinical scans, although in many cases, this may not be feasible because of the cost involved. Similarly, although the database of 14 scan pairs is a reasonable sample size for SPM (41,45), it is possible that an even larger database would improve our results. At least as a starting point, we have made our database of normal SPECT pairs available, so that other centers can independently test and validate the method with their own patient populations. Because the purpose of the database is mainly to provide an estimate of the variance in SPECT signal for each voxel in the brain between pairs of scans done on different days, it may be possible to use the database at other centers as long as the scan-to-scan variability is similar to ours. It should be noted that our SPECT scans were performed with [99mTc]-HPMAO (Ceratech, Medi-Physics, Amersham Healthcare), which has slightly different gray matter distribution from [99mTc]-ECD (Neurolite; Bristol-Myers Squibb Medical Imaging, N. Billerica, MA, U.S.A.) (46). We have not tested this method by using [99mTc]-ECD SPECT scans.

In the past, most epilepsy SPECT images were simply analyzed by visual comparison of ictal and interictal images. This method has many drawbacks, including the lack of intensity normalization, the difficulties in making comparisons between corresponding slices, and the lack of quantitative assessment (19). Subsequent improvements over simple visual interpretation have included ROI analysis (47) and ictal–interictal subtraction images coregistered to MRI (16,19). Recently, voxel-based statistical analysis using SPM has become a widely used method in analyzing functional neuroimaging studies. This method of analysis for SPECT data has been shown to be a useful diagnostic tool in localizing partial epilepsy (18,20,23).

In our investigation, we attempted to discern the appropriate parameters for analyzing ictal–interical SPECT images with SPM. We found that, although this method provides objective data on the statistical significance of SPECT changes, the interpretation of results may not be completely straightforward. One reason is that often in a single analysis, multiple clusters may be significant at the cluster level, even when using significance values that are corrected for multiple comparisons. Thus it was necessary to establish “reading rules” for determining which clusters to consider positive and how to determine the localization of positive clusters. The goal of noninvasive epilepsy localization is to identify a single region for surgical resection. Therefore criteria for interpreting SPM SPECT results were established here, aimed at identifying a single positive region whenever possible. The interpretation consisted of first identifying the most significant cluster of contiguous voxels, referred to as a “positive cluster.” Next, the region (lobe) identified by this cluster was identified and referred to as a “positive region.” This approach allowed us to use the statistically robust approach of SPM to identify significant clusters of voxels, correcting for multiple comparisons. We then determined the location of the cluster based on the lobe in which majority of the voxels in the cluster were located. Of note, we did not base the location on the most significant voxel. This was because we often found that the single most significant voxel was in the surgically incorrect lobe, whereas the majority of the voxels from the most significant cluster were in the correct lobe. These finding support the concept of an epileptogenic region, rather than a single epileptogenic focus (4,48).

Our analysis method differs from that of Lee et al. (23) in that we used the difference between an ictal and interictal scan for each patient, whereas they compared a single ictal scan with a database. Prior work has shown that omission of the interictal SPECT can lead to false localization (12,19). For example, increased CBF in the epileptogenic region may not show up in the face of baseline decreased CBF in this region. This can lead to pseudonormalization of CBF during seizures, which may not be detected without the interictal scan. Although omission of the interictal SPECT may still allow CBF increases to be seen in more obvious cases, we suspect that more subtle cases, especially with neocortical onset, may be difficult to localize correctly without the interictal comparison. Nevertheless, because omission of the interictal SPECT would lead to a considerable reduction in cost and time, a direct comparison of ISAS and the method of Lee et al. (23) should be pursued further. Additional direct comparisons of ISAS with difference imaging, as performed by Chang et al. (18), also would be worthwhile.

