Patients and Healthy controls
This retrospective study was permitted by the Research Ethics Committee of the Capital Medical University of Xuanwu hospital. Written informed consent was obtained from each participant.
Patients were included if they had (1) MRI normal or subtle changes on an official neuroradiology report. (2) a preoperative simultaneous 18F-FDG PET/MRI scan, (3) received a comprehensive presurgical evaluation included seizure semiology, video-electroencephalography (V-EEG), neuroimaging and intracranial EEG (IEEG) (4) postoperative follow-up of over 12 months, and (5) a postoperative MRI or CT scan. Patients were excluded if they had (1) poor PET/MRI quality hindering clinical read or (2) poor PET/MRI quality causing significant registration errors in the data processing procedures. None of the patients had intraoperative or perioperative complications, and the preoperative antiepileptic regimen was continued for all patients in the postoperative period.
Available microscopic slides from surgical resections were reviewed by a dedicated neuropathologist. Pathology results was classified according to the International League Against Epilepsy classification [15].
Seizure outcomes were categorized one-year after surgery by the Engel Epilepsy Surgery Outcome Scale. [16] Surgical outcome was then categorized as Engel I (seizure free) and Engel II-IV (non-seizure free). The demographic and clinical characteristics of all participants are presented in Table 1.
22 age- and gender-matched healthy controls (HCs) ((mean age, 29.32 ± 5.65 years, range, 17–39 years; 11 males) were enrolled. All HCs were free of psychiatric or neurologic disorders on the basis of a health screening measure.
PET/MRI acquisition
Interictal 18F-FDG PET and MRI data were simultaneously obtained using a simultaneous TOF-PET/MR scanner (SIGNA, GE Healthcare, WI, USA). The patients fasted at least 6 hours, and the level of fasting blood glucose was lower than11.1 mmol/L. 18F-FDG with radiochemical purity of > 95% was produced by the unit, and the injection dosage was calculated based on the patients’ body weight (3.7 MBq/kg). All patients were at rest in a dimmed environment for 40 min. Video surveillance was used to monitor the patients in order to exclude an ictal or postictal FDG administration. During the scanning, the subjects were instructed to keep their eyes open, keep their head as still as possible. Three-dimension T1 brain volume imaging (3D T1 BRAVO) (repetition time (TR) = 2300ms, echo time (TE) = 2.98ms, angle = 9°, slices/gap = 160/0.5mm, FOV = 256mm, matrix size = 256×256, voxel size = 1.0×1.0×1.0mm3) and other structural imaging sequences for diagnosis images were immediately obtained.
The PET bed position included a simultaneous 18-second 2-point Dixon scan for MRI. Attenuation correction, scatter correction, random correction, and dead-time correction were also performed. Scanning parameters for reconstructed images were as follows: matrix size = 192×192, voxel size = 1.82×1.82×2.78 mm3, 89 slices. PET image reconstruction with ordered subset expectation maximization algorithm (OSEM) 3 iterations, and 32 subsets, time-of flight and Sharp IR.
MRI postprocessing
MAP was performed on 3D T1 BRAVO images using an in-house code in MATLAB. The computed output consists of 3 volumetric statistical maps, called the junction, extension, and thickness maps. A blinded reviewer used the z score threshold of 4 to identify candidate MAP positive regions on the junction file, an accompanying region on the extension file (z > 6) and the thickness file (z > 4). The choice of z score threshold was consistent with previous literature [3]. Candidate MAP positive regions were searched in the whole brain. All candidate MAP positive regions were then addressed by two experienced neuroradiologists, who conducted a corresponding focused re-review of the preoperative clinical MRI (with 3D T1 BRAVO, T2-weighted FLAIR, and turbo spin echo sequences), if they have any opinions, a third neuroradiologists needed to determine. The MAP result was classified as concordant with cortical resection if the abnormal area included the resection site; otherwise, it was classified as non-concordant with cortical resection.
QPET analysis
All imaging data were preprocessed in SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12) and an in-house code in MATLAB. Glucose metabolism from all HCs were served as the normal databases. To assess the individual glucose metabolism changes, a two-sample t-test was performed between individual patient data and normal database, with age and gender regressed out as covariates to reduce the effects of these variables. Global nuisance effects were estimated by dividing the intensity in each image by the intensity of the cerebellum. Registered PET maps were spatially normalized to the Montreal Neurological Institute (MNI) space by the same transformation parameters from the segmentation procedure of the T1-weighted images. Preprocessed PET images in MNI space were converted to the standardized uptake value ratio (SUVr) relative to the cerebellum and then smoothed using a 6-mm full width at half maximum (FWHM) Gaussian kernel. The resulting SPM (t) maps were thresholded with a significance level of P < 0.05 without corrections. The SPM (t) maps reviewed by a dedicated nuclear medicine physician to exclude the focal metabolism caused by sulcus, if the physician has any questions, another physician is required to make a decision after consultation. The QPET result was interpreted as a normal or abnormal scan. If the SPM (t) maps was abnormal, the anatomic location of abnormal metabolism was recorded. The QPET result was classified as concordant with cortical resection if the abnormal metabolic area included the resection area; otherwise, it was classified as non-concordant with cortical resection.
For the combined MAP + QPET, if the results of both of the tests were concordant with cortical resection, the combined tests were considered as concordant; conversely, if the result of one of the tests was non-concordant, the combined tests were considered as non-concordant. For MAP/QPET, if the result of one of the tests were concordant, the tests were considered as concordant; conversely, MAP/QPET was considered non-concordant if both tests were non-concordant.
Statistical analysis
All data analysis was performed with SPSS software (IBM SPSS Statistics, Version 21.0). Group-level comparisons of demographic and clinical characteristics were carried out with independent-sample Student’s t-tests, one-way analysis of variance, or chi-square tests. P < 0.05 was considered statistically significant. True positive (TP) was classified as if the imaging results concordant with surgical resection in seizure-freedom patients (Engel I), imaging results non-concordant with surgical resection in patients with ongoing seizures (Engel II-IV) was classified as false positive (FP). The imaging results non-concordant with surgical resection in patients with ongoing seizures was classified as true negative (TN). Such a case neuroimaging found no focal lesion and surgery proceeded, but the patient did not improve. False negative (FN) was defined as imaging results concordant with the actual surgical resection in seizure-freedom patients. Sensitivity = TP/TP + FN, Specificity = TN/TN + FP, PPV = TP/TP + FP, NPV = TN/TN + FN. Associations between the regions identified by QPET and MAP concordant with cortical resection and seizure outcome at least one-year were tested by Chi-square tests.