Introduction

Epilepsy is one of the most common and serious chronic cerebral disorders, which affects over 70 million people worldwide [1, 2]. Almost 80% of patients with epilepsy reside in low- and middle-income countries or districts [3]. Among infants < 1 year old and people > 50 years old, the performances of epilepsy which appear a bimodal distribution with two peaks are more implicated than people in other age groups [4, 5]. Clinically, epileptic seizures vary widely in manifestation from abnormal sensations to motor symptoms.

Epilepsy is a multi-symptom disease with complex risk factors and in many cases has a strong inherited tendency, instead of a situation with a single cause and a single expression [3, 6]. For decades, the etiology, pathophysiology, and antiepileptic drugs (AEDs) are continually being explored and investigated [2, 7,8,9]. Currently, AEDs are effective in only about two-thirds of patients in developed countries, and despite being available of more than 25 medications worldwide, only a few of them are considered first-line [10]. Furthermore, epilepsy surgery is considered to be the most efficacious way to attain long-term seizure freedom, but it has been confined to individuals with drug-resistant epilepsy and still underused [11,12,13,14].

Nowadays, the insufficient understanding and absence of specific biomarkers of the epileptogenic process are the major constraints in the research and development of new AEDs that are able to prevent the underlying disease or improve prognosis [15]. The rapid advances in neuroimaging modalities have expanded our chances to investigate the disease of epilepsy by means of noninvasive research modalities. The field of neuroimaging has been remarkably developed in recent years. Neuroimaging techniques used in clinical settings for the assessment of patients with epilepsy include but not limited to non-contrast computerized tomography (CT), structural and functional MRI (fMRI), electroencephalograph combined with fMRI (EEG/fMRI), MR spectroscopy (MRS), positron emission tomography (PET), single-photon emission computed tomography (SPECT), and magnetoencephalography (MEG) [16]. The utilities of these modalities depend on specific circumstances and clinical questions to be addressed [17]. Currently, chronic intracranial EEG monitoring remains the gold standard in defining epileptic foci, but it also has boundedness and its application is highly dependent on other localization information [18, 19].

PET is manifested to be an available noninvasive method to guide intracranial electrode placement, and it can also reduce the number of patients requiring invasive EEG [20]. PET also plays a very important role in the evaluation of epilepsy. The most commonly used PET tracer in epilepsy is [18F]2-fluoro-2-deoxy-d-glucose (18F-FDG), usually performed in the seizure-free interval and aimed at the identification of cerebral regions with decreased glucose metabolism; it is considered to partially reflect the reduction of synaptic activity [21]. However, the epileptic areas are commonly smaller than the hypometabolic regions [18, 22, 23]. Besides 18F-FDG-PET, the GABA-A receptor ligand 11C-flumazenil (11C-FMZ) has displayed promising results in epileptic foci localization and lateralization [15, 24, 25]. Other potential PET tracers for detecting epileptic regions include [11C]a-methyl-L-tryptophan (AMT) which detects tryptophan metabolism, most recently 5-hydroxytryptamine type 1A (5-HT-1A) receptor ligands, and other radioligands that bind to opioid, histamine, N-methyl-d-aspartate (NMDA), “peripheral benzodiazepine” or acetylcholine receptors. Although the clinical application of most of them in epilepsy has not been systematically established, they have very important research value.

However, to our knowledge, only a few studies limited to small sample size directly compared the clinical performance of 18F-FDG-PET or 11C-FMZ PET compared with EEG or surgical outcome in localization of epileptogenic areas. We firstly did a meta-analysis by collating the available evidence to generate a precise estimation of the clinical utility of 18F-FDG-PET and 11C-FMZ-PET for the localization of the epileptogenic foci in patients with epilepsy, and secondly to provide evidence or clues for practitioners’ clinical decision-making and practice.