Some technical limitations and potential pitfalls of ISAS will require further studies in ongoing work. We excluded patients in the current study with significant anatomic abnormalities, prior resection, or intracranial hardware, because SPECT images are spatially warped to a template for analysis. However, it is feasible by using SPM to remove selective abnormal brain regions during spatial warping (49), so it is possible to analyze patients with anatomic abnormalities by ISAS as well (see http://spect.yale.edu for details). Although ISAS provides a straightforward and objective method for analysis, appropriate caution is required not to use a “black box” approach. It is essential to view the images at each stage of analysis and to assess image quality and registration visually and whether the results “make sense.” We have, therefore, automated the most straightforward processing steps (see http://spect.yale.edu/) to speed analysis time but have so far avoided further automation to encourage viewing results at intermediate stages of analysis and to allow greater user flexibility. For example, because anatomic warping is necessary for SPM analysis compared with a normal database, data should be visually inspected during processing for potential artifacts. In addition, results should be confirmed independently by methods that do not require spatial warping (16,19), which can also be achieved by using simple SPM tools (ImCalc), or other freeware (http://www.colin-studholme.net/software/software.html; see also http://spect.yale.edu). Although these other methods (without warping) do not allow objective statistical comparison with a standard database, they can be useful as a second check of results. Another enhancement that can be implemented in a relatively straightforward manner is unwarping of ISAS results back onto individual patient MRI scans rather than onto the MRI template (deformations toolbox in SPM2, http://www.fil.ion.ucl.ac.uk/spm/ see also http://spect.yale.edu). In this study, we also excluded patients with secondarily generalized seizures. A preliminary analysis demonstrated poor localization by SPECT in patients with secondarily generalized seizures (38); however, further investigation of this population is warranted.

One of the limitations of the current study is sample size. After applying our inclusion/exclusion criteria designed to study a population of patients with clear and definite localizations, only 47 image pairs were included, although our center has a relatively large volume of epilepsy patients. Further breaking down this group by localization, seizure duration, and injection time will result in small numbers within each subgroup. Important additional information may be gained by more detailed analysis of the time course of SPECT imaging changes. For example, from our data, we can speculate that temporal lobe seizures may propagate in the late ictal period, making localization of a single focus more difficult (e.g., patients 6 and 7 in Table 1). Previous studies also support this possibility (50), but a much larger sample size would be needed to investigate this phenomenon in detail. It also is known that the radiotracer can take between 30 and 60 s to be completely taken up in the brain (13,14); thus the injections done during the late ictal period may actually reflect postictal brain activity. On the other hand, the known delay between changes in neuronal activity and changes in CBF during seizures (51) may, at least in part, counteract this timing discrepancy. It is clear that to study fully the time course of CBF changes during and after seizures, a multicenter collaboration will be necessary.

In summary, ictal SPECT is a powerful noninvasive tool for presurgical localization. Our hope is to make a valid method of analysis and interpretation widely available, so that ictal SPECT will not be limited to a few specialized centers. Based on our results, obtaining true ictal injections remains a high priority. Although it is clear that ictal SPECT is far superior to postictal SPECT for presurgical localization, obtaining true ictal injections remains a challenge at most medical centers, mainly for logistic and financial reasons. A coordinated effort will most likely be needed among medical centers to guarantee that consistent ictal injections can be achieved and justified as medically necessary.

Acknowledgments

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgments
  7. REFERENCES

Acknowledgment:  We thank Drs. Thomas Kosten, Karen Tucker, and Christopher Gottschalk for providing the healthy normal SPECT scans from their prior work and for allowing us to publish the scans with this study. We also thank Jennifer Bonito and Judith Hess for management of the epilepsy clinical database at our center. This work was supported by an Epilepsy Foundation of America fellowship (to A.L.P.) and by the Betsy and Jonathan Blattmachr family. We also acknowledge the resources of the informatics core of the Yale GCRC supported by M01 RR00125.

