Dynamic FDG-PET in localization of focal epilepsy: A pilot study

Abstract

Epilepsy surgery remains underutilized, in part because non-invasive methods of potential seizure foci localization are inadequate. We used high-resolution, parametric quantification from dynamic 2-[18F] fluoro-2-deoxy-d-glucose positron emission tomography (dFDG-PET) imaging to locate hypometabolic foci in patients whose standard clinical static PET images were normal. We obtained dFDG-PET brain images with simultaneous EEG in a one-hour acquisition on seven patients with no MRI evidence of focal epilepsy to record uptake and focal radiation decay. Images were attenuation- and motion-corrected and co-registered with high-resolution T1-weighted patient MRI and segmented into 18 regions of interest (ROI) per hemisphere. Tracer uptake was calibrated with a model corrected blood input function with partial volume (PV) corrections to generate tracer parametric maps compared between mean radiation values between hemispheres with z-scores. We identified ROI with the lowest negative z scores (<−1.65 SD) as hypometabolic. Dynamic 2-[18F] fluoro-2-deoxy-d-glucose positron emission tomography ( found focal regions of altered metabolism in all cases in which standard clinical FDG-PET found no abnormalities. This pilot study of dynamic FDG-PET suggests that further research is merited to evaluate whether glucose dynamics offer improved clinical utility for localization of epileptic foci over standard static techniques.

1. Introduction

Non-invasive identification of a clear seizure focus remains elusive in some patients. This may contribute to underutilization of surgery, even though guidelines recommend its use in those with medically intractable focal epilepsy [1].

2-[18F] fluoro-2-deoxy-d-glucose positron emission tomography (FDG-PET) is an important component of the non-invasive presurgical localization [2–4] because it measures neuronal metabolism rather than anatomy, as is done by MRI [5], or electrical activity as is done by EEG. When measured interictally, metabolically hypoactive regions revealed by decreased glucose uptake correspond to seizure foci. Focal hypometabolism predicts better outcomes when concordant with other data [3]. When localizing information is discordant in the setting of normal MRI, FDG-PET aids in triage; for example, 63% of patients with normal MRI and discordant data had focal abnormalities on FDG-PET; of those, 41% went on to have invasive monitoring or epilepsy surgery guided by FDG-PET [2].

But, traditional “static” FDG-PET (sFDG-PET) has limited sensitivity when it fails to indicate a clear hypometabolic focus. For example, in a study of presurgical tests, identification of seizure foci with sFDG-PET predicted one-year seizure-remission with an odds ratio lower than other methods such as ictal single positron emission computed tomography, the presence of pre-operative auras, or concordance among presurgical procedures [6].

We developed a method of dynamic FDG-PET (dFDG-PET). Whereas sFDG-PET measures cerebral metabolism within a single time window of about 15 min, dFDG-PET maps radiation over a consecutive series of time windows for about 60 min. Analysis deconvolutes dynamic concentrations into a series of timedependent “snapshots” that reveal rates of glucose uptake rather than final absolute uptake.

The purpose of this early phase pilot study of dFDG-PET in patients with medically intractable focal epilepsy was to provide proof-of-concept to guide phase 2 study design.

2. Methods

2.1. Patients

This was an intrasubject comparison of hypometabolic lesions revealed by dFDG-PET in those whose clinical sFDG-PET images were normal. In this IRB-approved study, consented patients were enrolled if they were (1) age >17 years; (2) were reviewed in the University of Virginia’s Epilepsy Surgery Conference and determined to have medically intractable focal epilepsy; (3) had normal sFDG-PET as reviewed in the epilepsy surgery review process. Exclusions were (1) inability to obtain dFDG-PET without sedation; (2) diabetes mellitus; (3) psychogenic nonepileptic seizures or idiopathic generalized epilepsy; (4) weight >226 kg; and (5) implantation of a responsive neurostimulator.

A proposed localization and committee recommendations were determined from Epilepsy Surgery Conference. Results of dFDG-PET were not available to clinical reviewers.

2.2. dFDG-PET

Dynamic 2-[18F] fluoro-2-deoxy-d-glucose positron emission tomography ( studies were obtained on Siemens Biograph TOF mCT scanner [7] (Fig. 1). Dynamic acquisition consisted of an intravenous ~10 mCi tracer injection over 10 seconds with initiation of a 60-min scan in list-mode format (Fig. 1A). PET was preceded by a high resolution T1-weighted MPRAGE MRI (256 pixels × 256 pixels × 192 slices) using a Siemens 3T scanner for co-registration (Fig. 1B).

