Participants
We included 7 patients (24.6 ± 5.7 yrs, 5M/2F) diagnosed with VSS. All patients were referred for imaging by a neurologist (E.B.) specialized in migraine and VSS. Other underlying neurological conditions were excluded and ophthalmological examinations were performed to exclude other underlying ophthalmological conditions. Inclusion criteria were visual snow with dynamic, continuous, black and white tiny dots in the entire visual field lasting longer than 3 months, with at least two additional visual symptoms [4]. All patients completed a questionary that focused on the presence of continuous or episodic visual snow, additional visual symptoms (palinopsia, entopic eye phenomena, photophobia and nyctalopia), the beginning of their visual symptoms, association of migraine, migraine aura and tinnitus, their general current and past medical history, and illicit drug use [26]. Patients were asked to participate at referral to undergo a subsequent PET/MR investigation after their routine FDG PET/CT.
An age- and gender-matched subset of 15 healthy controls (CON) (mean age 28.0 ± 5.3 yrs; 8M/7F) was randomly selected from a large normal 18F-FDG PET/MR database obtained on the same imaging system. [27]. The main relevant exclusion criteria were history of major internal pathology, neurological and/or psychiatric disorders (including psychosis, depression, and anxiety), history of frequent migraine attacks, substance abuse or current use of any central acting medication, first-degree relatives with dementia. All CON had a normal neurological examination, Mini-Mental State Examination (MMSE) ≥ 28, Beck’s Depression Inventory (BDI) score ≤ 9, and a normal structural T1 and T2 MRI.
The study was approved by the KU Leuven Ethics Committee under study numbers S58764 (patients) and S58571 (control data set). The study was conducted in accordance with the ethical standards as laid down in the Declaration of Helsinki and its later amendments. All subjects signed written informed consent before enrollment in the study.
Image acquisition
All subjects fasted at least for 4 hours prior to 18F-FDG injection. 18F-FDG was injected intravenously (149 ± 10 MBq) in standard ambient conditions, supine in a dark, noise free room with eyes and ears open. 18F-FDG PET images were acquired for 20 min on a simultaneous GE Signa 3T PET/MR scanner with integrated Time-of-Flight (TOF) (GE Healthcare, Chicago, IL, USA), starting 64.7 ± 10.8 min postinjection, as routine 18F-FDG PET/CT was first acquired in routine setting after 30 minutes. For CON, dynamic PET data were acquired for 60 min starting from 18F-FDG injection (146 ± 9 MBq). From the list-mode data, the last 20-min were reconstructed and used as comparator for this study.
Simultaneous with the 18F-FDG PET/MR acquisition, zero-echo-time (ZTE) MR (3D radial acquisition; Flip Angle: 0.8°; Bandwidth: 62.5 kHz) images for attenuation correction [20] and a 3D volumetric T1-weighted BRAVO MR sequence (echo time (TE) = 3.2 ms; repetition time (TR) = 8.5 ms; voxel size = 1x1x1 mm) were acquired, using a vendor supplied high-resolution 8-channel phased array head coil (GE Healthcare, Milwaukee, USA).
The 20-min 18F-FDG PET list mode data were rebinned in 4 frames of 5 min, corrected for deadtime, randoms, scatter and time-offset. A previously validated MR-based attenuation correction using the ZTE MR images was applied [28]. PET images were reconstructed using OSEM (ordered subset expectation maximization; 28 subsets, 4 iterations), including TOF information, resolution modelling and smoothed with in-plane Gaussian kernel with a FWHM (full width at half maximum) of 4.5 mm.
18F-FDG PET image processing
18F-FDG PET frames were first corrected for motion by a rigid frame-by-frame coregistration and a single static 18F-FDG image was obtained as the average of all motion-corrected frames using PMOD software (v4.1; PMOD Inc. Zürich, Switzerland). 18F-FDG data were coregistered to the individual volumetric T1-weighted MR images and were then analyzed quantitatively by voxel- and volumes-of-interest (VOI)-based analyses. Before performing the voxel-based group comparison analysis, all the coregistered 18F-FDG PET were spatially normalized using the non-linear deformation fields generated by the CAT12 toolbox using a DARTEL template (voxel size: 1.5x1.5x1.5 mm), and subsequently smoothed using a Gaussian kernel with FWHM of 8 mm. PET images were analyzed using proportional scaling to the average GM activity.
Voxelwise and semiquantitative VOI-based 18F-FDG PET analysis
To assess the differences in glucose metabolism between the VSS and CON group, a whole-brain Statistical Parametric Mapping (SPM12; Welcome Trust Centre for Neuroimaging, University College, London, UK) implemented in Matlab (R2020b, The MathWorks Inc., Natick, MA, USA) group comparison was carried out (two-sample independent t-test; significance level set to pFWE (family wise error corrected) < 0.05 at cluster level, pheight−FWE < 0.05 unless stated otherwise, and extent threshold KEXT > 200 voxels (approximately 0.68 ml)). As this analysis was considered exploratory, we also applied a lower threshold of significance, Pheight < 0.001 uncorrected at cluster level. An explicit binary mask was created by first averaging the individual normalized GM and CSF probability maps, and subsequently by subtracting the binarized averaged CSF mask from the binarized averaged GM mask. Also, based on the possible effect of the comorbid conditions migraine and tinnitus, additional statistical designs were performed with binarized presence of migraine or tinnitus symptoms as covariates.
