Introduction
The majority of drug-resistant epilepsy are mesial temporal lobe epilepsy (MTLE) [
1]; MTLE comprises about 80% of epilepsy surgical resections that aim to remove localized epileptogenic zone (EZ) [
2]. In MTLE, the epileptogenic zone commonly involves the mesial temporal lobe structures, including the hippocampus, amygdala, and parahippocampal gyrus [
3,
4]. The most common finding in EZ of MTLE is hippocampal sclerosis (HS) [
4,
5], which is histologically characterized by neuronal loss and gliosis [
6‐
8]. Typical radiological features on magnetic resonance imaging (MRI) of HS include hippocampal atrophy, disrupted internal hippocampal structure, decreased T1-weighted signal, and increased T2-weighted signal [
5,
9,
10]. Patients with non-identifiable lesions on MRI (MR-negative) tend to have worse surgical outcomes than patients with MR lesions [
11]. Furthermore, over 80% of MR-negative MTLE patients who underwent surgery had abnormal histopathology [
12], thus calling for more sensitive imaging techniques.
18F-Fluorodeoxyglucose positron emission tomography ([
18F]FDG PET) has become widely used in presurgical workup of epilepsy, and is recommended by the neuroimaging subcommission of the International League Against Epilepsy (ILAE) [
13] despite a lack of correlation between hypometabolism and the severity of either MRI or histopathology [
14]. Particularly in MR-negative MTLE patients, FDG PET has been shown to have higher sensitivity in identifying temporal and hippocampal hypometabolism [
14,
15], which lead to improved surgical outcome [
16]. On the other hand, the detection rate of EZ in epilepsy using FDG PET or FDG PET/CT has been reported to be 36–73% [
17,
18], possibly due to subtle or extended hypometabolism.
Quantitative MR T2 relaxometry measures intrinsic tissue property, which may detect subtle pathology in the hippocampus even in the absence of hippocampal atrophy [
19‐
23]. Increased T2 value has been reported to be consistent with histopathologic findings of HS [
21,
24‐
26], which is found to provide more sensitive lesion detection compared to volumetric MRI at 1.5T [
21] and T2-weighted fluid-attenuated inversion recovery (FLAIR) at 3.0T [
27]. Furthermore, progress in fast imaging techniques allows reliable whole brain quantitative T2 relaxometry using multi-echo spin-echo within clinical feasible time [
28]. Thus, T2 mapping might play a role in the lateralization of MTLE in the presurgical evaluation of epilepsy.
Research with hybrid [
18F]FDG PET/MR system has been conducted in epilepsy patients, which demonstrated higher detection accuracy than PET/CT by fusion of PET images and high-resolution anatomical MR images [
18,
29]. Given that T2 relaxometry is sensitive to epileptogenic pathologies, we hypothesize that the combination of [
18F]FDG PET and T2 mapping would provide the benefit of both metabolic and intrinsic tissue property measurements, which could be manifested both in visual radiological assessments and quantitative analysis for the lateralization of MR-negative MTLE patients.
Method
Participants
This study has been approved by the Internal Review Board of Ruijin Hospital. Patients diagnosed with drug-resistant epilepsy between December 2017 and February 2020 were identified. The inclusion criteria of patients are as follows: (1) clinical history, neurological examination, seizure semiologies, scalp video-EEG findings, and neuropsychological deficit pattern that are consistent with the characteristics of unilateral MTLE; (2) MRI was either normal or disclosed patterns suggestive of HS; (3) stereoelectroencephalography (SEEG) examination was obtained to confirm presurgical localization of the EZ. On the other hand, patients with generalized epilepsy syndromes, posttraumatic epilepsy, brain tumors, or other nervous system lesions other than HS were excluded.
We also recruited two healthy control groups. The first control group (group I) consisted of 24 subjects with MRI scans, while the second control group (group II) consisted of 15 subjects with [18F]FDG PET/MR. None of the healthy participants had any history of neurologic or psychiatric illness or has taken chronic medications. Written informed consents were obtained from all participants.
