General considerations

The main role of imaging in a patient with focal epilepsy is to explore its cause. The MRI remains the mainstay examination in this situation and allows the distinction between a patient with “surgical” lesions that are able to explain the patient’s seizures, and MRI-negative patients who do not have lesions, or only show nonspecific or incidental findings (see Chapters 1 and 3).

Despite the availability of guidelines for MRI neuroimaging in patients with focal epilepsies (1, 2), in clinical practice patients may be referred to as “MRI-negative” when in fact expert review or repeat MRI reveal epileptogenic lesions (3). Similarly, the imaging features of hippocampal sclerosis and focal cortical dysplasia – among the most frequent epileptogenic lesions – were not well characterized until the late 1990s, and earlier studies may have missed relevant abnormalities. In this chapter, we will whenever possible try to include only studies of MRI-negative patients that fulfill minimum MRI quality criteria (1, 2).

Positron emission tomography (PET) is a nuclear medicine technique. Substances labeled with a radioactive isotope, usually ¹⁸F (half-life ~2h) or ¹¹C (half-life ~20 min), are injected intravenously and are taken up by neurons as an energy substrate ([18F]fluorodeoxyglucose, FDG) or are bound reversibly or irreversibly to receptors. The isotopes emit positrons which annihilate on encountering an electron in tissue, emitting two gamma rays at ~180° of each other, which can be detected by detector crystals arranged as rings in the PET camera and used for reconstruction of quantified tomographic images. Static or dynamic (time course) images of radioactivity distribution in the brain are acquired after injection. In the case of FDG, static images are generally used; receptor studies generally require dynamic acquisition of ~15–30 images or frames over 60–90 minutes. Arterial plasma concentration of radioactivity or radioactivity time courses in a region devoid of specific binding can then be used in mathematical models to derive binding parameters. Depending on which labeled substance has been injected, glucose metabolism (via FDG) or various receptor systems can be assessed. The FDG is by far the most widely used radiopharmaceutical. More detailed reviews are available elsewhere (e.g., 4–6).

Whatever the MRI finding, it is important to assess the PET finding with reference to the structural anatomy, i.e., coregister MRI and PET (7, 8). Otherwise, apparent hypometabolism or reduced receptor binding can be caused by focal atrophy or a wide sulcus; truly reduced signal, particularly at the bottom of a sulcus, can be missed as it may be similar to surrounding white matter (Figure 4.1).

Figure 4.1 Added value and complementarity of combined FDG-PET and MR imaging. Example of a focal pharmacoresistant epilepsy. The FDG-PET alone does not show a clear-cut asymmetry. The MRI alone only shows unusual gyration (arrow). The superposition of both images following posthoc coregistration reveals a focal area of hypometabolism, restricted to a single sulcal bottom. The patient has remained free of seizures for 3 years so far after a focal resection restricted to this gyrus.

(Data kindly provided by F. Chassoux; Figure reproduced from reference 58.)

Another important methodological advance of the past few years has been the addition of statistical analyses to the standard visual analysis of PET, often in the form of the widely used statistical parametric mapping software (SPM; www.fil.ion.ucl.ac.uk/spm) or via cortical asymmetry analyses (e.g., 9).

Various methodological pitfalls include image artefacts and presence of even only a few subjects with abnormalities in the control group (4), different ages between controls and patients (10), and details of quantification (11).

Blood flow PET uses H₂[¹⁵O]; the half-life of ¹⁵O is ~2 minutes. Interictal scans are unreliable, and in contrast to SPECT, ictal scans are generally impossible to obtain in practice due to the short half-life of ¹⁵O. For the assessment of brain function, H₂[¹⁵O] PET has been all but superseded by fMRI (see Chapter 8), even if some areas of application remain in the research setting (e.g., interest in brain areas affected by susceptibility artefact on MRI; requirement for noiseless scans).

FDG-PET

Hypometabolism is a hallmark of the epileptogenic zone as well as surrounding areas. The FDG-PET method has been used in epilepsy since the 1980s (12, 13), i.e., since well before routine availability of MRI. Areas of hypometabolism are often larger than a lesion or the epileptogenic zone, but are generally related to seizure onset zones and/or areas of seizure spread (14). Discrepancies between FDG-PET and other investigations are generally associated with poorer prognosis in MRI-negative temporal lobe epilepsies (15). Detailed recent reviews are available (16).