Methods

Search strategy and selection criteria

This meta-analysis was performed on the basis of the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) guidelines [26]. The research question of this study was raised in terms of PICOS, populations (participants with diagnosed epilepsy), interventions (FMZ or FDG-PET or coregistration of FDG-PET and MRI), comparators (EEG or surgical outcomes), outcomes (absolute numbers of participants with true-positive (TP), false-positive (FP), true-negative (TN), and false-negative (FN) results), and study designs (retrospective and prospective studies). The search strategy was restrictively based on the pre-designed protocol. We did a systematic search on PubMed and Embase to assay studies from inception to May 31, 2020, with articles in English considered. The following search terms were used: (11C-flumazenil OR 18F-FDG) AND (epilepsy OR epilepsies OR seizure disorder OR seizure disorders OR cryptogenic epilepsy) AND ((positron emission tomography) OR PET). We supplemented the online search with manual screen of the reference lists of all primary studies as well as relevant review articles. We considered studies using FMZ or FDG-PET or coregistration of FDG-PET and MRI for the assessment of patients with epilepsy. Inclusion criteria were as follows: FMZ or FDG-PET or FDG-PET/MRI used in participants with diagnosed epilepsy; use of EEG or surgical outcomes as the gold standard to evaluate diagnostic performance; corresponding outcomes such as concordance rates of PET or PET/MRI scan compared with the reference standard, absolute numbers of participants with TP, FP, TN, and FN results via FDG or FMZ PET. If studies recruited participants over the same period of time or from the same study center, only the research with the largest sample size or yielding the most pertinent outcomes was included to avoid duplications. Both retrospective and prospective studies were considered. Studies in abstract form, case reports, and successive cases seen in a unit were excluded.

Two independent investigators (Haiqun Xing and Meiqi Wu) conducted the process of literature search and study inclusion. When disagreement occurred, they discussed their arguments, and a third reviewer (Na Niu) was involved in case that no consensus was achieved.

Data extraction and quality assessments

Data were extracted from each selected publication by two investigators (Yanru Ma and Yimin Liu) independently. The following information were recorded: name of the first investigator, year of publication, number of participants, duration of epilepsy, age, gender ratio, type of epilepsy, reference standard, concordance rates of PET or PET/MRI compared with the reference standard, TP, FP, TN, and FN. To assess the methodological quality of the included studies and risk of bias and applicability concerns, we used the checklist of QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. This tool contains components in terms of patient selection, index test, and reference standard, as well as flow and timing.

Statistical analysis

All analysis was performed at the study level with the Stata15.0, R4.0.2 software and Review Manager 5.3 software. p < 0.05 was considered to be statistically significant. We calculated pooled concordance rates, sensitivity, specificity, positive likelihood ratio (LR) and negative LR, odds ratio (OR) with their respective 95% confidence intervals (CIs), and area under the summarized receiver operating characteristic (sROC) curves (AUCs). We used the Cochran Q and the I2 statistics to evaluate the heterogeneity of results between studies included. I2 values of 0–25%, 25–50%, 50–75%, and 75–100% indicate insignificant, low, moderate, and high heterogeneity, respectively. We created funnel plots to assess publication and related bias. Deeks’ method was used to statistically check the asymmetry of the funnel plot and detect publication bias. Moreover, subgroup analysis was performed to explore the potential sources of heterogeneity of different studies and influence analysis was used for the detection of outliers (studies) which affected the pooled results statistically.

Results

Study selection and characteristics

A total of 606 articles were identified from the databases searched. One hundred seventy duplicates were removed and 347 studies were excluded through an initial screening. After a full-text assessment for eligibility of the remaining 89 articles, 44 studies were identified for inclusion in this meta-analysis. No additional studies were found through the screening of references of the included full-text articles (see Fig. 1). The selected 44 studies containing a total of 2246 patients with diagnosed epilepsy. These articles were published from 1995 to 2020. More details of the studies included are shown in Table 1.

Fig. 1
figure 1

Search results and flow chart of the meta-analysis

Full size image
Table 1 Study characteristics
Full size table

Comparison between PET and reference standard

The pooled concordance rate of FDG-PET compared with the reference standard was 0.67 (95% CI: 0.60–0.73); as for FMZ-PET, the concordance rate was 0.75 (95% CI: 0.57–0.93). Concordance rate for FDG-PET/MRI coregistration was 0.93 (95% CI: 0.89–0.97) (see Figs. 2, 3, and 4). Subgroup analysis revealed that 18F-FDG-PET showed the highest concordance rate in children 0.84 (0.75; 0.92); 11C-FMZ had the highest concordance rate in the subgroup of adults 0.92 (0.74; 1.00). In patients with TLE, the pooled concordance rate of 18F-FDG-PET was 0.79 (0.63; 0.92) (see Table 2).