Data acquisition
PET and MRI scans were performed with integrated a 3.0-T hybrid PET/MR scanner (Biograph mMR; Siemens Healthcare). MRI sequences include 3D T1-weighted anatomical images using MPRAGE (resolution 0.5 × 0.5 × 1.0 mm
3, TR/TE/TI 1900/2.44/900 ms, FOV 250 × 250 mm
2, 192 slices), T2-weighted FLAIR (resolution 0.4 × 0.4 × 3.0 mm
3, TR/TE/TI 8460/92/2433 ms, FOV 220 × 220 mm
2, 45 slices), and a multi-echo spin-echo T2 mapping sequence (in-plane resolution 0.4 × 0.4 mm
2, 5.0 mm slice thickness, TR/TE
1/TE
2/TE
3/TE
4/TE
5/TE
6 2000/10.5/21.0/31.5/42.0/52.5/63.0 ms, 21 slices). All patients and controls were administered [
18F]FDG intravenously using a mean dose of 184.8 ± 29.0 MBq (range 133.2–247.9 MBq), with the scan being initiated 30~50 min after the injection. Static PET data were acquired in a sinogram mode for 15 min, matrix size 344 × 344, and post-filtered with an isotropic full-width half-maximum (FWHM) Gaussian kernel of 2 mm. Attenuation correction was performed using advanced PET attenuation correction with a unique 5-compartment model including bones [
30].
Image processing and analysis
Voxel-wise T2 maps were reconstructed with monoexponential nonnegative least-squares fitting of the multi-echo signals (MapIt; Siemens Medical Solutions). T2 maps were registered to T1-weighted images with the parameters of the registration between the first echo T2-weighted image and T1-weighted image. Meanwhile, voxels with T2 values larger than 170 ms were excluded to alleviate cerebrospinal fluid (CSF) contaminations [
31].
The FDG standard uptake value ratios (SUVRs) from PET images were obtained by intensity normalization via global mean scaling to correct individual variations [
32]. The SUVR maps were also registered to T1-weighted images. All image registrations were employed with rigid registration (6 degrees-of-freedom) using SPM12 (
https://www.fil.ion.ucl.ac.uk/spm/).
Physician visual assessment
The MR and PET images were visually analyzed by three experienced radiologists with certificates of both nuclear medicine and radiology. The readers were blinded for the clinical diagnosis of lateralization. The image evaluation was divided in three separate sessions with a break of at least 1 month: (1) routine MR imaging (T1-weighted, FLAIR, and T2-weighted images); (2) T2 (T2 map and z-score map); (3) PET (SUVR map and z-score map). In the image evaluation process, all cases were presented in a randomized order in every session. Any disagreement was further resolved through discussion.
Statistical analysis
Wilcoxon signed-rank tests were applied to compare hippocampal T2 and SUVR between the left and right hippocampus in healthy control, and between the ipsilateral and contralateral hippocampus in MTLE groups. For group comparisons, Mann–Whitney U tests were applied. A p value below 0.05 was considered statistically significant.
In order to evaluate the performance of hippocampal asymmetry of PET and T2 in the lateralization of MR-negative MTLE, logistic regression was used to discriminate left and right MR-negative MTLE groups. Left MTLE (LTLE) and right MTLE (RTLE) were defined as positive and negative samples respectively. Due to the limited sample sizes, a leave-one-out cross-validation strategy was implemented to corroborate the predictive generalizability of models. The area under the curve (AUC) of receiver operating characteristic curve (ROC) was calculated to evaluate the performance of each regression model. In addition, mean square error (MSE), the error between the true label and the predicted probability, was calculated to summarize the prediction error of each group. A lower MSE indicates that the classification model is more accurate.
All statistical analyses were performed using IBM SPSS v24. The logistic regression models were performed using the Python scikit-learn (sklearn).
Discussions
In this study, we utilized the hybrid PET/MR system to investigate the complementary lateralization capability of FDG PET and T2 mapping for MR-negative MTLE patients. Routine structural MRI and T2w-FLAIR lateralized 47.8% (22/46) of MTLE patients, and T2 mapping combined with [
18F]FDG PET correctly lateralized 95.6% (44/46) of MTLE patients. For MR-negative patients, FDG PET lateralized 19 out of 24 (79%) by visual assessment (Fig.
