Skip to main content
Advertisement

Main menu

  • Home
  • Content
    • Current Issue
    • Accepted Manuscripts
    • Article Preview
    • Past Issue Archive
    • AJNR Case Collection
    • Case of the Week Archive
    • Classic Case Archive
    • Case of the Month Archive
  • Special Collections
    • Spinal CSF Leak Articles (Jan 2020-June 2024)
    • 2024 AJNR Journal Awards
    • Most Impactful AJNR Articles
  • Multimedia
    • AJNR Podcast
    • AJNR Scantastics
    • Video Articles
  • For Authors
    • Submit a Manuscript
    • Author Policies
    • Fast publishing of Accepted Manuscripts
    • Graphical Abstract Preparation
    • Manuscript Submission Guidelines
    • Imaging Protocol Submission
    • Submit a Case for the Case Collection
  • About Us
    • About AJNR
    • Editorial Board
  • More
    • Become a Reviewer/Academy of Reviewers
    • Subscribers
    • Permissions
    • Alerts
    • Feedback
    • Advertisers
    • ASNR Home
  • Other Publications
    • ajnr

User menu

  • Alerts
  • Log in

Search

  • Advanced search
American Journal of Neuroradiology
American Journal of Neuroradiology

American Journal of Neuroradiology

ASHNR American Society of Functional Neuroradiology ASHNR American Society of Pediatric Neuroradiology ASSR
  • Alerts
  • Log in

Advanced Search

  • Home
  • Content
    • Current Issue
    • Accepted Manuscripts
    • Article Preview
    • Past Issue Archive
    • AJNR Case Collection
    • Case of the Week Archive
    • Classic Case Archive
    • Case of the Month Archive
  • Special Collections
    • Spinal CSF Leak Articles (Jan 2020-June 2024)
    • 2024 AJNR Journal Awards
    • Most Impactful AJNR Articles
  • Multimedia
    • AJNR Podcast
    • AJNR Scantastics
    • Video Articles
  • For Authors
    • Submit a Manuscript
    • Author Policies
    • Fast publishing of Accepted Manuscripts
    • Graphical Abstract Preparation
    • Manuscript Submission Guidelines
    • Imaging Protocol Submission
    • Submit a Case for the Case Collection
  • About Us
    • About AJNR
    • Editorial Board
  • More
    • Become a Reviewer/Academy of Reviewers
    • Subscribers
    • Permissions
    • Alerts
    • Feedback
    • Advertisers
    • ASNR Home
  • Follow AJNR on Twitter
  • Visit AJNR on Facebook
  • Follow AJNR on Instagram
  • Join AJNR on LinkedIn
  • RSS Feeds

Welcome to the new AJNR, Updated Hall of Fame, and more. Read the full announcements.


AJNR is seeking candidates for the position of Associate Section Editor, AJNR Case Collection. Read the full announcement.

 

Research ArticleBrain
Open Access

Impact of Methodologic Choice for Automatic Detection of Different Aspects of Brain Atrophy by Using Temporal Lobe Epilepsy as a Model

C. Scanlon, S.G. Mueller, D. Tosun, I. Cheong, P. Garcia, J. Barakos, M.W. Weiner and K.D. Laxer
American Journal of Neuroradiology October 2011, 32 (9) 1669-1676; DOI: https://doi.org/10.3174/ajnr.A2578
C. Scanlon
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
S.G. Mueller
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
D. Tosun
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
I. Cheong
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
P. Garcia
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
J. Barakos
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
M.W. Weiner
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
K.D. Laxer
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
  • Article
  • Figures & Data
  • Info & Metrics
  • Responses
  • References
  • PDF
Loading

Abstract

BACKGROUND AND PURPOSE: VBM, DBM, and cortical thickness measurement techniques are commonly used automated methods to detect structural brain changes based on MR imaging. The goal of this study was to demonstrate the pathology detected by the 3 methods and to provide guidance as to which method to choose for specific research questions. This goal was accomplished by 1) identifying structural abnormalities associated with TLE with (TLE-mts) and without (TLE-no) hippocampal sclerosis, which are known to be associated with different types of brain atrophy, by using these 3 methods; and 2) determining the aspect of the disease pathology identified by each method.

MATERIALS AND METHODS: T1-weighted MR images were acquired for 15 TLE-mts patients, 14 TLE-no patients, and 33 controls on a high-field 4T scanner. Optimized VBM was carried out by using SPM software, DBM was performed by using a fluid-flow registration algorithm, and cortical thickness was analyzed by using FS-CT.

RESULTS: In TLE-mts, the most pronounced volume losses were identified in the ipsilateral hippocampus and mesial temporal region, bilateral thalamus, and cerebellum, by using SPM-VBM and DBM. In TLE-no, the most widespread changes were cortical and identified by using FS-CT, affecting the bilateral temporal lobes, insula, and frontal and occipital lobes. DBM revealed 2 clusters of reduced volume complementing FS-CT analysis. SPM-VBM did not show any significant volume losses in TLE-no.

CONCLUSIONS: These results demonstrate that the 3 methods detect different aspects of brain atrophy and that the choice of the method should be guided by the suspected pathology of the disease.

