Professional Documents
Culture Documents
Z Score SISCOM
Z Score SISCOM
Z Score SISCOM
doi: 10.1111/epi.12139
*Department of Neurology General Neurology, Cleveland Clinic Foundation, Cleveland, Ohio, U.S.A.; †Department of Neurology,
Epilepsy Center, Cleveland Clinic Foundation, Cleveland, Ohio, U.S.A.; ‡Department of Neurology, Westmead Hospital,
Sydney, New South Wales, Australia; §Toshiba Medical Imaging Research, Mayfield Village, Ohio, U.S.A.; and
¶Department of Nuclear Medicine, Cleveland Clinic Foundation, Cleveland, Ohio, U.S.A.
Ictal single-photon emission computed tomography patients who have failed previous resections (Wetjen et al.,
(SPECT) is frequently used to aid presurgical work-up. 2006). Recent prospective studies have shown that SPECT
Analysis of ictal SPECT is usually performed in comparison can be used to generate implantation hypothesis and
with an interictal SPECT image. Ictal-interictal SPECT with improve surgical outcome in patients referred for invasive
co-registration to magnetic resonance imaging (MRI) (SIS- electroencephalography (EEG) evaluations (Matsuda et al.,
COM) has proven to be a highly valuable diagnostic tool in 2009; von Oertzen et al., 2011).
noninvasive localization of the seizure-onset zone (Kaibori- Proper selection of a z score (or threshold) is critical for
boon et al., 2002; Lee et al., 2006; Wichert-Ana et al., consistent, reliable, and successful analysis of SISCOM
2008), especially in challenging patients with nonlesional data (O’Brien et al., 1998b; Van Paesschen, 2004; Van
and/or extratemporal epilepsies (O’Brien et al., 2000), and Paesschen et al., 2007). Traditionally, a z score of 2 has
been proposed (O’Brien et al., 1998a). However, SISCOM
Accepted January 29, 2013. images often display multiple areas of hyperperfusion,
Address correspondence to Andreas V. Alexopoulos, Department of
Neurology, Epilepsy Center, Cleveland Clinic Foundation, 9500 Euclid
which include both the areas of ictal onset and seizure prop-
Avenue, Cleveland, OH 44195, U.S.A. E-mail: alexopa@ccf.org agation (Van Paesschen, 2004). Given the dynamic nature
Wiley Periodicals, Inc. of SPECT, a fixed, conservative threshold may not be opti-
© 2013 International League Against Epilepsy mal in all patients. Some users may actually vary the z score
793
794
C. R. Newey et al.
for different patients, but the sensitivity and specificity at 30 mAs, 5 mm slice) within 2 h after radiopharmaceutical
each z score have not been systematically studied. Further- injection. Interictal SPECT studies were accomplished after
more, different hyperperfusion patterns emerge at each z minimum of 24 h of no seizures. Timing of injection was
score, which may provide additional localizing information determined on retrospective video and EEG review. Injec-
(Dupont et al., 2006). It is the purpose of this study to sys- tion time was defined as the time point at which the radio-
tematically evaluate the localizing value of z scores of 1, pharmaceutical-containing syringe plunger had been
1.5, 2, and 2.5 and to examine the localizing information of completely depressed. Seizure onset was defined as the ear-
hyperperfusion patterns observed at each z score. liest ictal EEG or clinical evidence of seizure activity, and
seizure cessation when seizure activity was no longer pres-
ent on EEG (Wong et al., 2010).
