Chapter 5 – Advanced SPECT image processing in MRI-negative refractory focal epilepsy

Chapter 5 Advanced SPECT image processing in MRI-negative refractory focal epilepsy

MRI-Negative Epilepsy, ed. Elson L. So and Philippe Ryvlin. Published by Cambridge University Press. © Cambridge University Press 2015.

Elson L. So

Terence J. O’Brien

Benjamin H. Brinkmann

Introduction

The phenomenon of hyperemia at the region of the seizure focus has been known for decades. Perfusion SPECT is utilized to image patterns of altered regional cerebral blood flow (rCBF) for localizing the seizure onset zone. Thus far, perfusion SPECT is the only test that could image the altered physiology of all peri-ictal states (i.e., interictal, ictal, and postictal states).¹ The alteration during the interictal state is typically one of baseline hypoperfusion at the focus of seizure activity. In contrast, hyperperfusion transiently characterizes ictal activity, at the longest probably for 2 to 3 minutes beyond seizure termination. Ictal hyperperfusion switches to hypoperfusion during the early postictal state, typically and transiently to a degree markedly less than that of interictal perfusion (Figure 5.1).²

Figure 5.1 Graphic representation of the SPECT-detected perfusion changes corresponding to the different states of activity at the seizure focus.

(Adapted from,² with permission from Lippincott, Williams & Wilkins.)

The most frequently used SPECT radiotracers for peri-ictal perfusion imaging are ^99mTc-hexamethylpropylene amine oxime (Tc-HMPAO), and ^99mTc-Bicisate (Tc-ECD). Following injection of these ^99mTc-linked radiotracers during a seizure for ictal studies, they are quickly fixed to brain receptors because of their high first-pass brain extraction rate. The SPECT images can be acquired up to 3 to 4 hours after the injection, because of the relatively slow radioactivity decay of these ^99mTc-linked radiotracers.

Compared to ictal SPECT, interictal SPECT used alone has been shown to have low sensitivity in detecting the seizure focus. A major advancement in the use of SPECT in epilepsy evaluation has been the development of computerized subtraction techniques for detecting differences in perfusion patterns between the ictal state and the baseline interictal state in the same patient,³^, ⁴ or between the patient’s ictal perfusion pattern and the pattern in normal subjects.⁵^, ⁶ Both types of subtraction studies have been developed to improve the yield of peri-ictal SPECT studies.

The purpose of this chapter is to discuss and demonstrate the role of modern SPECT acquisition and analysis in the presurgical evaluation of patients with drug-resistant MRI-negative epilepsy. Strategies for enhancing the yield of ictal SPECT studies will also be discussed.

Subtraction SPECT techniques

Subtraction ictal–interictal SPECT coregistered to MRI

Injection of the SPECT radioisotope should be performed as soon as practically possible following seizure onset to ensure the blood flow pattern observed represents the characteristic increase at the seizure onset zone, rather than regions of propagated seizure or postictal activity. For subtraction ictal SPECT coregistered to MRI (SISCOM), the images are transferred to a workstation for processing after acquisition and reconstruction of ictal and interictal SPECT image.⁴^, ⁷ Cerebral activity is separated from extracerebral activity by using thresholding to create a binary mask of the cerebral region and the mean of the ictal and interictal cerebral pixel intensities are calculated. Ictal and interictal pixel intensities are rescaled to a constant mean pixel intensity to normalize differences in tracer uptake, retention, and decay. The interictal scan is coregistered to the coordinate space of the ictal SPECT using a voxel intensity-based algorithm, and the interictal image is transformed and resampled using a high-quality interpolation algorithm. The normalized and transformed interictal SPECT images are subtracted from the normalized ictal SPECT to produce a difference image, representing the differences in cerebral blood flow between the two scans. The standard deviation of cerebral pixels in the difference image is calculated, and a threshold based on the standard deviation is applied to identify the most prominent areas of increased and decreased blood flow. Hyperperfusion (positive difference) images are examined for ictal phenomena, while hypoperfusion (negative difference) images are examined for perfusion decreases indicating the postictal change from increased to decreased perfusion.

Following processing of the SPECT scans, the ictal SPECT is coregistered to a high-resolution structural magnetic resonance image (MRI) of the patient, typically using an algorithm maximizing the normalized mutual information between the two images. The ictal SPECT image can be displayed as an overlay on the MRI to verify the accuracy of registration. If the alignment is accurate, the thresholded hyperperfusion and hypoperfusion images are displayed as an overlay on the MRI to identify areas of increased and suppressed cerebral blood flow. This places a map of the physiological changes related to the seizure in the context of the structural anatomy in the MRI. These steps are diagrammed in Figure 5.2.

Figure 5.2. Diagram of steps in attaining subtraction ictal SPECT coregistered to MRI (SISCOM) images.