Abbreviations

DBM
deformation-based morphometry
EMS
expectation maximization segmentation
FDR
false discovery rate
FS
Freesurfer
FS-CT
FS-based cortical thickness
FSL
FMRIB Software Library
FWHM
full width at half maximum
GM
gray matter
ICV
intracranial volume
SPM
statistical parametric mapping
TLE
temporal lobe epilepsy
TLE-mts
TLE–mesial temporal sclerosis
TLE-no
TLE–normal-appearing hippocampus
ULD
unbiased large deformation
VBM
voxel-based morphometry
WM
white matter

The most commonly used methods for automated whole brain structural analysis are VBM,1,2 DBM,3,4 and cortical thickness methods.5,6 VBM is the most widely used method to date, with more than 22 studies published in the area of TLE alone.7–11 One of the main reasons for its popularity is probably that VBM is relatively easy to perform, with freely available software such as SPM (Wellcome Department of Cognitive Neurology, London, United Kingdom) and FSL (FMRIB Analysis Group, Oxford, United Kingdom). The first step of so-called optimized VBM12 is to generate a probabilistic GM map from the T1 gray-scale images by using a combination of voxel intensity and an a priori knowledge of the spatial distribution of GM. GM maps are then registered to a reference image. Registration to the reference image is based on a low-dimensional spatial transformation that aligns global differences but preserves local differences in GM distribution. GM volume differences between groups are then assessed voxel by voxel by using a general linear model.

DBM differs from VBM in 2 main aspects. First, the DBM image is not segmented, so the information from the full gray- scale brain is used in the analysis. Second, DBM registration is high-dimensional, eliminating individual subject physiologic and pathologic morphology differences. The anatomic differences then lie in the deformation fields that are required to transform the subject's brain. The more precise the registration, the more sensitive the method will be to detect subtle systematic structural differences that may not be possible with VBM.13,14 The downside of this technique, however, is that DBM registration algorithms are not widely available and are not as simple to implement as VBM. This is probably one of the main reasons VBM is generally used over DBM. To our knowledge, there are no whole brain DBM studies reported in TLE.

Cortical thickness is commonly computed by analyzing the 3D reconstruction of the brain's cortical surface from structural MR imaging. The freely available FS software (Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Boston, Massachusetts) is the most commonly used software for that purpose. FS-CT is determined as the distance from the GM/WM surface to the GM/CSF surface. The intersubject registration procedure is then based on alignment of the cortical folding patterns as opposed to the voxel intensities used for VBM and DBM. Although fluid warped images by using the DBM approach also can very precisely match the reference image, matching intensities may be less anatomically meaningful than matching cortical folding patterns. This may lead to a failure to align matching cortical regions across subjects, resulting in a lack of power in localizing subtle cortical differences in the voxel-based approaches.13 The drawback of cortical thickness analysis is that it will not detect subcortical abnormalities.

The overall aim of this study was to compare VBM, DBM, and FS-CT regarding their ability to detect different types of atrophy (subtle or microscopic atrophy versus mass or macroscopic atrophy). TLE is associated with both types of atrophy and thus was chosen as a model to investigate this question. Based on the appearance of the hippocampus, TLE can be divided into 2 subgroups: TLE with MR imaging signs (hippocampal atrophy and increased T2 signal intensity) of mesial temporal sclerosis (TLE-mts) and TLE with normal MR imaging, characterized by a normal-appearing hippocampus (TLE-no). In addition to the hippocampal volume loss, TLE-mts is characterized by ipsilateral mesial and lateral temporal, but also frontal, parietal, occipital, and cerebellar volume loss as well smaller subcortical structures (basal ganglia and thalamus).9,15,16 Structural changes in TLE-no by using both VBM and region of interest approaches however are more subtle and less consistent.17–19,11 Cortical thickness measurements, on the other hand, have shown widespread temporal and extratemporal cortical thinning.16 Due to differences in study design, eg, measurement parameter investigated, study population, field strength, and applied statistical analysis, it is not possible to draw conclusions regarding the sensitivity of the methods used to detect these abnormalities across studies.

The aim of this study was to compare the type of volume loss detected by each method—SPM-VBM, DBM, and FS-CT—and so to provide guidance on which method may be best to adopt in answering specific research questions. We expected that in TLE-mts, where macroscopic volume abnormalities occur, all 3 methods would be sensitive enough to detect these large-scale volume changes (CT on the cortex only). However, it was expected that in TLE-no DBM would be able to detect subtle subcortical changes over VBM due to its superior coregistration. We also hypothesized that FS-CT would detect subtle cortical abnormalities, without macroscopic volume losses, not detected by either voxel-based method.

Materials and Methods

Subjects

Participants in this investigation consisted of 15 patients with unilateral TLE-mts (6 men, 9 women; mean age, 40.1 ± 9.6 years), 14 patients with unilateral TLE-no (6 men, 8 women; mean age, 39.6 ± 8.3 years), and 24 healthy controls (12 men, 12 women; mean age, 37.9 ± 9.4 years). All patients were recruited during presurgical evaluation from the University of California, San Francisco and the Pacific Epilepsy Program, California Pacific Medical Center. A cortical thickness analysis including some of these subjects has previously been published from our laboratory.16 Laterality of seizure onset was made from prolonged ictal and interictal scalp video-electroencephalogram telemetry. Nine TLE-mts patients had a left temporal onset, and 6 TLE-mts patients had a right onset. In TLE-no, 7 patients were diagnosed with left temporal onset, and 7 patients had a right temporal onset. Patients were categorized as TLE-mts or TLE-no based on evidence of hippocampal atrophy and signal intensity changes on their 4T MR imaging that used an epilepsy-specific protocol, and all were reviewed by the same neuroradiologist (J.B.). Hippocampal volumetry was used to confirm the presence (TLE-mts) or absence (TLE-no) of significant hippocampal volume loss. Volumetry was performed on high-resolution T2-weighted hippocampal images by using a method of manual segmentation.20 There was a significant mean difference (P < .001) between the age at which TLE-mts and TLE-no patients developed epilepsy (TLE-mts, 5 ± 6.6 years; TLE-no, 23 ± 11.7 years) and also the duration of years patients have had epilepsy (TLE-mts, 34.6 ± 11.8 years; TLE-no, 17.1 ± 9.8 years).