Methods
Patient screening SISCOM
After obtaining approval by the institutional review Ictal and interictal SPECT studies were coregistered
board, we screened 94 consecutive patients who underwent using an automatic registration algorithm based on mutual
either temporal or extratemporal resection from 2007 to information (Maes et al., 1997). The interictal image was
2009 at the Cleveland Clinic Epilepsy Center. Patients were subtracted from the ictal study after normalization, and then
included if they: (1) underwent ictal and interictal SPECT the subtracted image was smoothed using three-dimensional
studies preoperatively; (2) had preoperative and postopera- (3D)–Gaussian smoothing kernel (full width at half maxi-
tive MRIs; (3) became seizure-free postoperatively (mini- mum = 12 mm). This information was transformed into a z
mum 6-month follow-up). Patients with previous history of score using the mean and standard deviation of the differ-
epilepsy surgery were excluded. In all patients, resective ences in all brain voxels. The mean activation map was then
surgery strategy and extent of resection were determined by used for coregistration to the preoperative MRI for anatomic
consensus at patient management conference on the basis of localization (O’Brien et al., 1998b; Dupont et al., 2006).
integration of all available clinical information and investi-
gations such as scalp and/or intracranial electroencephalog- Blinded SISCOM review
raphy, MRI [18F]fluorodeoxyglucose positron emission All SISCOM images were then reviewed for cortical
tomography (FDG-PET), SPECT, magnetoencephalogra- localization by two reviewers (CW, CN) blinded to clinical
phy (MEG), and neuropsychological assessment. During data. The whole brain was divided into 10 regions including
patient management conference, SPECT SISCOM threshol- left and right frontal, temporal, perirolandic, parietal, and
ded with a z score of 2 had been used for clinical interpreta- occipital based on previous studies (McNally et al., 2005).
tion and decision-making. Concordance between the SPECT hyperperfusion and
extent of resection was determined at the level of these 10
SPECT injection and imaging regions. Regions of hyperperfusion were superimposed on
Epilepsy-trained nurses sat by the bedside and injected the presurgical MRI (Fig. 1A), using MRICRO software
the radiopharmaceutical tracer in all studies. 99mTc-ECD (Georgia Institute of Technology, Atlanta, GA, U.S.A.) with
(ethyl-cysteinate dimer; 20–40 mCi) was injected into a varying z scores of 1, 1.5, 2, and 2.5. Reviewers were then
forearm vein immediately after noting seizure onset. asked to localize, if possible, each individual interictal-ictal
SPECT images were acquired on a Siemens (Erlangen, SPECT study to one or more of the 10 cerebral regions.
Germany) Symbia dual-head camera (SPECT: 15 s per stop Emulating the approach applied to patient management con-
960 stops, 128 9 128 matrix, iterative reconstruction with ference, the criteria used to determine localization included:
attenuation correction, six iteration, eight subsets; CT: SISCOM image intensity, location of hyperperfusion, and
A B
Figure 1.
SISCOM analysis and MRI coregis-
tration. Panel A shows SISCOM
analysis performed on preoperative
MRI (Z = 2). Panel B shows the
coregistered postoperative MRI,
which presents a resection consis-
tent with SISCOM localization.
Epilepsia ILAE
number of regions of hyperperfusion. Final localization was but not vice versa (Fig. 2B). Pattern 3 consists of multiple
based on agreement of the two reviewers. If disagreement SISCOM hyperperfusion areas not within or near the surgi-
occurred, a third reviewer (GW) evaluated the images. cal resection site.
Agreement between this reviewer and one of the two other
reviewers was the final determination. If no agreement Statistics In this statistics section is described quite nicely
occurred, the study was considered nonlocalizing. Continuous measures were described as means, standard
deviations, and percentiles. Categorical measures were
Determination of SISCOM localization value and summarized using frequencies and percentiles. Generalized
hyperperfusion patterns estimating equations (GEEs) were performed to evaluate
We unblinded the data after completing the above blinded the relationship between agreement and z score, and assess
portion of the study. We then judged correctness of localiza- the association between correct localization and injection
tion by introducing the patient’s postoperative MRI, which time as percent seizure time across all z scores. Because
was obtained 3–6 months after resection and was coregis- there are four z scores and multiple measurements for each
tered with the corresponding preoperative MRI (Fig. 1B). If patient (due to multiple reviewers), the correlation of the
the area determined by the final decision of localization in measurements within patient was adjusted by using a
the previous step fell within the surgical resection, then the compound symmetry covariance structure during the GEE
localization decision was considered correct. model selection procedure. Logistic regression was per-
Reviewers further identified the hyperperfusion patterns formed to assess the association between correct localiza-
as a separate unblinded step. By modifying the previously tion and injection time (or percent injection time over total
published descriptions by Dupont et al. (2006), hyperperfu- seizure time) for each z score. All analyses were performed
sion patterns were classified as follows: Pattern 1 describes at a significance level of p < 0.05 using SAS 9.2 software
the pattern with the largest and most intense (highest z (SAS Institute, Cary, NC, U.S.A.).