SPM-based SPECT analysis

The SPECT images of cerebral blood flow can also be compared statistically in reference to a population of paired scans of normal subjects. This approach has the advantage of incorporating normal physiological variation in a statistical comparison of the ictal and interictal SPECT scans,⁵^, ⁶^, ⁸^, ⁹ and it has been shown to have better yield than SISCOM. In these methods, a voxel-by-voxel estimate of normal interscan variation is constructed from paired resting scans of normal volunteers by statistical parametric mapping (SPM).¹⁰ The differences between normalized voxel intensities in the paired ictal and interictal scans of the patient are compared to the population of paired normal resting scans, thus generating a three-dimensional volume of t-statistics. Positive (hyperperfusion) and negative (hypoperfusion) thresholds are then applied to identify statistically significant perfusion changes. Because of the limited spatial resolution of SPECT imaging, a voxel cluster threshold corresponding to the spatial resolution of the imaging system is also applied to remove activations due to noise.⁹ These steps are diagrammed in Figure 5.3.

Figure 5.3 Diagram of steps in attaining statistical ictal SPECT coregistered to MRI (STATISCOM) images.

Yield of SISCOM in MRI-negative epilepsy

It is widely accepted that the greatest value of ictal SPECT in the presurgical evaluation of patients with drug-resistant epilepsy is when the MRI is negative, or when it shows either a large lesion or multifocal lesions.¹¹ In such cases, identifying a focal region of ictal perfusion abnormality, particularly with the use of subtraction techniques and MRI coregistration, can localize the seizure onset zone and provide a target for implantation of intracranial electrodes or surgical resection.¹² However, only few studies have specifically evaluated the yield of ictal SPECT in MRI-negative refractory epilepsy. In our initial study of the clinical utility of SISCOM, 26 of the 51 patients with drug-resistant temporal or extratemporal epilepsy had negative MRI. The SISCOM images in 92% (24/26) of the patients were judged to be localizing by blinded reviewers.¹³

In our subsequent series of 44 patients who underwent surgery for MRI-negative temporal lobe epilepsy, 33 had SISCOM studies, of which 82% (27/33) were localizing. The presence of a SISCOM abnormality localized to the resection site was found to be significantly associated with postsurgical seizure-free outcome. Other factors associated with the outcome were the absence of contralateral or extratemporal interictal epileptiform discharges, and the presence of subtle nonspecific MRI findings at site to the resection.¹⁴

We have also evaluated the usefulness of SISCOM in the presurgical evaluation of patients with MRI-negative extratemporal epilepsy. The SISCOM images were determined to be localizing by blinded reviewers in 77% (13/17) of the patients.¹⁵ Concordance of the SISCOM localization with the site of surgical resection was associated with a significantly better chance of excellent postsurgical outcome with respect to seizures, compared with those in whom the images were either nonconcordant or nonlocalizing (55% vs. 0%, p < 0.05). Moreover, the extent of resection of the region of primary cortical ictal hyperperfusion was predictive of postsurgical outcome. A subsequent study from our center shows that SISCOM was localizing in 68% (58/85) of the patients with MRI-negative extratemporal epilepsy.¹⁶ However, only 24 patients eventually underwent resective surgery, and no presurgical clinical factor or test result was found to be statistically associated with postsurgical outcome; 38% (9/24) were still seizure-free after 10 or more years after their surgery.

A more recent study shows that in 58% (74/130) of drug-resistant temporal or extratemporal epilepsy patients, SISCOM yielded results that permitted consideration for surgical resection or intracranial electrode implantation.¹¹ Half of these patients had negative MRI. However, only 38% (28/74) of these patients eventually underwent resective epilepsy surgery, ten of whom had negative MRI. Six of these ten patients had postsurgical seizure freedom. In another study, SISCOM was localizing in 93% (13/14) of the patients who had histologically proven cortical dysplasia but negative MRI.¹⁷

Yield of SPM-based subtraction SPECT in MRI-negative epilepsy

The use of normative dataset, against which ictal SPECT data are analyzed by SPM to discern seizure onset activity and location, could theoretically obviate the need for interictal SPECT study. Using this method set at a high threshold of p < 0.001, Lee and colleagues reported that hemispheric lateralization of the epileptogenic focus was correct in approximately 60% of their 21 temporal lobe epilepsy patients.¹⁸ (Focal or regional localization was not investigated.) The reason for the modest yield could be that when the seizure onset zone is interictally hypoperfused, seizure activity at the zone may increase perfusion only to a level at or near that of control subjects in the normative dataset. The SPM analysis then detects little or no difference between the degree of ictal perfusion of the patient and the degree of perfusion in the normal control dataset at the corresponding zone. In the same study, Lee and colleagues also analyzed with SPM the difference data from ictal–interictal subtraction. However, this method did not increase the sensitivity of detecting the epileptogenic zone.