Data Processing

MR Imaging Acquisition.

All subjects underwent MR imaging with an MedSpec 4T system (Bruker MedSpec, Madison, Wisconsin) controlled by a Trio console (Siemens, Erlangen, Germany) and equipped with an 8-channel array coil (USA Instruments, Aurora, Ohio). A volumetric T1-weighted magnetization-prepared rapid acquisition of gradient echo sequence was acquired with the following imaging parameters: 2300/3/950 ms (TR/TE/TI); flip angle of 7°; and 1 × 1 × 1-mm3 voxel resolution.

Image Processing

To combine left and right temporal onset patients in the analysis, MR imaging data for patients were reassembled according to brain hemisphere of seizure onset. Therefore, images of the subjects with right TLE onset were left-right flipped so that all subject ipsilateral hemispheres were on the left side. Because previous studies have revealed GM asymmetries between the left and right hemispheres in the normal population,21 customized symmetrical templates were generated for each analysis method from the same control subjects as described below.

Voxel-based Morphometry

T1 images were intensity inhomogeneity corrected and segmented into GM, WM, and CSF by using the EMS toolbox22 in SPM2 (http://www.fil.ion.ucl.ac.uk/spm), running in Matlab 6.1 (MathWorks, Natick, Massachusetts). The EMS toolbox was used as opposed to standard SPM2 due to its superior bias field correction algorithm that performs better with the more pronounced bias field of the 4T images. Optimized VBM was then carried out on probabilistic GM maps according to the optimized VBM protocol described in detail previously.12 To summarize, a study-specific symmetrical SPM GM template was first created by averaging 31 normalized (first affine followed by a nonlinear registration) control subjects (flipped and unflipped) and smoothing by using an isotropic 8-mm FWHM Gaussian kernel. All GM maps were then normalized to this symmetrical template first by using a 12-parameter affine transformation, followed by a nonlinear transformation to minimize the residual squared difference between the image and template.23 Voxel intensity values were multiplied (modulated) by the Jacobian determinant (ie, the local expansion/contraction factor) derived from the deformation map produced during spatial normalization, to preserve GM volume. The modulated GM images were smoothed with an isotropic 8-mm FWHM Gaussian kernel and used for statistical analysis.

Deformation-based Morphometry

The T1 images were skull-stripped, and intensity inhomogeneity was corrected by using the bias field generated by EMS. A study-specific symmetrical ULD template was created using the same 31 control subjects as for the VBM template by using a fluid registration algorithm.24 This template creation step is described in full elsewhere.25 To summarize briefly, it incorporates an unbiased approach in which each subject's flipped and nonflipped images are first simultaneously deformed to create a symmetrical subject brain. Symmetrical subject images are then simultaneously deformed iteratively into an average ULD template brain. This technique avoids the bias toward a particular subject's geometry introduced by selecting a single subject template.26 The ULD template also contains sharper features than the multisubject average intensity template as created in SPM.

Each subject's T1 image is first registered to the template with an affine 12-parameter transformation. Subject brains are then nonlinearly deformed to the template by using a fluid-flow warping technique.25 The Jacobian determinants of the deformation fields are used to gauge the local volume differences at each voxel between the individual image and the template. Each subject Jacobian map is smoothed with an isotropic 8-mm FWHM Gaussian kernel.

Cortical Thickness

Analysis of cortical thickness was carried out by using FS, version 3.05 (https://surfer.nmr.mgh.harvard.edu). Detailed descriptions of this method have already been published27,28 but are briefly summarized here. Based on a linear combination of voxel intensities and local geometric constraints, the cerebral WM is first segmented. The WM is divided into 2 hemispheres, and the brain stem and cerebellum are removed. Tessellation is then performed to produce a triangle-based mesh of the WM surface and refined to alleviate the voxel-based nature of the initial curvature. The WM surfaces are deformed outward to generate the pial (GM/CSF intersection) surface. Topologic defects in the surface are corrected by using an automated topology fixer. Visual quality checks are performed and inaccuracies are manually edited and corrected by reprocessing. The cortical surface is spherically inflated so that the entire cortical surface is exposed, including deep tissue inside the sulci. Using combined information from the pial and WM surfaces, cortical thickness is calculated at each vertex.

To perform surface-based analysis of cortical thickness between groups, a custom symmetrical template is constructed by using data from the same 31 control subjects as for VBM and DBM processing. To create a surface template, an average curvature map is created by averaging subject gyral and sulcal curvature patterns. Each study subject's surface is then registered to the template, and the deformation is guided by the cortical features of the template. Thickness data from each subject are then smoothed by using a 20-mm FWHM 2D Gaussian kernel and mapped to an average surface. This average surface is the average of all study subjects for the visualization of results and should not be confused with the template described above.