score) hyperperfusion falling within the surgical resection Sensitivity and specificity analyses were performed
site (Fig. 2A). Pattern 2 describes SISCOM hyperperfusion based on a modified version of the methodology by McNal-
with a bilobulated, that is, hourglass appearance with the ly et al. (2005). For each predetermined threshold, the 10
brightest lobule being outside the resection and the less cerebral regions (left and right frontal, temporal, perirolan-
intense of the two lobules residing within the resection site, dic, parietal, and occipital) were evaluated to assess whether
Figure 2.
Representative SPECT images at varying z scores superimposed on the postoperative MRI. In both examples, the z score of 1.5 shows
optimal sensitivity. Panel A shows the largest and most intense hyperperfusion overlapping with the resected site (perfusion pattern 1).
Panel B shows an hourglass pattern of hyperperfusion overlapping with the resected area (perfusion pattern 2). Center of resection is
denoted by blue cross.
Epilepsia ILAE
one or several lobar area(s) contained areas of hyperperfu- 1.5 is significantly superior (p = 0.005), when compared to
sion, based on the reviewers’ agreed decision. By definition the traditional z score of 2, as well as 1 and 2.5. There are no
the site of resection should encompass the true epilepto- significant differences in terms of localization value
genic zone in patients who experience sustained seizure between z scores 2 and 1, 2.5 and 1, and between 2 and 2.5.
freedom following epilepsy surgery. Therefore, for a given Figure 2 shows representative SPECT images at varying z
region, the sensitivity and specificity parameters in this scores superimposed on the patient’s postoperative MRI. In
cohort were defined as follows: true-positive (hyperperfu- both examples, a 1.5 z score shows optimal sensitivity.
sion and region resected), false-positive (hyperperfusion but Agreement between reviewers was poor with SISCOM
not resected), true-negative (no hyperperfusion and no threshold of 1; Kappa coefficient was 0.29 (95% confidence
resection), and false-negative (absence of hyperperfusion in interval [CI] 0.16, 0.74). A z score of 1.5 and 2 had moder-
resected region). Sensitivity and specificity values were cal- ate Kappa coefficients, 0.70 (95% CI 0.44, 0.96) and 0.60
culated at all four z scores. (95% CI 0.29, 0.91), respectively. A z score of 2.5 had a
high Kappa coefficient of 0.90 (95% CI 0.70, 1), where the
reviewers agreed on all but one case.
Results
Patient profile Localizing value of each hyperperfusion pattern
Twenty-six patients (nine male, 34.6%) met inclusion The bilobulated pattern 2 was encountered the least
criteria. Mean age was 36.5 years (standard deviation [SD] (n = 11); when present, it always correctly localized the
15.2 years; range 5–67 years). Preoperative mean seizure epileptogenic focus (100%; Fig. 2B). It was most frequently
burden was 9.5 seizures/week with 18 (69.2%) having a his- observed at z score 1.5 (n = 7). Pattern 1 was present in 41
tory of generalized motor seizures. MRI was abnormal in 19 instances and correctly localized in 38 (92.7% of occur-
(73%). Twenty-three (88.5%) had PET and 17 (65.4%) rences). It was also encountered most frequently at z score
underwent intracranial EEG. Half of these patients had 1.5 (n = 12). Pattern 3 was observed most frequently at z
extratemporal epilepsy, and the other half had temporal lobe score of 1. Of 52 encounters of pattern 3, reviewers had
epilepsy (nonmesial temporal in the majority of patients). localized correctly the epileptogenic zone in two cases
Surgical pathology was cortical dysplasia (n = 10, 38.5%), (3.8%). Results are summarized in Table 2. Patterns 1 and
infarct (n = 7, 26.9%), vascular malformation (n = 2, 2, exemplified in Fig. 2A,B, respectively, are both clearly
7.7%), hippocampal sclerosis (n = 3, 11.5%), tumor more localizing than 3 (odds ratio >1), as confirmed using
(n = 3), and nonspecific (n = 1, 3.8%). Mean postoperative the GEE model. We were unable to compare pattern 1
follow up was 20 months (range 7–35 months). Seven of against pattern 2, as the GEE model did not converge due to
the 26 patients had secondarily generalized seizures at the the much less frequent occurrence of pattern 2.