The FDG method, with its half-life of ~2 hours, is available commercially. Thanks to the success of FDG-PET/CT in oncology, most university hospitals with epilepsy surgery programs will have access to FDG-PET. While FDG uptake can be fully quantified with an arterial input function and dynamic scanning, static scans are used in clinical practice, usually acquired for about 10–15 minutes, about 30–50 minutes after injection. It is important to supervise the patient during the uptake period to minimize the risk of seizures leading to focal hypermetabolism (13) and possibly false lateralizations. Some recommendations include EEG during this period. This may pose logistical problems, in particular with modern PET/CT scanners where electrodes cannot be left in place during scanning due to artefacts when measuring attenuation with CT. In our experience, patient observation and direct questioning about possible seizures having occurred during the uptake period eliminates much of the risk, all the more so as seizures may not always be clinically obvious, but may also be silent on scalp EEG. As always, imaging results should be considered critically, in the context of all available information, and the possibility of false lateralizations due to ictal hypermetabolism (17) should be taken into account, particularly when using measures of asymmetry (see below).

In the context of this book, one of the most important contributions of FDG-PET, analysed in conjunction with coregistered MRI, is the detection of occult focal cortical dysplasia (FCD). Detecting FCD through focal FDG hypometabolism represents a step change in MRI, changing MRI-negative to post hoc MRI-positive patients with a good surgical prognosis (7, 8). Several series have retrospectively evaluated the contribution of FDG-PET in cases where FCD was detected on histology. In one such series of 14 MRI-negative patients, FDG-PET had been unremarkable in only three, and the degree of completeness of resection of the abnormalities (on FDG-PET, SISCOM or intracranial EEG) was associated with a good postsurgical outcome (55). A larger series of 62 patients with type II FCD contained 25 with normal or doubtful 1.5T MRI findings (56). Of those 25 MRI-negative patients, FDG-PET correctly identified the location of the FCD in 21 (84%), but was negative as well in two, and falsely localizing to the orbitofrontal cortex in another two. Of note, two FCDs were only identified after PET/MRI coregistration. The 25 MRI-negative patients had an indistinguishably good postoperative outcome (88% Engel class I and 56% entirely free of all seizures) compared with the 37 MRI-positive ones (94%/75%).

The FDG-PET technique may also be useful in retrospectively revealing pathology on MRI, or by increasing confidence in the relevance of subtle possible MRI abnormalities. In a series of 31 children with 1.5T MRI that was initially thought to be noncontributive (57), FDG-PET showed focal hypometabolism in 21. A second reading of the same initial MRIs in the light of the PET findings, and with both modalities coregistered, changed the conclusion from normal MRI to subtle MR abnormalities in seven, and from subtle changes – initially considered irrelevant – to probably pathological in another two. Histological verification was only available for five patients; all had FCD.

In the context of TLE, FDG-PET has also proven useful. For example, in TLE where no decision for or against surgery could be made based on MRI and video-EEG, FDG-PET led to a decision in 72/84 cases (18). Whereas negative or inconclusive MRI was more prevalent in the subgroup where PET was useful, it is important to note that surgical outcome (63% seizure-free) for MRI-negative patients was similar to that of the entire cohort of 302 operated patients (64% seizure-free). Several studies have underlined the good prognosis of “MRI-negative, PET-positive” TLE patients (19, 20).

A peculiarity of TLE to be taken into account in the organization of epilepsy surgery programs is that the ipsilateral hypometabolism is often clearly detectable on appropriately oriented images by visual analysis, yet remains “invisible” to more objective techniques like SPM (21). The yield of voxel-based techniques may thus be higher in extratemporal lobe epilepsies, with the possible exception of children (22).

In addition, the presence of focal hypometabolism may be a stronger predictor of postoperative seizure freedom than the presence of a focal MRI lesion (16, 23), reinforcing prior studies of the usefulness of FDG-PET for predicting postsurgical outcome in patients headed for intracranial EEG (24).