Fig. 1.

Steps in dynamic FDG-PET image acquisition, pre-processing, and parametric map creation. Dynamic PET, high-resolution MR and Destrieux Atlas defined on the MR brain template form the raw inputs (A–D). The motion corrected dynamic PET data are Co-Registered with patient MR followed by co-registration of MR image with a high-resolution T1-weighted MR brain template provided by the Montreal Neurological Institute using nonrigid transform (E). Model corrected blood input function (MCIF) is then computed from the internal carotid arteries for all patients (F). Each voxel of dynamic PET data along with blood input is independently fed into a graphical Patlak model to compute parametric Ki maps using linear regression and subsequently z score maps (G). The co-registered template and atlas was used to bin individual voxels of the generated uptake maps for regional analyses (H).

Subsequent processing was performed with custom tools developed in Matlab (Mathworks Inc., Natick, MA). Image pre-processing started with motion correction for the 60-min acquisition (Fig. 1C). PET data (400 pixels × 400 pixels × 111 slices × 38-time frames) were averaged across the first 14 frames to create a reference used to perform a rigid body transform across the 38 frames. The averages of all the motion-corrected PET frames were resliced and co-registered with T1-weighted MRI using nonrigid transform to generate a transformation matrix used, in turn, to generate a co-registered dynamic PET volume. Next, the MRI was co-registered with a high-resolution T1-weighted MRI template provided by the Montreal Neurological Institute (MNI) [8] (Fig. 1D) using a nonrigid transform, and a transformation matrix was generated. The total 164 regions of the Destrieux atlas [9], defined on the same MR brain template, were binned to generate 36 regions of interest (ROI) (18 regions/side). The above transformation matrix was inverted and applied to all ROIs to move them from the standard MNI template into the patient MRI (Fig. 1E). All the above processes were performed using MRtrix functions [10]. ROIs were then dropped onto the parametric maps (voxel-by-voxel maps) generated from dFDG-PET images.

Objective parametric PET maps were generated from model corrected blood input function (MCIF) corrected for partial volume (PV) averaging and spill-over (SP) contamination, using our technique developed for rodent brains [11]. Briefly, image-derived blood input function (IDIF) from internal carotid artery in an early time frame (Fig. 1F) for each patient was computed from the cohort average of 4 ROIs of the left internal carotid artery. These ROIs were applied to all the motion-corrected 38 PET frames to generate blood time activity curves (PET_IDIF). A model IDIF correcting the blood input for PV and tissue SP contamination can be written as

$$Mode{l}_{IDIF,i}=\frac{{\int }_{t_b^i}^{t_e^i}\left[S_{Tb}{C}_T\left(t\right)+r_b{C}_a\left(t\right)\right]dt}{t_e^i-t_b^i}$$

in which

S_Tb = SP contamination from the tissue to the blood at late time points

r_b = blood recovery coefficient

t_b and t_e = beginning and end of a time frame.

C_T(t), the model tissue, was obtained by solving FDG transport differential equations from blood to tissue spaces. C_a(t) is 7-parameter model blood for FDG transport as described [12,13]. The above model IDIF was optimized using the following objective functions [11]:

$$O_1\left(p\right)=\sum _{i=1}^n{\left(Mode{l}_{IDIF,i}-PE{T}_{IDIF,i}\right)}^2$$

$$O_2\left(p\right)={\left(ModelPea{k}_{IDIF}-PETPea{k}_{IDIF}\right)}^2$$

$$O\left(p\right)=O_1\left(p\right)+O_2\left(p\right)$$

ModelPeak was computed from the model equations for the IDIF (Model_IDIF) (Eq. (1)). PETPeak values were derived from the dynamic PET blood images for each patient. Optimization of O(p) used non-linear regression analysis yielding the estimate of MCIF.

Each voxel of the dynamic data was then independently fed into a graphical Patlak model 11, together with the computed PBIF to compute whole-brain parametric Ki and z-score maps (Fig. 1G) applied to ROIs. The Patlak model performed a linear regression (starting at 10 min and beyond for which the image data were linear) where the slope provided a measure for the rate of FDG uptake at that voxel. Parallelization of multicore high-performance computers was used to compute whole-brain parametric maps for all patients. Data analysis was performed with (Fig. 1G) z-score parametric mapping and (Fig. 1H) segmented into ROI.

