Optimizing statistical parametric mapping analysis of ¹⁸F-FDG PET in children

Visual inspection of 18F-fluoro-2-deoxy-d-glucose-positron-emission tomography (FDG-PET) images is extensively used in pediatric clinical practice, but it remains subjective and depends on the expertise and experience of the observer. By contrast, statistical parametric mapping (SPM) is an objective and therefore observer independent method of analysis [1, 2], but it requires comparison to a control data set that is quite impossible to obtain in healthy children for ethical reasons. Based on FDG-PET in pediatric epilepsy, the present study proposes to generate and validate a pseudo control group of pediatric patients for SPM studies in children.

PET with 18F-fluoro-2-deoxy-d-glucose or FDG-PET currently plays a key role in the investigation and management of patients with refractory focal epilepsy, particularly when surgery is a therapeutic option [3, 4]. The use of FDG-PET over the past decades has shown that regions of interictal hypometabolism are strongly associated with the seizure onset zone (also called the epileptogenic zone) in adults as in children, although the hypometabolism may often extend beyond this zone [5–7] or occasionally miss it [8]. Findings are usually analyzed visually, and this is considered to carry powerful detecting value, although it is highly variable according to the location (temporal/extratemporal) and type (positive/negative MRI) of epilepsy [9–11]. PET/MRI coregistration also impacts visual detection [12–14]. SPM proved to be a useful strategy for FDG-PET in adults with refractory focal epilepsy not only in temporal lobe cases [5, 7, 15–19], but also in extratemporal epilepsy and/or negative MRI, where sensitivity of visual analysis is lower [20]. For instance, in MRI-negative frontal lobe epilepsy patients, SPM sensitivity was equivalent or superior to visual analysis [20, 21]. As a result SPM is more and more used as a complementary procedure to study focal epilepsy in adults.

By contrast, SPM in children with epilepsy remains limited to a few teams [22–26] and provides discordant results [14, 27]. SPM procedure requires image normalization to a template before a statistical comparison is done to a control group. Because of ethical constraints regarding PET in normal children, pediatric SPM application most often has to use an adult template and adult controls, but such a procedure has major limitations. One issue is the difference in size of the head of adults and children which cannot be solved with spatial correction and requires the use of a pediatric template for children below the age of 6 years [23]. Another issue is the major age-related changes of cerebral glucose metabolism, in cortical as in the subcortical structures, throughout the entire childhood [28, 29]. Regional metabolism increases from birth to around 4 years up to values over twice the adult values, which were maintained until the end of the first decade, when they began to decline and reach adult rates by the end of the second decade. Metabolism increases earlier in the sensorimotor and occipital cortex than in the frontal cortex and increases higher in the cerebral cortex than in subcortical structures and cerebellum [28]. In the anterior cingulate cortex and the thalamus in particular, metabolism continues to increase up to 25 years and then remains relatively stable [29]. Such widespread hypometabolic changes related to the maturation of cortical and subcortical structures may complicate the interpretation of SPM results in children when compared to adults (is a given hypometabolism pathological or physiological?) and may decrease the sensitivity of detection of hypometabolic areas using SPM. These issues point out the need of an age-matched control group.

Despite continuous debate, ethical constraints prevent investigating healthy children as control subjects for PET studies. To our knowledge, only two authorizations have been delivered up to now, and they did not include patients under 6 years of age because they usually need sedation to be scanned [30, 31]. As an alternative, populations of pediatric patients had therefore to be used as control groups. A first option was to focus on children with another pathology and to test the difference with the pathology to be studied. Using this approach, Zilbovicius et al. succeeded identifying bitemporal dysfunction with H₂0¹⁵ PET in autistic children compared to children with idiopathic mental retardation [32], and Juhasz et al. identified a medial prefrontal hypometabolic network with FDG-PET in epileptic children with aggressive behavior compared to non-aggressive ones [33]. However, such control groups cannot be extended to other studies since they provide PET images with pathology-related anomalies. A second option was to generate pediatric control groups from patients with normal PET images. Two historical studies provided FDG-PET imaging from respectively sick children considered retrospectively as normal and pediatric idiopathic focal epilepsy, thus permitting to establish the maturational profile of cortical and subcortical metabolism, but the data are not compatible with current PET machines [28, 34]. The present study provides compatible data from the epilepsy field.

The aim of this study was to evaluate a pediatric age-paired pseudo-control group compared to a classical adult control group in order to optimize the use of SPM analysis in children with refractory epilepsy.

We retrospectively studied the 71 consecutive children referred for FDG-PET examination at our institution during a 4-year period. Inclusion criteria were the following: diagnosis of refractory focal epilepsy, previous extensive work-up including video-EEG (electroencephalogram) recordings and MR imaging, and age over 2 years (based on recruitment). The protocol was approved by the institutional ethical standards committee on human experimentation, and written informed consent was obtained from all guardians of patients participating in the study.

We defined two groups according to MRI and PET analysis (using both visual and SPM analysis, see Methods) (Figure 1). Group I (pseudo-control group) comprised the 24 children with negative MRI and negative (normal) PET; group II (patient group) comprised the remaining 47 children with positive (abnormal) PET (20 had positive MRI, 27 had negative MRI). The clinical characteristics of groups I and II were similar regarding age, sex, and lateralization and location of the SOZ (Table 1). They were compared to an in-house adult control group (21 healthy adults, aged 21 to 40 years, mean 30.5 ± 5.8 years), in which PET scans were performed as a research protocol approved by the local ethics committee (CPP Pitié-Salpêtrière, Paris, France) and written informed consent was obtained.