Statistical Analysis

Linear regression was performed to determine the effect of “group” (patient or control) on the measurement parameter at each voxel/vertex. ICV was entered as a nuisance variable for VBM analysis only because this was already accounted for during the initial affine registration of DBM and cortical thickness is not confounded by ICV.29 Contrasts were defined to detect differences at each voxel/vertex between 1) controls and TLE-mts and 2) controls and TLE-no. Given the large number of voxels/vertices being tested in each analysis, it is necessary to correct for multiple comparisons to reduce the probability of obtaining false-positives (type I errors). The most commonly used methods, such as random-field theory and FDR are dependent on the number of voxels/vertices tested. Given that SPM-VBM and DBM images contain approximately 2 million brain voxels, whereas FS provides cortical thickness information at 320,000 vertices on average, these methods of multiple comparison correction make it difficult to fairly compare results across the morphologic methods used in this study. Permutation analysis is a nonparametric technique that has been demonstrated to be an effective multiple comparison correction technique in neuroimaging30 and is also independent of the number of voxels/vertices tested. A null distribution for the effect of group at each voxel/vertex was constructed by using 10 000 random permutations of the data. For each test, the subject's diagnosis was randomly permuted, and t tests were conducted to identify voxels more significant than P=.05. The group differences more significant than P=.05 were computed for the real experiment and for the random assignments. Finally, a ratio, describing the fraction of the time the suprathreshold group difference was greater in the randomized maps than the real effect (the original labeling), was calculated and a new P value was reported for the significance at that point. Voxelwise analysis was conducted by using FSL's “randomize” tool (FMRIB Software Library, version 4; http://www.fmrib.ox.ac.uk/fsl). FS-CT analysis was conducted with the same parameters at each vertex by using FS's statistical analysis tool.

Results

TLE-mts versus Controls

Figure 1 displays GM volume loss (SPM-VBM), volume loss (DBM), and cortical thinning (FS-CT) in TLE-mts compared with controls after permutation correction. Both SPM-VBM and DBM show 1 large cluster of volume losses extending from the ipsilateral hippocampus and mesial temporal lobe to bilateral thalamus, brain stem, and cerebellum. In theSPM-VBM analysis, this cluster extends to the lateral temporo-occipital cortex (ipsilateral > contralateral). FS-CT demonstrates 1 cluster of cortical thinning of the ipsilateral temporo-occipital-parietal region. Table 1 outlines the size of the significant clusters and the maximum t-statistic. The largest cluster was found by using VBM analysis; however, DBM revealed the largest cluster t-statistic.

Fig 1.
  • Download figure
  • Open in new tab
  • Download powerpoint
Fig 1.

Controls versus TLE-mts. A, VBM-SPM GM differences. B, DBM Jacobian differences. C, FS-CT differences between groups. All results corrected for multiple comparisons by using permutation analysis (P < .05).

View this table:
  • View inline
  • View popup
Table 1:

TLE-mts versus controls

TLE-no versus Controls

Figure 2 shows volume loss (DBM) and cortical thinning (FS-CT) in TLE-no compared with controls after permutation correction. SPM-VBM analysis revealed no changes between groups after correction for multiple comparisons. FS-CT showed a large cluster of bilateral cortical thinning in both temporal lobes extending to the frontal, occipital, and parietal lobes bilaterally. DBM demonstrated 2 significant clusters. The first cluster included the ipsilateral temporal lobe, extending to the brain stem and cerebellum. The second cluster included the bilateral superior frontal cortex, pre- and postcentral cortex, and superior parietal cortex. Table 2 outlines the size of the significant clusters and the maximum t-statistic. Figure 3 shows SPM-VBM, DBM, and FS-CT results after FDR correction for both TLE-mts and TLE-no versus controls.

Fig 2.
  • Download figure
  • Open in new tab
  • Download powerpoint
Fig 2.

Controls versus TLE-no. A, VBM-SPM GM differences. B, DBM Jacobian differences. C, FS-CT differences between groups. All results corrected for multiple comparisons by using permutation analysis (P < .05).

View this table:
  • View inline
  • View popup
Table 2:

TLE-no versus controls

Fig 3.
  • Download figure
  • Open in new tab
  • Download powerpoint
Fig 3.

Controls versus TLE-mts and controls versus TLE-no after FDR < .05 correction for multiple comparisons. A, VBM-SPM GM differences. B, DBM Jacobian differences. C, FS-CT differences between controls and TLE-mts. D, VBM-SPM GM differences. E, DBM Jacobian differences. F, FS-CT differences between controls and TLE-no.

Discussion

In this study, we aimed to investigate the aspects of disease atrophy detected by 3 different automated methods of brain morphometry. There were 2 main findings: 1) In TLE-mts, voxel-based methods SPM-VBM and DBM identified the most pronounced volume losses in the ipsilateral hippocampus and mesial temporal region, the ipsi- and contralateral thalamus, and cerebellum. FS-CT showed cortical thinning in the ipsilateral temporo-occipital region. 2) SPM-VBM showed no significant volume loss in TLE-no. In contrast, DBM detected a region of ipsilateral temporal, bilateral superior frontal, and cerebellar volume loss. The most widespread changes covering the bilateral temporal, frontal, occipital, and parietal cortex were identified by using FS-CT. Based on these findings, we conclude that VBM, DBM, and FS-CT detect different types of atrophy and thus that the choice of the volumetry method should be guided by the knowledge about the disease process.

TLE-mts

Significant findings after FDR correction were consistent with previous studies, the clinical implications of which have been demonstrated previously and are not discussed here.9,15,16,18,31,32 The results differ when permutation analysis was used to correct for multiple comparisons instead of the more commonly used FDR approach. These methods of multiple comparison correction are based on different principles (see Statistical Analysis). The permuted results were used to compare between methods because this method is independent of the number of voxels/vertices tested and therefore represents a more unbiased approach.