Amorim and colleagues conducted a study similar in principle to that of Lee and colleagues’ study.¹⁹ Amorim and colleagues found that SPM comparison of ictal studies with normal control data detected the ictal onset zone in 64% of their 22 patients, even when the sensitivity threshold was set low at p value of < 0.05. The yield was improved to 77% when SPM analysis was applied to ictal–interictal subtraction data. However, either method fared lower than their visual review of ictal and interictal scans, which observed the “epileptogenic focus” in all patients.

The method of ictal–interictal SPECT analyzed by SPM (ISAS) was used to analyze data from ictal or postictal Tc-HMPAO injections in 28 studies of mesial temporal epilepsy and 19 studies of neocortical epilepsy.⁹ Analysis of the ictal injection studies had a sensitivity of 93% and specificity of 87% in localizing the epileptogenic temporal lobe, whereas the analysis for localizing the neocortical epileptogenic zone had a sensitivity of 77% and specificity of 93%. On the other hand, ISAS analysis of postictal injection studies was poor for localizing the seizure onset zone, but lateralization to the hemisphere of seizure onset was correct in approximately 80% of the studies. These favorable rates should be appreciated in the context of well-localized epilepsy; because all mesial temporal epilepsy patients in the report had lesional epilepsy (the proportion of lesional epilepsy in the neocortical group is unreported). With few exceptions, patients in Lee and colleagues’ and Amorim and colleagues’ reports also had lesional epilepsy.¹⁸^, ¹⁹ A study using a method similar to ISAS has already demonstrated high yields in lateralizing the epileptogenic zone in epilepsy with mesial temporal sclerosis.²⁰ In the study, ictal–interictal difference SPECT analysis of the hippocampus subregion by SPM achieved correct lateralization in 91% of the patients, and 87% with analysis of the amygdala subregion. These rates were comparable with the yield of MRI in these patients, which was 89% for FLAIR and 78% for volumetric abnormalities. However, all foregoing studies have not demonstrated the usefulness of SPM-based ictal SPECT in nonlesional refractory epilepsy. Peri-ictal perfusion characteristics at lesional seizure onset foci should not be presumed to be identical to those at nonlesional foci.

The foregoing SPM-based studies consisted mostly of patients who had responded to epilepsy surgery. Patients who were not ideal surgical candidates were mostly excluded from the reports, such as those with indeterminate or multifocal seizure onsets. In clinical practice, patients undergoing epilepsy surgery evaluation include those who are good surgical candidates such as the subjects in these studies, as well as those who are not. Patients who had prior nonlocalizing or conflicting data are the ones who most need functional imaging studies such as peri-ictal SPECT. It should also be noted that seizures with secondary generalization were excluded from the ISAS study.⁹

Thus far, statistical ictal SPECT coregistered to MRI (STATISCOM) is the only SPM-based method of ictal–interictal subtraction SPECT using normative data that has been shown to be useful in both MRI-positive and MRI-negative refractory epilepsies.⁶ Blinded reviewers in a study compared STATISCOM with SISCOM in 87 consecutive temporal lobe epilepsy surgery patients who were enrolled in the study regardless of their postsurgical outcome. The agreement between reviewers was better with STATISCOM than with SISCOM (kappa score of 0.81 vs. 0.36) A hyperperfusion focus was detected by STATISCOM in 84% of the patients and in 66% by SISCOM (p < 0.05). The STATISCOM correctly distinguished between mesial and lateral neocortical temporal epilepsy in 68% of the patients, compared with only 24% by SISCOM (p = 0.02). The superiority of STATISCOM over SISCOM in this regard was also demonstrated in the subgroup of 35 patients with MRI-negative epilepsy (80% correct with STATISCOM versus 47% with SISCOM; p = 0.04). Correct localization by STATISCOM within mesial vs. lateral neocortical subregion was associated with 81% postsurgical seizure freedom, whereas indeterminate subregional localization was associated with postsurgical seizure-free rate of 53% (p = 0.03). This association between subregional localization and surgical outcome was not present in SISCOM studies. The findings indicate that the epileptogenic zone is more accurately identified when a very focal subregional SPECT abnormality is detected, than when an abnormality is regional or undetectable, and that STATISCOM is better suited for detecting focal subregional temporal lobe abnormalities.