The goal of this study was to evaluate the performances of SPM analysis of FDG-PET in children using an age-matched pseudo-control group of epileptic patients with robustly negative imaging versus classical healthy young adult controls. Both SPM analyses achieve a similar rate of detection of hypometabolic clusters and perform as well as visual analysis, not only in MRI-positive patients but also in MRI-negative ones. Although the sensitivity of SPM is not increased using pediatric vs. adult controls (79% vs. 87%), specificity is increased (97% vs. 89%) due to the reduction of the number of hypometabolic artifacts detected. As a result the positive predictive value of the method rises from 65% to 86%. That provides a relevant clinical impact especially in the younger children.

Visual inspection of FDG-PET images is extensively used in clinical epilepsy practice and remains the reference method of analysis in children [14, 27]. The hypometabolic areas detected have been proved to include the seizure onset zone, although they usually are more extended and do not delineate it [6, 8]. The localizing value of visual PET analysis varies from 36% to 73% in extratemporal cases, depending whether MRI is negative or not [9, 20], and reaches 90% exclusively in temporal lobe cases [26]. This detection value can be improved, particularly in extratemporal epilepsy with negative MRI, using PET/MRI coregistration [12, 13]. That has been confirmed by intracranial EEG and post-operative data. The present 56% detection rate of relevant hypometabolic areas is therefore in accordance with the literature, if one considers that we did not use coregistration and that the series mostly comprises MRI-negative patients and children with extratemporal epilepsy. In addition, the ability to detect a hypometabolic area visually is subjective and depends on the expertise and experience of the observer. Most reported series tend to limit this bias by achieving a consensus between closely performing readers, as we presently did with a first concordance of 0.81.

By contrast, SPM analysis is observer independent [1, 2]. Taking the whole brain into account, it allows assessing the spatial extent and location of the abnormal site on the brain map with no a priori hypothesis. Initially dedicated to the comparison of data sets among groups of subjects, SPM method also proved to be reliable for quantitative analysis of individual FDG-PET scans in various neurological disorders, including epilepsy [37]. SPM analysis has been widely applied to the study of adults with epilepsy, with thresholds varying from p < 0.001 (uncorrected) to p < 0.05 (corrected) and >20 voxels to >250 voxels [5, 7]. Since SPM showed comparable sensitivity to visual assessment, it is considered as an aid in diagnosing seizure onset zones in temporal as well as in extratemporal lobe epilepsy and in lesional as well as in non-lesional cases [5, 7, 12, 15, 16, 38].

In children, experience with SPM in epilepsy is more limited. Only one study formally assessed the role of statistical threshold on sensitivity and specificity [27]: they found p < 0.001 as the best compromise, as we presently do. Based on young adults as controls, the results are not univocal. De Tiege et al. did succeed detecting areas of remote inhibition in children aged 5 to 11 years with a particular epileptic encephalopathy and to see them vanish when epilepsy and cognitive deficits recover [24, 39, 40]. Some authors showed a high SPM performance (86%) for correct localization of the seizure onset zone in adolescents (12 to 15 years) [26], but others achieved a rate of only 13% in children from the age of 3 [14]. In another series, the same procedure gave an SPM sensitivity of 74% in children over 6 years old [27], whereas it failed in children under 6 years of age due to a significant proportion of artifacts [23]. We also found a high number of such false positive hypometabolic clusters when using an adult template, at a threshold stringent for such an individual SPM analysis [12]. One main issue is the different sizes of the head of adults and young children [23]. We confirm here that this artifact rate is higher in the younger ages and can be reduced by using a pediatric template [36]. Another potential advantage of a pseudo-normal epilepsy group could be that the potential effect of the antiepileptic drugs is largely cancelled out since both the patients and the control groups were on antiepileptic medication [35]. To generate even better templates, one could speculate on data from other pathologies that justify whole-body PET scanning but may respect the brain, like pediatric lymphoma, for example [41].

In order to deal with the pediatric controls the closest to “normals”, we confirmed PET negativity using the Signorini’s method [37]: each data set of the pseudo-control group was compared to the rest of the group using SPM. The complete procedure excluded three additional children. Our main result is a gain of specificity and this pediatric procedure reaches the highest rates ever reported in epilepsy for SPM. Conversely, the predictive positive value is improved, thus decreasing the number of artifacts and optimizing the clinical relevance of SPM analysis compared to the classical adult procedure. We assume that these advantages of pediatric SPM can compensate the slightly decreased sensitivity compared to adult SPM (79% vs. 87%). Note that these rates of sensitivity and specificity apply to SPM analysis and not to FDG-PET in general, since we excluded from their calculation the major part of negative cases. Nevertheless, this method was robustly able to identify some clinically relevant hypometabolic areas missed by visual analysis in 6% of cases.

Although our template presents with the usual balance of specificity versus sensitivity, we recently showed the whole procedure to be beneficial also in multifocal childhood epilepsy when visual PET analysis fails to detect bilateral abnormalities, as in school-age children with a fever-induced epileptic encephalopathy [42].

Such a pediatric SPM procedure is a way to optimize SPM analysis of FDG-PET in children with epilepsy, useful for clinical purpose and for research. It might also be considered for other brain pathologies in pediatrics in the future.