When comparing across the methodologies, the most prominent volume changes in TLE-mts were demonstrated below the cortex and identified by using both SPM-VBM and DBM. Both methodologies detected the pathology because the observed changes are macroscopic, and the ability of DBM to detect subtle morphologic changes is not essential. In addition, however, the SPM-VBM cluster extended to the ipsilateral temporal-occipital and contralateral occipital cortex that was not detected by DBM or by previous VBM studies. Although this cortical finding may seem counterintuitive when DBM registration is theoretically more accurate than SPM-VBM, there are several possible explanations. 1) An increased intersubject variance in the cortical region by using DBM-SPM-VBM coregistration corrects macroscopic volume effects but maintains most of the individual morphologic differences at the gyral level. DBM, as used in this study, corrects most of the physiologic and disease-related interindividual differences at the gyral level, resulting in a higher between subject variability of the transformation matrix and thus a lower power to detect disease-related differences. 2) Although corrected for head size, the resulting deformation matrix of SPM-VBM contains the affine and the nonlinear transformations, whereas the affine coregistration is not integrated in the DBM transformation matrix. This would suggest that the significant cortical cluster may be a reflection of the volume decrease in the deep WM of the temporal lobe.

FS-CT detected a region of cortical thinning in the temporo-occipital lobe also detected by SPM-VBM. However, volume loss detected by SPM-VBM and DBM mostly affected the hippocampus and subcortical structures that are not part of the FS-CT analysis. These results suggest that TLE-mts is a disease with mostly macroscopic volume loss affecting a large region predominantly including the ipsilateral hippocampus, thalami, and cerebellum, and extending to the lateral temporal cortex, which may be detected best by using VBM analysis. Cortical thinning is present in TLE-mts but, at least in this analysis, is less prominent than the subcortical volume losses that might suggest that the cortical thinning is secondary to the subcortical atrophy.

TLE-no

The 4T VBM findings presented here were in agreement with previous VBM studies at 1.5T, demonstrating no significant findings after FDR correction.11,19 The DBM findings were consistent with some previous region of interest volume analyses that show ipsilateral temporal lobe atrophy.17,18 The FDR corrected cortical thickness results for TLE-no have been reported previously and discussed by our laboratory,16 but they have not been demonstrated elsewhere.

In the permutation analysis of TLE-no patients versus controls, DBM found significant changes affecting the ipsilateral temporal lobe, cerebellum, and brain stem in 1 cluster, and the superior frontal and parietal cortex in a second cluster. No significant volume loss was detected by using SPM-VBM. This suggests that the fluid-registration algorithm used by DBM may be better suited to detect the subtle volumetric abnormalities associated with this disease type than the spatial basis function algorithm implemented in SPM2. A region of interest–based DBM study focused on the thalamus has previously been carried out on the same study subjects in our laboratory.33 This study demonstrated a subtle volume loss in this region. If it is necessary to study subcortical structures in disease types with very subtle volume changes such as TLE-no, it may be beneficial to have an a priori hypothesis to avoid having to correct for multiple comparisons across the whole brain, where such a finding may not survive a stringent correction.

FS-CT analysis revealed the most widespread cortical thinning in TLE-no affecting the bilateral temporal lobes, insula, and frontal and occipital lobes. Only the thinning of the superior frontal cortex also was detected by DBM. The reason for this may be physiologic, methodologic, or a combination: 1) Physiologic: Subtle thinning of the cortex associated with TLE-no is not detected through macroscopic voxel-based volume analysis. The nature of the abnormalities associated with TLE-no may be primarily confined to the thinning of the cerebral cortex, subtle effects of which may be diluted in analysis of volume (a combination of thickness and surface area). In TLE-mts, however, the cortical changes are large and hence can be detected through analysis of either volume or FS-CT. 2) Methodologic: Image registration methods of matching gyral patterns help to precisely colocalize identical cortical regions of subjects without sacrificing the details contained in the underlying measure of interest (ie, FS-CT). Matching intensities, as with SPM-VBM and DBM, is less anatomically meaningful and may lead to a less precise colocalization of subject cortical region and a decrease in the power to detect structural changes in a disease group whose cortical abnormalities are already quite subtle. In addition, the 3D smoothing step used by the voxel-based methods leads to a blurring of tissue across neighboring banks of a sulcus that are not anatomically related and thus could again lead to a reduced ability to detect cortical abnormalities. FS smoothing however is performed across the 2D inflated brain surface, preserving the relationship between neighboring sulcal and gyral structures.

Limitations

There are several methodologic differences between the techniques that have not been controlled: 1) The methods use different segmentation techniques (probabilistic in SPM-VBM versus binary in FS-CT). 2) Each method requires smoothing for statistical analysis, but because of the different smoothing approaches, it is difficult to determine a comparable smoothing kernel size for surface-based and voxel-based methods. Therefore, in this study, recommendations were taken from previous studies for both voxel- and surface-based smoothing kernel sizes. 3) A more recent SPM toolbox, DARTEL (compatible with SPM, versions 5 and higher) includes an improved atlas creation and registration method that adopts a diffeomorphic registration algorithm that is similar to the DBM approach used in this study.34 Theoretically, this would lead to more comparable results between VBM-SPM and DBM and is currently being tested in our laboratory. 4) Effort was made to remove brain dura during the skull stripping process before DBM analysis, particularly at the most superior part of the interhemispheric fissure. However, this was difficult to accomplish in some subjects without the removal of cortical voxels. Although the percentage of subjects with some remaining dura between controls and TLE-no subjects was equal, it cannot be disregarded that the superior frontal cortical cluster may be due to this artifact and should be considered a limitation of this method. 5) None of the 3 methods depicts the “whole truth,” ie, just because a brain region seems normal in a VBM or cortical thickness analysis, it cannot be concluded for sure that this structure is not affected by the disease process but only that it is less likely to be affected by the type of pathologic abnormalities to which the chosen method is particularly sensitive.