Other radioligands used clinically

Many of these radioligands come from the research arena and are labeled with ¹¹C. As by definition all organic molecules contain carbon atoms, ¹¹C is very versatile for labeling a wide range of substances by replacing one of their carbon atoms, i.e., without changing their biological properties. In addition, radiation burden tends to be a little bit lower. However, with its half-life of ~20 minutes, use of substances labeled with ¹¹C is restricted to centres with a cyclotron to produce the isotope on site.

GABA_A receptors

The ligand which has perhaps found the most widespread use is [¹¹C] flumazenil (FMZ), an antagonist at the benzodiazepine site of GABA_A receptors containing the subunits alpha 1, 2, 3, or 5 (for a review, see 5, 25); GABA is the most important inhibitory neurotransmitter, and epileptogenic foci are often associated with reduced [¹¹C]FMZ binding. Such areas of reduced [¹¹C]FMZ binding are, as a rule, much smaller than areas of hypometabolism seen with FDG-PET (Figure 4.2).

Figure 4.2 FMZ (left) vs. FDG (right) decreases in the same patient: FMZ abnormalities are smaller; FMZ more clearly indicates bilateral hippocampal damage. Note both show the same parasagittal abnormality due to the patient’s brain structure.

In TLE, reduced [¹¹C]FMZ binding is correlated with hippocampal volume loss due to partial volume effects, but clinical utility in patients without MRI abnormalities was suggested by the fact that decreases are over and above volume loss (e.g., 26). In a review of the literature, 45 patients with TLE and negative MRI were identified (25). Of these, 38/45 had abnormal [¹¹C]FMZ PET scans, confirming [¹¹C]FMZ PET’s higher sensitivity compared to MRI, and in 21/45 these abnormalities were thought to be surgically useful. Unsurprisingly, surgical series showing data on postoperative patients showed a higher proportion of [¹¹C]FMZ PET scans judged useful.

In the same review, [¹¹C]FMZ PET was also more sensitive than MRI in patients with focal epilepsy of extratemporal origin and negative MRI (25). In total, 73/102 MRI-negative patients with extratemporal seizure origin identified in the literature showed abnormalities of FMZ binding. These were definitely surgically useful in 27/102 and definitely or potentially useful (e.g., preventing potentially harmful further investigations) in 54/102. Again, there was evidence of recruitment bias when comparing surgical with nonsurgical series.

There are ¹⁸F labeled FMZ analogs which would allow more widespread use due to ¹⁸F having a half-life of about 2 hours. However, only a handful of studies so far have used them, albeit with generally promising results (reviewed by 16).

GABA_A receptors are present on the majority of neurons. Hence, FMZ can also to an extent serve as a neuronal marker, particularly in the white matter (see (27) and references therein). There is a tight correlation between temporal lobe heterotopic white matter neurons and white matter [¹¹C]FMZ volume-of-distribution (VD). The presence of mainly temporal white matter [¹¹C]FMZ-VD increases in a sizeable proportion (11/18) of MRI-negative TLE cases, and of mainly periventricular increases in 7/44 MRI-negative patients with extratemporal seizure origin suggest that occult migration disturbances may be the underlying basis for this observation in a number of these cases. The presence of periventricular increases was associated with poorer outcome in two independent groups of patients with hippocampal sclerosis (HS) (27, 28). Larger control groups have recently increased the precision of surgical outcome predictions, which may now be useful for individual preoperative counseling (28).

As [¹¹C]FMZ binds to the benzodiazepine site of the GABA_A receptor, benzodiazepines given therapeutically will decrease the binding. The situation is less clear for other antiepilepsy drugs having GABA-ergic mechanisms, but these considerations add an additional constraint to using [¹¹C]FMZ PET clinically.

Radioligands used in research

Using PET allows the pursuit of several interesting avenues in the research setting.

There are efforts to develop novel ligands, e.g., for glutamate NMDA receptors (43).