time of ictal SPECT.
Injection time
Average injection time was 18.4 s (range 3–63 s), with Table 1. Pair-wise comparisons of each z score with
the majority of patients (61.4%) having injections under regards to its localization accuracy
20 s. Average ictal duration was 75.6 s (range 17–260 s). z score pairs Odds ratio 95% CI p-Value
With an estimated peak brain uptake of 30 s (Andersen, 1.5 vs. 1 4.55 1.78 11.67 0.002
1989; Kaminska et al., 2003; Dupont, 2008; Pastor et al., 1.5 vs. 2 4.23 1.54 11.62 0.005
2008), the majority of SPECT studies (17 of 26) occurred 1.5 vs. 2.5 7.53 2.56 22.18 0.0002
ictally; 6 patients had short seizures lasting <30 s. Despite 2 vs. 1 1.08 0.42 2.78 0.88
2 vs. 2.5 1.78 0.88 3.59 0.11
expected heterogeneity of seizures in this group, we found
2.5 vs. 1 0.60 0.22 1.65 0.33
no significant correlation between injection time
(p = 0.36), or percentage of injection time over total seizure
time (p = 0.39), and localization accuracy as indicated by Table 2. Localization accuracy of each pattern indicated
the GEE model. by number of correct localization at each z score
z score Total
Localizing value of each z score Total no. of correct no. of
At z score of 1, 34.6% (n = 9) of studies were correctly 1 1.5 2 2.5 localization (%) occurrence
localized. The correctly localized studies increased to Perfusion 7 12 10 9 38 (92.7) 41
76.9% (n = 20) at a z score of 1.5. Higher z scores were Pattern 1
associated with a lower percentage of correctly localized Perfusion 1 7 3 0 11 (100) 11
studies: 50% (n = 13) at 2% and 34.6% (n = 9) at 2.5. Pattern 2
Perfusion 1 1 0 0 2 (3.8) 53
Table 1 shows pair-wise comparisons of each individual z
Pattern 3
score with regard to its localization correctness. A z score of
Epilepsia, 54(5):793–800, 2013
doi: 10.1111/epi.12139
797
Varying z Score in SPECT SISCOM
Patterns of ictal hyperperfusion of the hourglass perfusion pattern and true ictal versus
We found that all occurrences of the hour-glass configu- “post-ictal” injection. Earlier injections have been sug-
ration (pattern 2) led to a correct decision regarding locali- gested to be more localizing compared to later injections
zation of the ictal focus. This bilobulated hourglass (Noachtar et al., 1998; Hong et al., 2008). This is particu-
configuration is thought to indicate ictal propagation, with larly important in extratemporal epilepsies, where propaga-
the lobule of highest z score residing outside the seizure- tion is more likely to occur early after ictal onset (Noachtar
onset zone, whereas the less intense lobule lies closest to the et al., 1998). Theoretically, the most useful ictal SPECT is
area of seizure generation (Dupont et al., 2006). Because of one that has the radioisotope injected seconds before ictal
its infrequent occurrence, we were unable to determine sta- onset. This is demonstrated in few “preictal” SPECT studies
tistically if pattern 2 has higher localizing value when com- that were captured either by serendipity or in special cir-
pared to pattern 1. Nonetheless, our study suggests that the cumstances when seizures can be provoked (Hong et al.,
identification of an hourglass-appearing SISCOM pattern 2008; Pastor et al., 2008; Barba et al., 2009). These studies
can be highly valuable and leads to more accurate localiza- showed a high degree of colocalization between the area of
tion of the ictal-onset zone. “preictal” SPECT hyperperfusion and the corresponding
For our study, we used a modified version of the images epileptogenic focus. Lastly, 7 of 26 patients in our study had
proposed by Dupont et al. to determine patterns of hyper- secondarily generalized seizures at the time of SPECT
perfusion. The previous study examined hyperperfusion study. This observation may decrease localization value of
patterns in relationship to a preoperatively identified/ SISCOM images and confound the correlation analysis
known dysplastic lesion on MRI (Dupont et al., 2006). (Varghese et al., 2009). Nonetheless we did not find any
In our study, initial determination of SISCOM localization correlation between the existence of the hourglass perfusion
was made based on agreement among blinded reviewers. pattern and the presence or absence of secondarily general-
Subsequent analysis of perfusion patterns was performed ized seizures.