Fig. 2
figure 2

Forest plot of pooled concordance rate for FDG-PET compared with the reference standard. The pooled concordance rate of FDG-PET was 0.67. Heterogeneity was high and statistically significant

Full size image
Fig. 3
figure 3

Forest plot of pooled concordance rate for FMZ-PET compared with the reference standard. The pooled concordance rate of FMZ-PET was 0.75. Heterogeneity was high and statistically significant

Full size image
Fig. 4
figure 4

Forest plot of pooled concordance rate for PET/MRI coregistration compared with the reference standard. The pooled concordance rate of PET/MRI coregistration was 0.93. Heterogeneity was statistically insignificant

Full size image
Table 2 Subgroup analysis for the pooled concordance rate of 18F-FDG-PET, 11C-FMZ-PET, and FDG-PET/MRI
Full size table

Diagnostic performances of 11C-FMZ and 18F-FDG PET

Data from the 12 studies were used in the bivariate mixed-effects regression model to compute the pooled results on the basis of threshold analysis on FDG-PET (p = 0.67) and 7 studies with respect to FMZ-PET (p = 0.94) were analyzed. FDG-PET demonstrated an overall sensitivity of 0.66 (95% CI: 0.58–0.73) and specificity of 0.71 (95% CI: 0.63–0.78), with an AUC of 0.71 (95% CI: 0.67–0.75), positive LR of 2.3 (95% CI: 1.7–3.0), negative LR of 0.48 (95% CI: 0.38–0.61), and diagnostic OR of 5 (95% CI: 3–8) for the localization of epileptogenic zone in patients with epilepsy. FMZ-PET showed an overall sensitivity of 0.62 (95% CI: 0.49–0.73) and specificity of 0.73 (95% CI: 0.59–0.84), with an AUC of 0.71 (95% CI: 0.67–0.75), positive LR of 2.3 (95% CI: 1.3–4.0), negative LR of 0.52 (95% CI: 0.36–0.77), and diagnostic OR of 4 (95% CI: 2–11) (Fig. 5). The results of subgroup analysis manifested that there was no statistical significance in different subgroups of 18F-FDG-PET performance. 11C-FMZ-PET showed better diagnostic performance in the TLE subgroup and adults, respectively. Besides, pooled sensitivity and specificity for FDG-PET were 0.67 (0.55–0.79) and 0.76 (0.64–0.87) (Table 3).

Fig. 5
figure 5

Forest plots of studies included in the meta-analysis for localization of epilepsy with 18F-FDG PET (a) and 11C-FMZ PET (b). FDG-PET demonstrated an overall sensitivity of 0.66 and specificity of 0.71; heterogeneity for pooled sensitivity and specificity was statistically insignificant. FMZ-PET demonstrated an overall sensitivity of 0.62 and specificity of 0.73; heterogeneity for pooled sensitivity and specificity was statistically insignificant

Full size image
Table 3 Subgroup analysis for the diagnostic performance of 18F-FDG-PET and 11C-FMZ-PET
Full size table

Heterogeneity and quality of studies

A forest plot showed no heterogeneity for the sensitivity of 18F-FDG PET (Cochran Q = 1.82, p = 0.38, I2 = 6.9%), and there was no heterogeneity for the specificity of 18F-FDG PET (Cochran Q = 11.65, p = 0.38, I2 = 5.6%). As for the results of 11C-FMZ PET, no significant heterogeneity of sensitivity (Cochran Q = 9.13, p = 0.17, I2 = 34.3%) and specificity (Cochran Q = 9.85, p = 0.13, I2 = 37.1%) were found (see Fig. 5).

Quality assessment by QUADAS-2 scale showed that 37 studies had low risk of bias for patient selection, 2 studies had high risk of bias, and 5 studies had unclear risk of bias. Thirty-one studies had low risk of bias for index test, 3 studies had high risk of bias, and 10 studies had unclear risk of bias. Thirty-seven studies had low risk of bias for reference standard and 7 studies had unclear risk of bias. Clinical applicability concerns of each study included were also evaluated (Supplementary Figures 1 and 2).

The results of subgroup analysis demonstrated statistical difference between subgroups of population in the concordance rate analysis for FDG PET (p = 0.001). As for FMZ PET, difference was present in subgroups of duration and population (p = 0.009, p = 0.011). Subgroup analysis of diagnostic performance of 11C-FMZ-PET showed statistically significant difference in the TLE patients and adults (see Tables 2 and 3).

Influence analysis showed that no single study had significant impact on the overall outcomes in all pooled analysis.

Publication bias

Deeks’ funnel plot asymmetry tests yielded a p value of 0.59 for 18F-FDG PET and a p value of 0.24 for 11C-FMZ PET, respectively (see Fig. 6).