REFERENCES

  1. Top of page
  2. Abstract
  3. METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgments
  7. REFERENCES
  • 1
    Begley CE, Beghi E, Beran RG, et al. ILAE Commission on the Burden of Epilepsy, Subcommission on the Economic Burden of Epilepsy: final report 1998–2001. Epilepsia 2002;43: 66873.
  • 2
    Cascino GD. Surgical treatment for epilepsy. Epilepsy Res 2004;60: 17986.
  • 3
    LudersHO, ComairYG, eds. Epilepsy surgery, 2nd ed. Philadelphia : Lippincott Williams & Wilkins, 2001.
  • 4
    Rosenow F, Luders H. Presurgical evaluation of epilepsy. Brain 2001;124: 1683700.
  • 5
    Lee SA, Spencer DD, Spencer SS. Intracranial EEG seizure-onset patterns in neocortical epilepsy. Epilepsia 2000;41: 297307.
  • 6
    Hamer HM, Morris HH, Mascha EJ, et al. Complications of invasive video-EEG monitoring with subdural grid electrodes. Neurology 2002;58: 97103.
  • 7
    Simon SL, Telfeian A, Duhaime AC. Complications of invasive monitoring used in intractable pediatric epilepsy. Pediatr Neurosurg 2003;38: 4752.
  • 8
    Duncan JS. Neuroimaging methods to evaluate the etiology and consequences of epilepsy. Epilepsy Res 2002;50: 13140.
  • 9
    Spencer SS, Bautista RE. Functional neuroimaging in localization of the ictal onset zone. Adv Neurol 2000;83: 28596.
  • 10
    Spencer SS. The relative contributions of MRI, SPECT, and PET imaging in epilepsy. Epilepsia 1994;35(suppl 6):S7289.
  • 11
    Spencer SS, Theodore WH, Berkovic SF. Clinical applications: MRI, SPECT, and PET. Magn Reson Imaging 1995;13: 111924.
  • 12
    Spanaki MV, Spencer SS, Corsi M, et al. Sensitivity and specificity of quantitative difference SPECT analysis in seizure localization. J Nucl Med 1999;40: 7306.
  • 13
    Devous MD Sr, Leroy RF, Homan RW. Single photon emission computed tomography in epilepsy. Semin Nucl Med 1990;20: 32541.
  • 14
    Andersen AR. 99mTc-D,L-hexamethylene-propyleneamine oxime (99mTc-HMPAO): basic kinetic studies of a tracer of cerebral blood flow: Cerebrovasc Brain Metabol Rev 1989;1: 288318.
  • 15
    Shulman RG, Rothman DLE. Brain energetics and neuronal activity: applications to fMRI and medicine. New York : John Wiley & Sons, 2004.
  • 16
    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.
  • 17
    Devous MD Sr, Thisted RA, Morgan GF, et al. SPECT brain imaging in epilepsy: a meta-analysis. J Nucl Med 2004;39: 28593.
  • 18
    Chang DJ, Zubal IG, Gottschalk C, et al. Comparison of statistical parametric mapping and SPECT difference imaging in patients with temporal lobe epilepsy. Epilepsia 2002;43: 6874.
  • 19
    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.
  • 20
    Knowlton RC, Lawn ND, Mountz JM, et al. Ictal SPECT analysis in epilepsy: subtraction and statistical parametric mapping techniques. Neurology 2004;63: 105.
  • 21
    Koo WC, Devinsky O, Hari K, et al. Stratifying differences on ictal/interictal subtraction SPECT images. Epilepsia 2003;44: 37986.
  • 22
    Brinkmann BH, O'Brien TJ, Webster DB, et al. Voxel significance mapping using local image variances in subtraction ictal SPET. Nucl Med Commun 2000;21: 54551.
  • 23
    Lee JD, Kim HJ, Lee BI, et al. Evaluation of ictal brain SPECT using statistical parametric mapping in temporal lobe epilepsy. Eur J Nucl Med 2000;27: 165865.
  • 24
    Avery RA, Spencer SS, Spanaki MV, et al. Effect of injection time on postictal SPET perfusion changes in medically refractory epilepsy. Eur J Nucl Med 1999;26: 8306.
  • 25
    Newton MR, Berkovic SF, Austin MC, et al. Ictal postictal and interictal single-photon emission tomography in the lateralization of temporal lobe epilepsy. Eur J Nucl Med 1994;21: 106771.
  • 26
    Rowe CC, Berkovic SF, Austin MC, et al. Patterns of postictal cerebral blood flow in temporal lobe epilepsy: qualitative and quantitative analysis. Neurology 1991;41: 109613.
  • 27