as a separate unblinded step, after coregistering the SIS-
COM images on the patient’s postoperative MRI. Despite Limitations and future directions
the different approaches, the two studies yielded similar Inherent limitations exist due to the retrospective nature
conclusions. It is important to note areas of hyperperfusion of our current study. Although we included all ictal SPECT
identified by ictal SPECT may point not only to the studies performed in a consecutive patient series, it should
presumed ictal onset, but also to the pathways and regions be noted that not all patients undergoing presurgical evalua-
of seizure propagation (Noachtar et al., 1998; Shin et al., tion are referred for ictal SPECT studies, especially when
2002; Dupont et al., 2006). Hyperperfusion patterns that there is clear concordance between video-EEG and MRI
involve the largest area and/or exhibit the highest z score abnormalities, such as in patients with unilateral hippocam-
may in fact represent propagated activity. Therefore, the pal sclerosis (Velasco et al., 2011). This is likely the reason
area of highest z score may not need to be completely that very few patients with mesial temporal lobe epilepsy
resected or included in the resection to achieve seizure were included in this study (only three). Moreover, ictal
freedom (Van Paesschen et al., 2007). Rather, the resection SPECT is a cost-intensive and labor-intensive procedure
site should always be determined by a multimodal reserved for the more challenging patients undergoing pre-
approach, in the context of a full presurgical evaluation surgical evaluation.
(Van Paesschen, 2004; Kimura et al., 2012). Overall, our seizure-free cohort comprises a mixture of
temporal (50%) and extratemporal patients (50%), and it
Injection time included patients with both lesional and nonlesional preopera-
We did not find injection time, in its absolute value or its tive MRI studies. Although we screened a large number of
percentage over total seizure time, to have a significant cor- patients undergoing their first resective epilepsy surgery, we
relation with the probability of correct localization. This included only those who clearly achieved documented and
may be due to the small sample size and/or the relatively sustained seizure freedom. The relatively small sample size
rapid radioisotope injection time. Indeed the average injec- of 26 eligible patients did not allow for subgroup analyses.
tion time in this series was 18.4 s, with seizure duration Nonetheless, our study is the largest study to date to evaluate
averaging 75.6 s, and 50% of patients having completed the the localizing value of various z scores in patients who under-
tracer injection within 15 s of seizure onset. As a result the went SISCOM analysis and successful epilepsy surgery.
majority of patients (17/26) in our study had a true ictal Earlier studies evaluating SISCOM localization judged
SPECT study, even after allowing for the generally accepted the efficacy of SPECT localization by comparison with sei-
minimum of 30 s of seizure duration postinjection for tran- zure foci suggested by other clinical information such as
sit time to occur (Andersen, 1989; Kaminska et al., 2003; EEG (Newton et al., 1995). In this situation, it could be dif-
Dupont, 2008; Pastor et al., 2008). Six of the remaining nine ficult to judge whether false SPECT localization was truly
patients had short seizures lasting <30 s. On the basis of false, or attributable to incorrect localization of seizure
these data we did not find correlation between the existence focus. In our study, we selected patients who became
Epilepsia, 54(5):793–800, 2013
doi: 10.1111/epi.12139
799
Varying z Score in SPECT SISCOM
Subtraction ictal SPECT co-registered to MRI improves clinical (2009) Clinical use of ictal SPECT in secondarily generalized tonic–
usefulness of SPECT in localizing the surgical seizure focus. Neurology clonic seizures. Brain 132:2102–2113.
