The committees of human research at the University of California, San Francisco (UCSF); California Pacific Medical Center, San Francisco; and VA Medical Center, San Francisco approved the study, and written informed consent was obtained from each subject. Twenty-nine patients with drug-resistant TLE who agreed to participate in the 4T research protocol, which was administered in addition to the standard imaging procedures, were recruited between mid-2005 and the end of 2007 from the Pacific Epilepsy Program, California Pacific Medical Center, and the Northern California Comprehensive Epilepsy Center, UCSF, where they underwent evaluation for epilepsy surgery. Fifteen patients (mean age 41.3 ± 10.4 years, left TLE/right TLE 9/6, female/male 10/5) had evidence for mesiotemporal lobe sclerosis on their 1.5 T MR images (TLE-MTS) and 14 patients (mean age 39.8 ± 8.1 years, left TLE/right TLE 6/8, female/male 7/7) had normal-appearing hippocampi on their 1.5T MR images (TLE-no) and normal 1.5 T MR results. All the 1.5T reads were done by an experienced neuroradiologist. Hippocampal volumetry was used to confirm the presence (TLE-MTS) or absence (TLE-no) of significant hippocampal volume loss on the 4T images. The identification of the epileptogenic focus was based on seizure semiology and prolonged ictal and interictal video/EEG/telemetry (VET) in all patients. The control population consisted of 29 healthy volunteers (mean age 37.7 ± 9.2 years, female/male 18/11). displays the patient characteristics.
All imaging was performed on a Bruker MedSpec 4T system (Bruker MedSpec, Madison, WI, U.S.A.) controlled by a Siemens Trio (Siemens, Iselin, NJ, U.S.A.) console and equipped with a USA instruments (Aurora, OH, U.S.A.) eight-channel array coil. The following sequences were acquired: (1) for cortical thickness and thalamus measurements a volumetric T1-weighted gradient echo MRI (TR/TE/TI = 2,300/3/950 ms, 1.0 × 1.0 × 1.0 mm3 resolution, acquisition time 5.17 min); (2) for the measurement of hippocampal subfields, a high-resolution T2-weighted fast-spin echo sequence (TR/TE: 3,500/19 ms, 0.4 × 0.4 mm in-plane, 2 mm slice thickness, 24 slices acquisition time 5:30 min); and (3) for the determination of the intracranial volume (ICV), a T2-weighted turbospin echo sequence (TR/TE 8,390/70 ms, 0.9 × 0.9 × 3 mm nominal resolution, 54 slices, acquisition time 3.06 min).
Cortical thickness measurement
images were segmented using the expectation-maximization segmentation (EMS) algorithm (Van Leemput et al., 1999a
). The bias field maps and tissue maps obtained from this process were used for bias correction and skull stripping of the T1
image. To allow for a combination of left and right TLE in the analysis, the T1
images of all patients with right TLE were side flipped so that the focus was on the left side in all patients. The same was done with all control images. FreeSurfer was used for cortical surface reconstruction and cortical thickness estimation of the original (controls and left TLE) and side-flipped (controls and right TLE) images. The procedure has been described extensively elsewhere (Dale et al., 1999
; Fischl et al., 1999a
; Desikan et al., 2006
). All outputs underwent a quality check and were manually corrected if necessary (between 15 and 30 min/subject). A project-specific average spherical representation (average of original and side-flipped control data, original left TLE data, and side-flipped right TLE data) using a nonrigid high-resolution surface-based averaging method for an optimal alignment of the cortical folding pattern was generated. Each subjects’ inflated brain was morphed onto this average for the vertex-based statistical analysis. The data were smoothed with a 20-mm full-width half-maximum (FWHM) Gaussian kernel to improve the signal-to-noise ratio. Mean cortical thickness in the ipsilateral inferior temporal gyrus (part of the presumed epileptogenic focus in TLE-no) (Vossler et al., 2004
; Mueller et al., 2007a
) for the correlation analyses was taken from the corresponding label of the FreeSurfer output. A region of interest (ROI) was manually drawn consisting of the regions in entorhinal cortex (ERC), parahippocampal gyrus, and lingual gyrus (ERC-paralingual ROI) in which cortical thickness was positively correlated with ipsilateral thalamic volume in TLE-MTS and controls, and TLE-no and controls (c.f. ) to identify the thalamic nuclei driving the overall thalamus volume loss.
Figure 1 (A) Region of interest (ROI, lilac) containing parts of the ERC, parahippocampal, and lingual gyrus (ERC-paralingual ROI). Size and shape of the ROI were based on the region that showed significant correlations with the total ipsilateral thalamus volume (more ...)