To monitor for seizures during acquisition, patients were monitored with video-EEG obtained with MRI-compatible scalp electrodes (Ives EEG Solutions, Manotick, Ontario, Canada).

2.3. Data analysis

Whole-brain voxel-level rate of FDG uptake (Ki) was computed from the motion-corrected, co-registered dFDG PET data for each patient. Ki maps for each patient were converted to voxel level z-score maps by normalizing to the whole-brain mean and standard deviation within each patient. Average regional z-scores were computed for 36 ROI (18 regions/side). All regions with z-scores less than −1.65 standard deviations (SD), dictated by the default z-score threshold in MIM (https://www.mimsoftware.com/), were identified as hypometabolic compared to its contralateral side.

3. Results

Nine patients underwent dFDG-PET without complications (Table 1). One patient (patient 2) was excluded because of a seizure during dFDG-PET acquisition. Technical problems prevented another (patient 5) from obtaining MRI; data are presented for subjective review only.

Table 1.

Patients in pilot study of dFGD-PET. Phase 1 (non-invasive) localizing procedures are presented in the top table, and the proposed localization, the location of dFDG-PET findings, and clinical outcomes in the duration between the dFDG-PET scan and last follow-up. Tmp = temporal; SOZ = clinical localization of phase 1 epilepsy surgery conference; P2 = phase 2 evaluation with intracranial EEG; LITT = laser interstitial thermal therapy; SUDEP = sudden unexpected death in epilepsy. MRI: Magnetic Resonance Imaging. SPECT: Single Photon Computed Tomography.

ID	Age/Sex	MRI brain	Interictal EEG	Ictal EEG onset	Interictal SPECT	Ictal SPECT
1	28/M	normal	L tmp slow	L tmp	normal	normal
3	52/M	normal	L anterior tmp spikes	L tmp	normal	L tmp
4	60/M	R > L MTS, infarct L occipital	R tmp spikes, L tmp slow	R tmp	L tmp	R tmp
5	35/M	R MTS (subtle)	L slow	n/a	n/a	n/a
6	60/M	normal	L mid-tmp spikes	n/a	R medial tmp lobe	n/a
7	44/M	R frontal encephalomalacia	R + L tmp spikes, R slow	Broad, non-localizing	R frontal	n/a
8	29/M	normal	normal	R hemisphere broad	R tmp-parietal	n/a
9	43/F	2 mm anterior midline meningioma	R tmp spikes	Broad, non-localizing	normal	R frontal
ID	Age/Sex	Proposed SOZ	Clinical Read, Static PET of Dynamic PET	dFDG-PET	F/U (m)	Outcome
1	28/M	L tmp	No seizure focus	L tmp	14	P2 declined, lost to f/u
3	52/M	L tmp	L Lateral Tmp	L tmp	40	P2 scheduled
4	60/M	R tmp	R Tmp mesial, ant and Lateral	R Hippocampus, L Tmp Mesial	39	LITT R hippocampus, seizure free 30 months
5	35/M	L vs R tmp	No seizure focus	N/A	23	P2 declined, SUDEP 29 months
6	60/M	L vs R tmp	L ant Tmp	R tmp	34	P2 declined
7	44/M	R frontal, L vs R tmp	No seizure focus	R Tmp	33	P2, R frontal rsxn, repeat P2 multifocal SOZ
8	29/M	L vs R frontal	No seizure focus	R Tmp	29	P2 declined
9	43/F	R > L tmp frontal	No seizure focus	R Tmp	18	P2 declined

Supplementary Tables 2, 3, and 4 document partialPV factors and SP contamination, FDG-uptake rates (Ki), and z-score intrahemispheric for all patients, respectively. Supplementary Fig. 3 illustrates the process of calculation of model IDIF and MCIF which in turn allows for the semi-automated computation of Ki maps for each patient.

Table 1 and Fig. 2 show dFDG-PET results in the context of localization provided by Epilepsy Surgery committee review. Focal hypometabolism (<−1.65 SD) was identified in all 7 patients with quantitative results. Patient 5 had similar subjective hypometabolism on visual review (Supplementary Fig. 4). Patients fell into three groups according to relationships between proposed localization and dFDG-PET.

Fig. 2.