    Newton MR, Berkovic SF, Austin MC, et al. Postictal switch in blood flow distribution and temporal lobe seizures. J Neurol Neurosurg Psychiatry 1992;55: 8914.
  • 28
    Zubal IG, Spanaki MV, MacMullan J, et al. Influence of technetium-99m-hexamethylpropylene amine oxime injection time on single-photon emission tomography perfusion changes in epilepsy. Eur J Nucl Med 1999;26: 127.
  • 29
    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.
  • 30
    Penfield W. The evidence for a cerebral vascular mechanism in epilepsy. Ann Intern Med 1933;7: 30310.
  • 31
    Lee HW, Hong SB, Tae WS. Opposite ictal perfusion patterns of subtracted SPECT: hyperperfusion and hypoperfusion. Brain 2000;123: 21509.
  • 32
    Menzel C, Grunwald F, Klemm E, et al. Inhibitory effects of mesial temporal partial seizures onto frontal neocortical structures. Acta Neurol Belg 1998;98: 32731.
  • 33
    Rabinowicz AL, Salas E, Beserra F, et al. Changes in regional cerebral blood flow beyond the temporal lobe in unilateral temporal lobe epilepsy. Epilepsia 1997;38: 10114.
  • 34
    Van Paesschen W, Dupont P, Van Driel G, et al. SPECT perfusion changes during complex partial seizures in patients with hippocampal sclerosis. Brain 2003;126: 110311.
  • 35
    Blumenfeld H, McNally KA, Vanderhill S, et al. Positive and negative network correlations in temporal lobe epilepsy. Cereb Cortex 2004;14: 892902.
  • 36
    EngelJ, PedleyTA, eds. Epilepsy: a comprehensive textbook. Philadelphia : Lippincott Williams & Wilkins, 1997.
  • 37
    Studholme C, Novotny E, Zubal IG, et al. Estimating tissue deformation between functional images induced by intracranial electrode implantation using anatomical MRI. Neuroimage 2001;13: 56176.
  • 38
    McNally KA, Davis K, Doernberg SB, et al. Cerebral blood flow changes during secondarily generalized tonic-clonic seizures. Soc Neurosci Abst Online at http://web.sfn.org/ 2004.
  • 39
    Gottschalk PCH, Kosten TR. Isradipine enhancement of cerebral blood flow in abstinent cocaine abusers with and without chronic perfusion deficits. Am J Addict 2002;11: 2008.
  • 40
    Acton PD, Friston KJ. Statistical parametric mapping in functional neuroimaging: beyond PET and fMRI activation studies. Eur J Nucl Med 1998;25: 6637.
  • 41
    Friston KJ, Holmes A, Poline JB, et al. Detecting activations in PET and fMRI: levels of inference and power. Neuroimage 1996;4: 22335.
  • 42
    Blumenfeld H, Westerveld M, Ostroff RB, et al. Selective frontal, parietal and temporal networks in generalized seizures. Neuroimage 2003;19: 155666.
  • 43
    Metz CE. Basic principles of ROC analysis. Semin Nucl Med 1978;VIII: 28398.
  • 44
    Swets JA. ROC analysis applied to the evaluation of medical imaging techniques. Invest Radiol 1979;14: 10921.
  • 45
    Friston KJ, Holmes AP, Worsley KJ. How many subjects constitute a study? Neuroimage 1999;10: 15.
  • 46
    Asenbaum S, Brucke T, Pirker W, et al. Imaging of cerebral blood flow with technetium-99m-HMPAO and technetium-99m-ECD: a comparison. J Nucl Med 1998;39: 6138.
  • 47
    Ramsay SC, McLaughlin AF, Greenough R, et al. Comparison of independent aura, ictal and interictal cerebral perfusion. J Nucl Med 1992;33: 43840.
  • 48
    Blumenfeld H, Gale K, Bertram EH, et al. Epilepsy as a disorder of large-scale neural networks. Epilepsia 2000;41(suppl 7):75.
  • 49
    Brett M, Leff AP, Rorden C, et al. Spatial normalization of brain images with focal lesions using cost function masking. Neuroimage 2001;14: 486500.
  • 50
    Shin WC, Hong SB, Tae WS, et al. Ictal hyperperfusion patterns according to the progression of temporal lobe seizures. Neurology 2002;58: 37380.
  • 51
    Nersesyan H, Herman P, Erdogan E, et al. Relative changes in cerebral blood flow and neuronal activity in local microdomains during generalized seizures. J Cereb Blood Flow Metab 2004;24: 105768.