Hippocampal subfield volumetry
The method used for subfield marking including assessment of measurement reliability has been described in detail previously (Mueller et al., 2007b
). The marking scheme depends on anatomic landmarks, particularly on a hypointense line representing myelinated fibers in the stratum moleculare/lacunosum (Eriksson et al., 2008
), which can be reliably visualized on these high-resolution images. Therefore, external and internal hippocampal landmarks are used to further subdivide the hippocampus into subiculum, cornu ammonis sectors (CA) CA1, CA1-2 transition zone, and CA3/dentate gyrus (c.f. ).
Thalamus deformation-based morphometry (DBM) and volumetry
An unbiased, symmetrical normal atlas was created from the original and side-flipped images of the control group. These images were coregistered (12 parameter affine) to a randomly selected nonflipped control image. Next, a fluid registration algorithm (Christensen et al., 1996
) was used to generate the final symmetrical normal atlas using the technique described by Lorenzen et al. (2005)
and Joshi et al. (2004)
. All images (original and side-flipped control images, original left TLE images, and side-flipped right TLE images) were then linearly coregistered to this symmetrical atlas (12 parameter affine). A 75 × 96 × 51 mm ROI centered on the thalamus was selected in the atlas, and in each image and the subject structures contained within this ROI were warped onto the corresponding structures in the atlas using a fluid registration algorithm (Christensen et al., 1996
A deformation map for each subject was generated by calculating the jacobian determinant of the resulting nonlinear transformation matrix. The left and right thalamus region in each Jacobian map was identified and extracted from these deformation maps using thalamus labels obtained by manually marking both thalami in the atlas (see next paragraph) and smoothed with a 4-mm FWHM Gaussian spatial kernel for voxel-based statistical analysis. Because the normal atlas had been generated from the control population, a voxel expansion in patients compared to controls can be interpreted as evidence for volume loss in patients and a voxel contraction as volume gain.
The thalamus in the symmetrical atlas was manually traced using anatomic landmarks (Natsume et al., 2003
). The so-obtained labels for the right and left thalamus were then applied to each subject’s image by warping the thalamus region of the atlas onto the subject brain using a combination of linear and nonlinear coregistration steps. The resulting subject thalamus labels were visually inspected for accuracy and manually corrected if necessary. shows the manual thalamus tracing in the atlas and an example of the automatically obtained thalamus label in an individual subject.
Figure 2 (A) Overview of the generation of the thalamus deformation map. The yellow region in the right-most image indicates the left-sided thalamus label that was used to select the thalamus. (B) Left side: Atlas with manual thalamus tracings. The enhanced contrast (more ...)
Thalamic side differences (left vs. right, and ipsi- vs. contralateral) were tested using a paired t
test. After exclusion of side differences in the control group, left and right thalamus was averaged for the comparisons with TLE. Group differences between ipsi- and contralateral thalamus volumes were assessed using linear regression analysis followed by Tukey’s post hoc tests. Thalamus volume was entered as dependent and “age,” ICV, and “group” as independent measures. Regions of cortical thinning in TLE-MTS and TLE-no (side-flipped images for right TLE, original images for left TLE) compared to controls (original and side-flipped images) were tested for in a regionally unbiased way by computing a general linear model of the effect of “group” on thickness at each vertex using the statistical tool provided by FreeSurfer. A linear regression analysis was used to identify ipsilateral regions with significant correlations between ipsilateral thalamus volume (normalized to ICV because it had been shown to be highly dependent on ICV in the preliminary analysis) and cortical thickness (not a volume and hence not corrected for ICV) in controls and TLE-MTS and in controls and TLE-no. Age was used as a nuisance variable in these analyses because thalamus volume as well as cortical thickness had been found to be negatively correlated with age. False discovery rate (FDR) of p ≤ 0.05 was applied to all FreeSurfer analyses to correct for multiple comparisons. SPM2 (Wellcome Department of Cognitive Neurology, http://www.fil.ion.ucl.ac.uk
) running MATLAB 6.1 (The MathWorks, Natick, MA, U.S.A.) was used to search for differences between groups [TLE-MTS vs. controls, TLE-no versus controls, analysis of covariance (ANCOVA), age as nuisance variable, family-wise-error rate (FWR), p ≤ 0.05 to correct for multiple comparisons] in the thalamus deformation maps. Partial correlation analyses with age as a nuisance variable were used to test for positive correlations between regions of thalamic volume loss and ipsilateral CA1 volume, ipsilateral cortical thickness of the inferior temporal gyrus, and the ERC-paralingual ROI in controls and TLE-MTS and in controls and TLE-no.