dataset.zip
Literature search and inclusion
A comprehensive and systematic search was performed in PubMed and ScienceDirect Database (from January 1999 to June 2016) for VBM studies of IGE， by using the following keywords: “gray matter” AND (“epilepsy” OR “JME” OR “juvenile myoclonic epilepsy” OR “GTCS” OR “generalized tonic-clonic seizure” OR “AE” OR “childhood or juvenile absence epilepsy” OR “IGE” OR “idiopathic generalized epilepsy”) AND (“Voxel-based morphometry” OR “VBM” OR “Voxel-wise”). For the studies obtained, each was respectively scanned by a professor who specializes in neuroimaging and an experienced neurologist, to determine if it met the inclusion criteria. The references in the selected studies were also reviewed to identify the relevant papers.
The included studies should meet these criteria: (1) full text is accessible and published in English with peer review; (2) reporting a VBM comparison on gray matter volume between IGE patients and healthy controls; (3) reporting the stereotactic coordinates of significant GMV abnormalities across the whole brain; (4) corrected significance levels for multiple comparisons, or uncorrected levels with spatial extent thresholds were used.
A study was eliminated if (1) there was no healthy control group; (2) this study were not reported on Talairach or Montreal Neurological Institute (MNI) stereotactic coordination; (3) the data overlapped with other articles; (4) the reported changes were uncorrected and the spatial extent threshold was not reported; (5) the patients were infants or newborns; (6) the subject information was insufficient. The method used in the present study was according to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines for observational studies[18].
ALE meta-analysis
For comparing the GMV difference and seeking a consistent anatomical bias among IGE subsyndromes, we used BrainMap database (http://www.brainmap.org/), which is an online database of structural neuroimaging studies in the form of stereotactic(x,y,z) coordinates, including most of the published VBM studies (most papers included in our study can be found in this database). Thus, the above collected VBM studies, if included in this database, were firstly retrieved, and the associated information and coordinates were extracted from the Brainmap database using the tool Sleuth 2.3.6 (http://www.brainmap.org/). Experiments unavailable in the database were manually encoded using Scribe 2.3.1 (http://www.brainmap.org/). Coordinates in Talairach & Tournoux (T&T) space were converted to Montreal Neurological Institute (MNI-152) space by using the icbm2tal transform in ALE.
In this study, we performed pooled analysis to compare the difference of GMV abnormalities between IGE and controls, and within-subtypes analysis to compare the difference of GMV abnormalities between JME and controls, or AE and controls, by using cluster-level thresholding of ALE algorithm. When using cluster level inference, the simulated data are thresholded by using a “cluster-forming threshold”. GingerALE finds the contiguous volumes of clusters above the threshold, and then tracks the distribution of their volumes by using permutation threshold. False discovery rate (FDR)-corrected threshold inference was used to control the rate of false positive. In this study, pooled and within-subtypes analysis was performed by using a cluster level threshold of 0.05, a permutation threshold of 1000, and a FDR-corrected P value of less than 0.05. Notably, we did not perform meta-analysis for GTCS, since there were very few studies (1 for GMV increase, 2 for GMV decrease) about this subtype.
For comparing the difference of GMV abnormalities between JME and AE, we made a contrast analysis for these two subtypes. In this between-subtypes contrast analysis, GingerALE created simulated data by pooling the foci datasets and randomly dividing them into two new groups, which had the same size as the input data sets. For example, JME contained 20 foci, while AE consisted of 29 foci, then the pooled data of 49 foci were therefore randomly divided into two groups, one 20 and the other 29 foci. ALE values were calculated for each group, and then compared to the ALE values of the true data. After 1000 permutations, a null distribution was generated for the difference in ALE values between JME and AE. The true difference in ALE values was then tested against this null hypothesis at each voxel, generating a voxel-wise P-value image that was thresholded with a FDR <0.05 and a minimum cluster size of 100 mm3.
In addition, conjunction analysis was carried out for JME and AE to assess the common regions with GMV abnormalities. This was derived from the voxel-wise minimum value of the input ALE images. The resulting conjunction image reflects the statistically significant similarities between the JME and AE.
Mango software (http://ric.uthscsa.edu/mango) was employed to visualize the ALE results, which were overlaid on the MNI-152 brain template in MNI coordinate space.