Conclusions

The findings of this study show that each of the 3 methods detects different types of structural abnormalities and that the choice of the method has to be guided by the nature of the suspected pathology. Some of the differences are obvious because they are inherent to the method, eg, FS-CT is not suited to detect subcortical abnormalities because these structures are not part of the cortex. However, the results also can differ in structures that are assessed by all 3 methods. Based on the findings in this study, we conclude that SPM-VBM and DBM will detect cortical and subcortical abnormalities in diseases associated by macroscopic volume losses and that FS-CT will detect the cortical component. In diseases without macroscopic volume losses, FS-CT is the optimal method to detect cortical abnormalities. DBM is the optimal method to detect subcortical abnormalities, but DBM also will pick up some of the cortical pathology in these diseases.

Footnotes

  • This work was supported by the National Institutes of Health grant R01NS0311966 to K.D.L.

  • Disclosures: Paul Garcia: Research Support (including provision of equipment or materials): Medtronic, UCB Pharma Details: RCT of thalamic stimulation (Medtronics) and brivaracetam (UCB Pharma). Michael Weiner: Research Support (including provision of equipment or materials): NIH, VA, DOD, Merck, Avid Consultant: Astra Zeneca, Araclon, Medivation, Pfizer, Ipsen, TauRx, Bayer Healthcare, Biogen Idec, Exonhit Therapeutics, Servier, Synarc.