A subsequent study compared STATISCOM, ISAS, and SISCOM in 21 patients who had standard anterior temporal lobectomy for refractory MRI-negative epilepsy.²¹ All patients had no potentially epileptogenic lesions on their MRI studies that had been optimized for detecting the lesions. Radiotracer used in this study was Tc-ECD, and all injections were ictal with a mean latency of 26 seconds after seizure onset. Normative perfusion data were derived from 30 control subjects who each had two SPECT studies performed. The two scans on each control subject were performed on different days. Both hyperperfusion and hypoperfusion maps were generated with threshold set at p value of < 0.001 for ISA, and < 0.027 for STATISCOM. Three reviewers of the studies were blinded to all other clinical and laboratory data, and each type of SPECT study was assessed without knowledge regarding the other types of studies. Localization was concordant with the resected site in 71% of the patients with STATISCOM, 67% with ISAS, and only 38% with SISCOM (Figure 5.4). The localization rates were not statistically different between STATISCOM and ISAS, but either method was significantly better than SISCOM (both p < 0.001). Kappa score for interrater agreement was 0.82 for STATISCOM, 0.70 for ISAS, and 0.20 for SISCOM. Moreover, the reviewers’ mean confidence rating was also significantly higher with either STATISCOM or ISAS than with SISCOM (2.94 and 2.92 respectively, vs. 1.89; p < 0.001).

Figure 5.4 (a) STATISCOM and (b) ISAS showing similar hyperperfusion (red–orange) temporal hyperperfusion focus, whereas (c) SISCOM does not show the abnormality.

(Blue foci are nonlocalizing hypoperfusion changes.)

The STATISCOM, ISAS, and SISCOM methods were also compared in a separate study of MRI-negative refractory extratemporal epilepsy (personal communication). Similar to findings in MRI-negative temporal epilepsy, interobserver agreement between blinded reviewers was better for STATISCOM or ISAS than for SISCOM (0.66, 0.44 vs. 0.36). Rate of detecting a hyperperfusion focus was also higher with STATISCOM or ISAS than with SISCOM (90%, 92% vs. 62%). In seizure-free patients, the rate of concordance between the focus and resection was also superior for either STATISCOM or ISAS than SISCOM (80%, 77% vs. 47%).

Techniques for optimizing yield of SPECT

The rapidity of SPECT radiotracer uptake is generally considered the advantage of SPECT imaging over other functional imaging tests, because when the radiotracer is injected very soon after seizure onset, the focus of maximal radiotracer uptake may represent the seizure onset zone. However, if the radiotracer injection is delayed, the radiotracer uptake pattern may represent seizure propagation pathways or destination, giving rise to false localization. Prompt radiotracer injection is aided by presence of reliable epileptic auras, obvious early clinical or EEG manifestations of seizures, and absence or delayed seizure spread. Seizure generalization consistently diminishes the yield of peri-ictal SPECT for seizure localization.²² Moreover, seizures in some patients have short duration, especially extratemporal-onset seizures. In which case, the radiotracer may be injected postictally rather than ictally.²³ The yield of postictal SPECT is less than that of ictal SPECT.¹² Table 5.1 shows the techniques for optimizing yield of ictal SPECT studies for seizure localization.

Table 5.1 Techniques for optimizing yield of ictal SPECT studies

Case example

A 21-year-old right-handed man had afebrile unprovoked seizures since 15 months of age. Seizures in the first decade of life consisted of staring and unresponsiveness to verbal communication occurring for a few seconds, but up to 50 times a day. Since 10 years of age, seizure manifestations had been noted as preferentially nocturnal sleep-related brief attacks of stiffening and jerking of the right extremities. However, these seizures have later been described as tingling all over the body, followed by unresponsiveness and observed pupillary dilatation, rocking motion of the torso, and flailing of all extremities. Each seizure was reported to last only 15 seconds at the longest, with only 25% of the seizures reported to occur at night. Multiple antiepileptic medications had failed to control his seizures, including phenytoin, primidone, carbamazepine, and topiramate. Neurological examination showed mild mental retardation but no other abnormalities. Neuropsychological evaluation revealed full scale IQ of 80, verbal IQ of 88, and performance IQ of 75. Relative compromise of phonemic fluency and complex problem-solving skill was thought to be consistent with left anterior brain dysfunction.

Epilepsy protocol brain MRI showed no lesion. Interictal EEG recording disclosed left frontotemporal epileptiform discharges. Video-EEG monitoring recorded multiple episodes of sudden arousal from sleep and violent rocking to-and-fro of the torso with flailing of all extremities, each episode lasting about 20 seconds. There was mild postictal confusion, which lasted for a little over 1 minute. The EEG during seizure onset showed sudden appearance of muscle and movement artifacts from a background of sleep activity, and no discernible seizure discharge.

A Tc-ECD was injected 7 seconds after the termination of a 19-second typical seizure. The SISCOM analysis showed left lateral frontal convexity focus of hyperperfusion (see Figures 5.5 a and b). The focus was subsequently confirmed, by a total of 44 contacts of the subdural grid and strip recording, to be concordant with the EEG seizure onset zone. Electrocortical stimulation of the contacts showed that the expressive speech area was superior and posterior to the SISCOM focus and the ictal onset zone. Subsequent resection of the SISCOM and ictal EEG onset regions was followed by seizure freedom while still taking antiepileptic medications for more than 5 years.