Fig. 6
figure 6

Funnel plots for the pooled analysis of 18F-FDG PET (a) and 11C-FMZ PET (b) in the localization of epilepsy. Deeks’ funnel plot asymmetry tests yielded a p value of 0.59 for 18F-FDG PET and a p value of 0.24 for 11C-FMZ PET

Full size image

Discussion

PET is considered to be a neuroimaging technique that provided satisfactory insights into the molecular functioning of the brain in a living human [27]. 18F-FDG was developed to assess brain glycometabolism, which led to the original application of PET in epilepsy [28]. Although 18F-FDG is widely used in patients with epilepsy, unfortunately, it is not a desired tracer: the distribution of glucose hypometabolism regions is not related accurately to the level of hippocampal sclerosis, as results from MRI or histopathological test [29]. Since the 1990s, plentiful researches have assayed the implementation of 11C-FMZ PET in the field of epilepsy [25]. Interestingly, it was demonstrated that the increase of seizure frequency correlated inversely with the tracer intake in the frontalis piriform cortex in an 11C-FMZ PET study in epilepsy patients [30]. The site was inconsistent with the location of seizure onset and has been consistent with the results of morphometric MRI and EEG-fMRI researches [31]. Nevertheless, the tracer has not seen widely clinical utility, mainly due to its short half-life.

For decades, endeavors have been made to compare the utility of 18F-FDG and 11C-FMZ in the localization of epileptogenic zone [18, 32,33,34,35,36,37]. Nonetheless, the conclusions were less powerful to be extrapolated to clinical practice due to limited sample sizes and heterogeneity of different studies. We performed a meta-analysis to generate a more precise effect size of performances of FDG and FMZ in epilepsy localization and to provide a convincing evidence for healthcare professionals in counseling patients with epilepsy.

In this meta-analysis, we did a detailed literature search to improve the potential to capture all relevant studies as we can. Data extraction was conducted by two independent investigators using a pre-designed form. Furthermore, we assessed the heterogeneity between studies included, source of heterogeneity, quality of each study, and publication bias. The quality of individual study included in this meta-analysis was evaluated as high according to the QUADAS-2 scale. Furthermore, no significant heterogeneity between studies was observed; the analysis of threshold indicated that the results of studies can be consolidated with sufficient reasons. The influence analysis showed no single study had significant impact on the overall results. However, as we performed subgroup analysis, the results revealed that as for 11C-FMZ-PET, there was statistically significant difference in the TLE patients and adults. This contradiction may be attributed to the small numbers of studies in subgroups. Deeks’ funnel plot asymmetry tests indicated that publication bias may not affect the results between studies in either 18F-FDG or 11C-FMZ PET. This meta-analysis showed higher sensitivity of FDG-PET over FMZ-PET for the localization of epilepsy; however, the specificity of FDG-PET was lower than that of FMZ-PET. With respect to the type of epilepsy, both modalities showed better performance in patients with TLE compared with other epilepsy types. In consideration of clinical applications, given the positive and negative LRs of FMZ-PET (2.3 (95% CI: 1.3–4.0) and 0.52 (95% CI: 0.36–0.77)), it might help in excluding and confirming the localization of epilepsy foci in contrast to relevant indicators of FDG-PET. With respect to concordant detection results compared with reference standards (EEG/surgery outcome), FDG-PET/MRI coregistration reached a concordance rate of 93%; FMZ-PET manifested superior concordance rate over FDG-PET imaging. FMZ-PET showed higher concordance rate than FDG-PET.

Nevertheless, there are limitations in this meta-analysis. All analysis was performed at the study level, so we were unable to extract information at individual level based on the information in each study. Therefore, the division of subgroups for age of participants and the duration of epilepsy was deemed to be less specific. Although we detailed the inclusion and exclusion criteria, heterogeneity in studies still existed. Even though the subgroup analysis was conducted, the corresponding interpretation should be made with caution.

On the basis of our findings in this analysis, we may conclude that both 11C-FMZ PET and 18F-FDG PET can provide helpful complementary information for the localization of epileptogenic zone, especially when combined with other noninvasive technologies such as MRI. Interestingly, the recent development of 18F-FMZ, an alternative tracer of 11C-FMZ, might overcome the issues including short half-life [38]. Further assessments of potentially powerful tracers such as 18F-FMZ are needed.