Indicates open access to non-subscribers at www.ajnr.org

References

  1. 1.↵
    1. Mechelli A,
    2. Price CJ,
    3. Friston KJ,
    4. et al
    . Voxel-based morphometry of the human brain: methods and applications. Curr Med Imaging Rev 2005; 1: 105–13
    CrossRef
  2. 2.↵
    1. Wright IC,
    2. McGuire PK,
    3. Poline JB,
    4. et al
    . A voxel-based method for the statistical analysis of gray and white matter density applied to schizophrenia. Neuroimage 1995; 2: 244–52
    CrossRefPubMed
  3. 3.↵
    1. Ashburner J,
    2. Hutton C,
    3. Frackowiak R,
    4. et al
    . Identifying global anatomical differences: deformation-based morphometry. Neuroimage 1998; 6: 348–57
  4. 4.↵
    1. Davatzikos C,
    2. Vaillant M,
    3. Resnick SM,
    4. et al
    . A computerised approach for morphological analysis of the corpus callosum. J Comput Tomogr 1996; 20: 88–97
    CrossRef
  5. 5.↵
    1. Fischl B,
    2. Dale AM
    . Measuring the thickness of the human cerebral cortex from magnetic resonance images. Proc Natl Acad Sci U S A 2000; 97: 11050–55
    Abstract/FREE Full Text
  6. 6.↵
    1. Lerch JP,
    2. Evans AC
    . Cortical thickness analysis examined through power analysis and a population simulation. Neuroimage 2005; 24: 163–73
    CrossRefPubMed
  7. 7.↵
    1. Bonilha L,
    2. Edwards JC,
    3. Kinsman SL,
    4. et al
    . Extrahippocampal gray matter loss and hippocampal deafferentation in patients with temporal lobe epilepsy. Epilepsia 2010; 51: 519–28
    CrossRefPubMed
  8. 8.↵
    1. Eriksson SH,
    2. Thom M,
    3. Symms MR,
    4. et al
    . Cortical neuronal loss and hippocampal sclerosis are not detected by voxel-based morphometry in individual epilepsy surgery patients. Hum Brain Mapp 2009; 30: 3351–60
    CrossRefPubMed
  9. 9.↵
    1. Keller SS,
    2. Roberts N
    . Voxel-based morphometry of temporal lobe epilepsy: an introduction and review of the literature. Epilepsia 2008; 49: 741–57
    CrossRefPubMed
  10. 10.↵
    1. Labate A,
    2. Cerasa A,
    3. Aguglia U,
    4. et al
    . Voxel-based morphometry of sporadic epileptic patients with mesiotemporal sclerosis. Epilepsia 2010; 51: 506–10
    CrossRefPubMed
  11. 11.↵
    1. Riederer F,
    2. Lanzenberger R,
    3. Kaya M,
    4. et al
    . Network atrophy in temporal lobe epilepsy: a voxel-based morphometry study. Neurology 2008; 71: 419–25
    Abstract/FREE Full Text
  12. 12.↵
    1. Good CD,
    2. Johnsrude IS,
    3. Ashburner J,
    4. et al
    . A voxel-based morphometric study of ageing in 465 normal adult human brains. Neuroimage 2001a; 14: 21–36
    CrossRefPubMed
  13. 13.↵
    1. Apostolova LG,
    2. Thompson PM
    . Brain mapping as a tool to study neurodegeneration. Neurotherapeutics 2007; 4: 387–400
    CrossRefPubMed
  14. 14.↵
    1. Studholme C,
    2. Cardenas V,
    3. Blumenfeld R,
    4. et al
    . Deformation tensor morphometry of semantic dementia with quantitative validation. Neuroimage 2004; 21: 1387–98
    CrossRefPubMed
  15. 15.↵
    1. Bernhardt BC,
    2. Worsley KJ,
    3. Besson P,
    4. et al
    . Mapping limbic network organization in temporal lobe epilepsy using morphometric correlations: insights on the relation between mesiotemporal connectivity and cortical atrophy. Neuroimage 2008; 42: 515–24
    CrossRefPubMed
  16. 16.↵
    1. Mueller SG,
    2. Laxer KD,
    3. Barakos J,
    4. et al
    . Widespread neocortical abnormalities in temporal lobe epilepsy with and without mesial sclerosis. Neuroimage 2009; 46: 353–59
    CrossRefPubMed
  17. 17.↵
    1. Bernasconi N,
    2. Bernasconi A,
    3. Caramanos Z,
    4. et al
    . Entorhinal cortex atrophy in epilepsy patients exhibiting normal hippocampal volumes. Neurology 2001; 56: 1335–39
    Abstract/FREE Full Text
  18. 18.↵
    1. Mueller SG,
    2. Laxer KD,
    3. Schuff N,
    4. et al
    . Voxel-based T2 relaxation rate measurements in temporal lobe epilepsy (TLE) with and without mesial temporal sclerosis. Epilepsia 2007; 48: 220–28
    CrossRefPubMed
  19. 19.↵
    1. Mueller SG,
    2. Laxer KD,
    3. Cashdollar N,
    4. et al
    . Voxel-based optimized morphometry (VBM) of gray and white matter in temporal lobe epilepsy (TLE) with and without mesial temporal sclerosis. Epilepsia 2006; 47: 900–07
    CrossRefPubMed
  20. 20.↵
    1. Mueller SG,
    2. Laxer KD,
    3. Barakos J,
    4. et al
    . Subfield atrophy pattern in temporal lobe epilepsy with and without mesial sclerosis detected by high-resolution MRI at 4 Tesla: preliminary results. Epilepsia 2009; 50: 1474–84
    CrossRefPubMed
  21. 21.↵
    1. Good CD,
    2. Johnsrude IS,
    3. Ashburner J,
    4. et al
    . Cerebral asymmetry and the effects of sex and handedness on brain structure: a voxel based morphometric analysis of 465 normal adult human brains. Neuroimage 2001b; 14: 685–700
    CrossRefPubMed
  22. 22.↵
    1. Van Leemput K,
    2. Maes F,
    3. Vandermeulen D,
    4. et al
    . Automated model based bias field correction of MR images of the brain. IEEE Trans Med Imag 1999; 18: 885–96
    CrossRefPubMed
  23. 23.↵
    1. Ashburner J,
    2. Friston KJ
    . Voxel-based morphometry—the methods. Neuroimage 2000; 11: 805–21
    CrossRefPubMed
  24. 24.↵
    1. Christensen GE,
    2. Rabbitt RD,
    3. Miller MI
    . Deformable templates using large deformation kinematics. IEEE Trans Image Process 1996; 5: 1435–47
    CrossRefPubMed
  25. 25.↵
    1. Lorenzen P,
    2. Prastawa M,
    3. Davis B,
    4. et al
    . Multi-modal image set registration and atlas formation. Med Image Anal 2006; 10: 440–51
    CrossRefPubMed
  26. 26.↵
    1. Joshi S,
    2. Davis B,
    3. Jomier M,
    4. et al
    . Unbiased diffeomorphic atlas construction for computational anatomy. Neuroimage 2004; 23: 151–60
    CrossRef
  27. 27.↵
    1. Dale AM,
    2. Fischl B,
    3. Sereno MI
    . Cortical surface-based analysis I: segmentation and surface reconstruction. Neuroimage 1999; 9: 179–94
    CrossRefPubMed
  28. 28.↵
    1. Fischl B,
    2. Sereno MI,
    3. Dale AM
    . Cortical surface-based analysis II: inflation, flattening, and a surface-based coordinate system. Neuroimage 1999; 9: 195–207
    CrossRefPubMed
  29. 29.↵
    1. Barnes J,
    2. Ridgway GR,
    3. Bartlett J,
    4. et al
    . Head size, age and gender adjustment in MRI studies: a necessary nuisance? Neuroimage 2010; 53: 1244–55
    CrossRefPubMed
  30. 30.↵
    1. Nichols TE,
    2. Holmes AP
    . Nonparametric permutation tests for functional neuroimaging: a primer with examples. Hum Brain Mapp 2001; 15: 1–25
    CrossRef
  31. 31.↵
    1. Lin JJ,
    2. Salamon N,
    3. Lee AD,
    4. et al
    . Reduced neocortical thickness and complexity mapped in mesial temporal lobe epilepsy with hippocampal sclerosis. Cereb Cortex 2007; 17: 2007–18
    Abstract/FREE Full Text
  32. 32.↵
    1. McDonald CR,
    2. Hagler DJ,
    3. Ahmadi ME,
    4. et al
    . Regional neocortical thinning in mesial temporal lobe epilepsy. Epilepsia 2008; 49: 794–803
    CrossRefPubMed
  33. 33.↵
    1. Mueller SG,
    2. Laxer KD,
    3. Barakos J,
    4. et al
    . Involvement of the thalamocortical network in TLE with and without mesiotemporal sclerosis. Epilepsia 2010; 51: 1436–45
    PubMed
  34. 34.↵
    1. Ashburner J
    . A fast diffeomorphic image registration algorithm. Neuroimage 2007; 38: 95–113
    CrossRefPubMed
  • Received November 1, 2011.
  • Accepted after revision January 25, 2011.
  • © 2011 by American Journal of Neuroradiology
PreviousNext
Back to top

In this issue

American Journal of Neuroradiology: 32 (9)
American Journal of Neuroradiology
Vol. 32, Issue 9
1 Oct 2011
  • Table of Contents
  • Index by author
Advertisement
Print
Download PDF
Email Article

Thank you for your interest in spreading the word on American Journal of Neuroradiology.