50:445–454. Velasco TR, Wichert-Ana L, Mathern GW, Araujo D, Walz R, Bianchin
O’Brien TJ, So EL, Mullan BP, Hauser MF, Brinkmann BH, Jack CR Jr, MM, Dalmagro CL, Leite JP, Santos AC, Assirati JA Jr, Carlotti CG Jr,
Cascino GD, Meyer FB, Sharbrough FW. (1999) Subtraction SPECT Sakamoto AC. (2011) Utility of ictal single photon emission computed
co-registered to MRI improves postictal SPECT localization of seizure tomography in mesial temporal lobe epilepsy with hippocampal
foci. Neurology 52:137–146. atrophy: a randomized trial. Neurosurgery 68:431–436; discussion 436
O’Brien TJ, So EL, Mullan BP, Cascino GD, Hauser MF, Brinkmann BH, von Oertzen TJ, Mormann F, Urbach H, Reichmann K, Koenig R,
Sharbrough FW, Meyer FB. (2000) Subtraction peri-ictal SPECT is Clusmann H, Biersack HJ, Elger CE. (2011) Prospective use of
predictive of extratemporal epilepsy surgery outcome. Neurology subtraction ictal SPECT coregistered to MRI (SISCOM) in presurgical
55:1668–1677. evaluation of epilepsy. Epilepsia 52:2239–2248.
Pastor J, Dominguez-Gadea L, Sola RG, Hernando V, Meilan ML, De Dios Wetjen NM, Cascino GD, Fessler AJ, So EL, Buchhalter JR, Mullan BP,
E, Martinez-Chacon JL, Martinez M. (2008) First true initial ictal O’Brien TJ, Meyer FB, Marsh WR. (2006) Subtraction ictal single-
SPECT in partial epilepsy verified by electroencephalography. photon emission computed tomography coregistered to magnetic
Neuropsychiatr Dis Treat 4:305–309. resonance imaging in evaluating the need for repeated epilepsy surgery.
Shin WC, Hong SB, Tae WS, Kim SE. (2002) Ictal hyperperfusion patterns J Neurosurg 105:71–76.
according to the progression of temporal lobe seizures. Neurology Wichert-Ana L, de Azevedo-Marques PM, Oliveira LF, Fernandes RM,
58:373–380. Velasco TR, Santos AC, Araujo D, Kato M, Bianchin MM, Sakamoto
Van Paesschen W. (2004) Ictal SPECT. Epilepsia 45(Suppl. 4):35– AC. (2008) Ictal technetium-99 m ethyl cysteinate dimer single-photon
40. emission tomographic findings in epileptic patients with polymicrogyria
Van Paesschen W, Dupont P, Sunaert S, Goffin K, Van Laere K. (2007) syndromes: a subtraction of ictal-interictal SPECT coregistered to MRI
The use of SPECT and PET in routine clinical practice in epilepsy. study. Eur J Nucl Med Mol Imaging 35:1159–1170.
Curr Opin Neurol 20:194–202. Wong CH, Mohamed A, Larcos G, McCredie R, Somerville E, Bleasel A.
Varghese GI, Purcaro MJ, Motelow JE, Enev M, McNally KA, Levin AR, (2010) Brain activation patterns of versive, hypermotor, and bilateral
Hirsch LJ, Tikofsky R, Paige AL, Zubal IG, Spencer SS, Blumenfeld H. asymmetric tonic seizures. Epilepsia 51:2131–2139.