Regional z score maps of dFDG-PET co-registered on MRI for all 7 patients indicating temporal lobe hypometabolism. Inclusion criteria required normal standard sFDG-PET.

1. Patients 1, 3, 4, and 9 had unilateral mesial temporal/hippocampal regions of hypometabolism on dFDG-PET concordant with anticipated localization. Patient 4 underwent laser interstitial thermal therapy of the ipsilateral hippocampus and was seizure free for 30 months at last follow-up. Patients 1, 3, and 9 declined further surgical consideration.

2. Patient 7 had previous right frontal lobectomy that did not improve seizure frequency. Intracranial monitoring disclosed multifocal seizure foci including the right mesial temporal region concordant to dFDG-PET results.

3. Patients 5, 6, and 8 had dFDG-PET findings indicating unilateral mesial temporal hypometabolism in the context of bilateral mesial foci proposed by the Epilepsy Surgery Committee review. Each declined intracranial monitoring; patient 5 died in follow-up (sudden unexpected death in epilepsy, SUDEP).

4. Discussion

In this early-phase study of focal epilepsy surgery candidates for whom sFDG-PET showed no abnormalities, dFDG-PET found focal regions of hypometabolism in all cases. No patients had adverse events related to the study. This pilot study of dFDG-PET suggests that further research is merited to evaluate the specificity and sensitivity of dFDG-PET in larger cohorts to determine whether glucose uptake dynamics offer improved localization of epileptic foci over standard static PET techniques.

This work builds on innovations around dFDG-PET and explored in rodent brains [11] and models of movement disorders [14], human ambulatory metabolic studies [15], and brain cancer [16]. A critical and innovative step is our development of a blood input function that calibrates the calculation of FDG uptake. In animals, direct measurement through arterial blood or in non-invasive estimates that derive blood input function from images of left ventricular blood are possible. In human brain imaging, however, the requirement for non-invasive estimation within a limited image field led to our technique of estimations via images of the internal carotid arteries with model based PV corrections [11]. Our other development was the “turn-key” application of automated movement-correction software required over the 1-h scan time, a process in the past that was largely performed manually.

The goal of this pilot study was to provide initial data to drive further work in validation of dFDG-PET. This pilot study suggests that dFDG-PET may indicate focal regions of hypometabolism in epilepsy surgery subjects whose standard static FDG-PET was unhelpful and whose non-invasive, “phase 1” evaluations did not disclose clear or single surgical targets. Our plan is to design trials that will calculate specificity and sensitivity of dFDG-PET to the “ground truth” of seizure freedom following epilepsy surgery. We note that in the one patient who underwent successful surgery (patient 4), dFDG-PET predicted the surgical target. In contrast, in the one patient who had prior unsuccessful surgery (patient 7), dFDG-PET indicated a different surgical target than the one undertaken (subsequent intracranial monitoring disclosed probable multifocal foci).

Dynamic 2-[18F] fluoro-2-deoxy-d-glucose positron emission tomography ( may offer additional sensitivity over sFDG-PET to reveal epileptic networks since it captures the kinetics of glucose wash-in, metabolism, and wash-out from the point of injection. An advantage of our technique is that it is potentially transportable to any facility with an appropriate PET scanner, a possible transformative method to allow current hardware to improve non-invasive localization. Further software development is needed to create a complete imaging package that can be implemented among different centers.

The limitations of an early phase study include small numbers of patients who may or may not precede with epilepsy surgery; thus, we were unable to confirm the localization of epileptic foci uniformly through the “gold standard” outcome of seizure remission through resection of identified regions, except in the one patient (patient 4) in whom dFDG-PET predicted the surgical target and who experienced long-term seizure remission after laser ablation. Therefore, the specificity of our technique remains to be evaluated.

5. Conclusion

In conclusion, this early phase trial has demonstrated that dFDG-PET may offer non-invasive localization of potential epileptic foci. With further validation, dFDG-PET may offer more patients the advantages of presurgical localization and possibly convert those who may be non-candidates into candidates for transformative epilepsy surgery.

Abbreviations:

sFDG-PET

standard 2-[18F] fluoro-2-deoxy-D-glucose positron emission tomography

dFDG-EPT

dynamic 2-[18F] fluoro-2-deoxy-D-glucose positron emission tomography

ROI

regions of interest

MCIF

model corrected blood input function

PV

partial volume averaging

SP

spill-over contamination

IDIF

image-derived blood input function

Ki

FDG-uptake rate in brain tissue