VOI-based analysis was performed using the PMOD PNEURO tool (v4.1; PMOD Inc. Zürich, Switzerland) and the N30R83 Hammers probability atlas [22] resulting in 83 automatically delineated brain VOIs. Individual VOI 18F-FDG activities were normalized to the individual average grey matter (GM) activity to obtain the relative VOI metabolic activity. Aside from lingual gyrus, cuneus and lateral occipital cortex which were the main focus based on previous literature, the other regions were grouped into 9 larger bilateral composite regions: frontal cortex, cingulate cortex, temporal cortex, mesotemporal cortex, parietal cortex, (total) occipital cortex (lingual gyrus, cuneus and lateral occipital cortex), striatum, thalamus, insula and cerebellum. VOI-based group comparisons were assessed using a two-tailed unpaired t-test and Bonferroni post-hoc tests (p < 0.05/12 = 0.004).
Voxel-based morphometry and MR volumetry
MR volumetric differences between VSS and CON were assessed both at the voxel-level (voxel-based morphometry,VBM) and at the VOI level. Individual specific tissue probability maps for GM, WM and cerebrospinal fluid (CSF) were obtained by segmenting the spatially normalized 3D T1-weighted MR scans using the CAT12 toolbox (standard setting for parameters) of SPM12. The modulated warped GM probability maps were first smoothed with a kernel with 8mm FWHM and then used as input for the VBM analysis.
For VBM, the preprocessed images were entered into a statistical unpaired t-test design, with total intracranial volume (TIV) as a nuisance covariate to correct for different brain sizes. To exclude extracerebral clusters, the same explicit binary GM mask as used for the voxel-based 18F-FDG PET analysis was applied. Data were explored at pcluster < 0.05, and two thresholds were used at the voxel-level: a stringent pheight−FWE < 0.05 and a lower exploratory threshold pheight < 0.001 uncorrected. The cluster extent (kext) level was set at 200 voxels.
For the VOI-based analysis, GMV values were also extracted from the Hammers atlas after segmentation of the T1-weighed MR images within native MRI space in the PMOD PNEURO tool, and grouped into the same 12 larger composite VOIs as used for 18F-FDG. The GM VOI values were subsequently normalized to the TIV. VOI-based group comparisons were assessed using a two-tailed unpaired t-test and Bonferroni post-hoc tests (p < 0.05/12 = 0.004).
Descriptive statistics and discriminant analysis.
Descriptive statistics were performed with GraphPad Prism 9.1 (GraphPad Software, La Jolla, CA, USA). Significance was accepted at the 95% probability level.
Discriminant analysis was performed using SPSS Statistics v26.0 for Windows (IBM Armonk, NY, USA) with general discriminant modeling. The independent variables (predictor variables) used to predict the grouping variable (CON vs VSS) were regional relative VOI uptake data for 18F-FDG and MR GMV values. VOI data were entered independently into the discriminant function at once. For all analyses, a leave-one-out post-hoc classification was performed, only these data are reported. Receiver operating characteristic (ROC) analysis was performed for the significant discriminant 18F-FDG and MR GM VOIs to assess the diagnostic accuracy in discriminating between the VSS and CON groups.
Visual analysis of 18F-FDG PET
Prior to visual analysis, all 18F-FDG scans were fully anonymized, randomized and spatially normalized and processed using MIM- Neuro® as done in our clinical setting with transverse, sagittal and coronal slices and a 3D surface rendering of the PET data (v7.0; MIM software Inc., Cleveland, OH, USA) made available to the observers. Two experienced nuclear medicine physicians (K.V.L. and K.G., with 25 and 10 years of experience in brain imaging, respectively) visually analyzed and rated all images in a blinded fashion unaware of clinical information at time of the scan. The following instructions were given: firstly, to score activity in relevant regional left and right predefined areas (frontal, temporal, medial temporal, parietal, occipital, primary and secondary visual cortex, lingual gyrus, striatum, thalamus, and cerebellum), by using a 5-point scale (-2 = strongly decreased, -1 = slightly decreased, 0 = normal, 1 = slightly increased, 2 = strongly increased). Secondly, the observers were asked to binary classify the subject as “VSS patient” or “CON” with knowledge of previous literature data of Schankin [13], the SPM result in our group analysis and the number of subjects in each group. Finally, observers also gave a confidence rating for their binary classification, scaled as : 1 = very uncertain, 2 = rather uncertain, 3 = reasonably certain, 4 = certain, and 5 = very certain. The diagnostic accuracy (sensitivity, specificity, and accuracy) was calculated for each observer by direct comparison to the ground truth. Visual assessment of the MRI data was not attempted since the GM changes were deemed too subtle for visual detection conform earlier literature [1, 4, 16, 20].