1; Table
S1). Together with quantitative T2 mapping, 23 out of 24 (95.83%) MR-negative patients were correctly lateralized, suggesting the potential benefit of combining T2 mapping with FDG PET. Our findings indicated that using hybrid classification models based on the combination of these two modalities results in higher AUC in predicting lateralization for MR-negative MTLE patients.
Epileptogenesis is typically characterized by neuronal damage and gliosis [
34]. Neuronal damage can lead to hypometabolism and hence may be identified by FDG PET, which was shown to be a promising biomarker for neural injury and dysfunction [
35]. However, reactive gliosis in epileptogenesis can also cause partial recovery of glucose hypometabolism [
36,
37]. The incidence of hypometabolism recovery from gliosis may further limit the capability of FDG PET in detecting epileptogenesis, which may explain why it did not correlate with histological analysis of tissue damage [
14]. On the other hand, strong correlation has been found between T2 signal and histological evidence of gliosis [
5,
38]. This may explain why T2 mapping could play a complementary role to FDG PET for the lateralization of EZs in cases when hypometabolism fails to be a good indicator.
Quantitative T2 relaxometry is sensitive to recognize subtle changes, even for epileptogenic hippocampus with normal MR findings [
19,
21]. In this work, we also demonstrate that quantitative hippocampal T2 relaxometry can improve lateralization in MTLE patients, which becomes especially valuable for those ambiguous cases in FDG PET. Simultaneous T2 mapping and PET can allow acquisitions under the same physiological and pathophysiological conditions, thus more accurate EZ localization [
17]. Compared with single mode modality, T2 mapping combined with [
18F]FDG PET could improve the lateralization accuracy by correctly lateralizing 95.6% (44/46) of MTLE patients.
Several notes should be taken when considering T2 mapping incorporation into clinical routine. (1) Scan parameters: Considering the multiexponential nature of brain T2 relaxometry [
39], difference in acquisition parameter selection could partially contribute to difference in absolute T2 values among literature reports [
40]. (2) CSF removal: Because CSF has very long T2, it appears very bright on T2 maps, interfering visual inspections. In this study, we erode the boundary of hippocampal segmentation and employ a threshold of 170 ms to minimize CSF contamination [
31]. (3) Quantitative comparisons: We observed longer T2 relaxation time (ΔT2
L-R = 1.17 ± 2.38 ms,
p = 0.04) and slightly higher SUVR (ΔSUVR
L-R = 0.02 ± 0.03,
p = 0.01) in the left hippocampus compared to the right hippocampus in healthy subjects (Supplementary Material
2). The left-right asymmetry of healthy subjects may confound the subtle ipsilateral-contralateral asymmetry of MRI-negative patients, complicating visual assessment of MR-negative cases. As commonly done for FDG PET images, automated quantitative approach of hippocampal asymmetry by
z-score normalization would avoid the visual bias of having asymmetric baseline T2. Quantitative hippocampal T2 can objectively characterize the presence and laterality of hippocampal abnormalities in MTLE with the accuracy comparable or even better than the performance of visual diagnosis from experts [
19,
31,
41]. Finally, with current fast imaging strategy and computation power, whole brain T2 mapping is feasible in clinical scans, with the possibility of incorporating image processing steps into radiological workstation to aid clinical assessments.
This study has limitations. The ages of healthy control for PET are not well-matched with patients. Since glucose metabolism of the hippocampus has been reported as age-independent from age 16 to 80 years [
42‐
44], we did not recruit extra healthy volunteers to radioactive PET imaging. Another limitation is the small sample size, which is why leave-one-out cross-validation was used in testing the classification models, and also due to subpial aspiration during surgery, which is supposed to minimize the damage to surrounding structures during the operation. As a result, it was very challenging to obtain the en bloc resected hippocampal specimens for histopathological studies. Future studies will be designed to collect relatively intact hippocampus specimens and expand larger cohorts to confirm our findings.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.