Ictal/postictal/interictal comparisons have elucidated the role of neurotransmitters in stopping seizures, particularly the opioids (e.g., 44; review in 4), and are starting to show differences between patients with normal MRI versus patients with hippocampal sclerosis. Difference images were not, however, able to pinpoint the seizure onset zone in individual patients. Cannabinoid type 1 (CB1) receptors have about a tenfold higher concentration than opioid receptors. There is evidence that CB1 receptor availability increases focally after seizures (45), and postictal/interictal comparisons could in principle be used similarly to SISCOM for determining seizure onset zones (45).

There is a variety of ligands available to probe the dopamine system. Dopamine is thought to have an anticonvulsant modulatory action (see references in 4). The dopaminergic system has been studied in various syndromes, including some where MRI is typically negative, e.g., autosomal dominant nocturnal frontal lobe epilepsy. Very few patients with “MRI-negative” focal epilepsy in the sense of the focus of this book have been studied, and if so, they have not been separately analyzed due to the small number of patients in each study (46). However, recent morphological studies have uncovered that the substantia nigra is smaller ipsilaterally to the seizure focus in TLE with and without hippocampal sclerosis (47); so it is to be expected that MRI-negative patients would show PET abnormalities of their dopamine system as well.

One of the major hypotheses to explain drug resistance in epilepsy is the “transporter hypothesis,” postulating an increase of P-glycoprotein (P-GP) activity in the epileptogenic focus. P-GP is a component of the blood–brain barrier and actively removes a large number of lipophilic molecules from the brain, including several antiseizure drugs. The use of PET with labeled P-GP substrates has contributed to evidence for PGP action in the epileptogenic focus (e.g., 48, 49).

Probing receptor subtypes with heterogenous distribution in the brain will not replace the “workhorse tracers” with ubiquitous uptake that are able to probe the entire brain for abnormalities during the presurgical work-up. However, such highly selective tracers offer possibilities for investigating specific cerebral dysfunctions associated with epilepsy, such as the possible molecular basis of memory impairment in epilepsy (4).

Case example (Figure 4.3)

Figure 4.3 Multimodal imaging in a child with pharmacoresistant epilepsy. Top: axial slices. Top left, FDG-PET superimposed on MRI. Top middle, SPM analysis of FDG-PET. Top right, MEG synthetic aperture magnetometry (SAM) analysis. Middle left, SPM analysis result superimposed on to sagittal slice. Middle right, position of intracranial depth electrodes; electrodes q’ and n’ highlighted. Bottom: extract from interictal and ictal icEEG demonstrating seizure onset in q’ and n’ electrodes.

(Data courtesy of Philippe Ryvlin, IDEE, Lyon.)

Outlook and conclusions

The PET technique is only one of several useful techniques when MRI fails to reveal a potentially epileptogenic lesion; FDG-PET remains the mainstay of PET investigations and can be substantially enhanced by joint interpretation with anatomical MRI, and by additional voxel-based analysis against a control group.

Of the other tracers, 5-HT_1A tracers may be useful for lateralizing TLE, and, to a degree, further describe the TLE syndrome. However, nonlimbic epileptogenic foci are likely to be missed. FMZ may be useful in selected cases and has the advantage of a widespread cortical signal, but is not widely available and performs best with invasive quantification. AMT has low sensitivity in MRI-negative epilepsy, but may on occasion be spectacularly useful.

Depending on the clinical situation, other investigations may be preferable over PET. MEG is particularly useful when combined with spatial filtering techniques (50), but requires the presence of frequent interictal spikes (see Chapter 6). Ictal SPECT or interictal–ictal SPECT comparisons (e.g., SISCOM) depend on seizures that are frequent enough (or can be provoked), tracer availability, and seizures that are long enough for successful injection of the radioligand (see Chapter 5).

As pointed out throughout this chapter, PET requires interpretation alongside MRI findings. This is an example of a joint indication for both imaging techniques for which the novel technology of hybrid (simultaneous) MRI-PET may be extremely useful.

In addition, more quantitative analyses of MRI images and derived features like gray–white matter junction maps are becoming more widely available (see Chapter 3). Combining such feature maps with PET, particularly when adding their respective statistical comparison against control groups, should be useful and can be combined with machine-learning techniques (e.g., 47) for automatic classification either of each voxel in the image or of the patient into a certain category.