NOTE: We only request your email address so that the person you are recommending the page to knows that you wanted them to see it, and that it is not junk mail. We do not capture any email address.

Enter multiple addresses on separate lines or separate them with commas.
Impact of Methodologic Choice for Automatic Detection of Different Aspects of Brain Atrophy by Using Temporal Lobe Epilepsy as a Model
(Your Name) has sent you a message from American Journal of Neuroradiology
(Your Name) thought you would like to see the American Journal of Neuroradiology web site.
CAPTCHA
This question is for testing whether or not you are a human visitor and to prevent automated spam submissions.
Cite this article
C. Scanlon, S.G. Mueller, D. Tosun, I. Cheong, P. Garcia, J. Barakos, M.W. Weiner, K.D. Laxer
Impact of Methodologic Choice for Automatic Detection of Different Aspects of Brain Atrophy by Using Temporal Lobe Epilepsy as a Model
American Journal of Neuroradiology Oct 2011, 32 (9) 1669-1676; DOI: 10.3174/ajnr.A2578

Citation Manager Formats

  • BibTeX
  • Bookends
  • EasyBib
  • EndNote (tagged)
  • EndNote 8 (xml)
  • Medlars
  • Mendeley
  • Papers
  • RefWorks Tagged
  • Ref Manager
  • RIS
  • Zotero
0 Responses
Respond to this article
Share
Bookmark this article
Impact of Methodologic Choice for Automatic Detection of Different Aspects of Brain Atrophy by Using Temporal Lobe Epilepsy as a Model
C. Scanlon, S.G. Mueller, D. Tosun, I. Cheong, P. Garcia, J. Barakos, M.W. Weiner, K.D. Laxer
American Journal of Neuroradiology Oct 2011, 32 (9) 1669-1676; DOI: 10.3174/ajnr.A2578
del.icio.us logo Twitter logo Facebook logo Mendeley logo
  • Tweet Widget
  • Facebook Like
  • Google Plus One
Purchase

Jump to section

  • Article
    • Abstract
    • Abbreviations
    • Materials and Methods
    • Results
    • Discussion
    • Conclusions
    • Footnotes
    • References
  • Figures & Data
  • Info & Metrics
  • Responses
  • References
  • PDF

Related Articles

  • No related articles found.
  • PubMed
  • Google Scholar

Cited By...

  • Frontotemporal Dementia Subtyping using Machine Learning, Multivariate Statistics, and Neuroimaging
  • Association of specific biotypes in patients with Parkinson disease and disease progression
  • Spatial patterning of tissue volume loss in schizophrenia reflects brain network architecture
  • Crossref (25)
  • Google Scholar

This article has been cited by the following articles in journals that are participating in Crossref Cited-by Linking.

  • Network structure of brain atrophy in de novo Parkinson&#039;s disease
    Yashar Zeighami, Miguel Ulla, Yasser Iturria-Medina, Mahsa Dadar, Yu Zhang, Kevin Michel-Herve Larcher, Vladimir Fonov, Alan C Evans, D Louis Collins, Alain Dagher
    eLife 2015 4
  • Grey and white matter abnormalities in temporal lobe epilepsy with and without mesial temporal sclerosis
    Cathy Scanlon, Susanne G. Mueller, Ian Cheong, Miriam Hartig, Michael W. Weiner, Kenneth D. Laxer
    Journal of Neurology 2013 260 9
  • Sex effects on brain structure in de novo Parkinson’s disease: a multimodal neuroimaging study
    Christina Tremblay, Nooshin Abbasi, Yashar Zeighami, Yvonne Yau, Mahsa Dadar, Shady Rahayel, Alain Dagher
    Brain 2020 143 10
  • How reliable are gray matter disruptions in specific reading disability across multiple countries and languages? insights from a large‐scale voxel‐based morphometry study
    Katarzyna Jednoróg, Artur Marchewka, Irene Altarelli, Ana Karla Monzalvo Lopez, Muna van Ermingen‐Marbach, Marion Grande, Anna Grabowska, Stefan Heim, Franck Ramus
    Human Brain Mapping 2015 36 5
  • Sex is a defining feature of neuroimaging phenotypes in major brain disorders
    Lauren E. Salminen, Meral A. Tubi, Joanna Bright, Sophia I. Thomopoulos, Alyssa Wieand, Paul M. Thompson
    Human Brain Mapping 2022 43 1
  • Cortical Thickness in Children Receiving Intensive Therapy for Idiopathic Apraxia of Speech
    Darren S. Kadis, Debra Goshulak, Aravind Namasivayam, Margit Pukonen, Robert Kroll, Luc F. De Nil, Elizabeth W. Pang, Jason P. Lerch
    Brain Topography 2014 27 2
  • Asymmetric cortical surface area and morphology changes in mesial temporal lobe epilepsy with hippocampal sclerosis
    Saud Alhusaini, Colin P. Doherty, Lena Palaniyappan, Cathy Scanlon, Sinead Maguire, Paul Brennan, Norman Delanty, Mary Fitzsimons, Gianpiero L. Cavalleri
    Epilepsia 2012 53 6
  • Regional transcriptional architecture of Parkinson’s disease pathogenesis and network spread
    Benjamin Freeze, Sneha Pandya, Yashar Zeighami, Ashish Raj
    Brain 2019 142 10
  • Cortical thinning and caudate abnormalities in first episode psychosis and their association with clinical outcome
    Cathy Scanlon, Heike Anderson-Schmidt, Liam Kilmartin, Shane McInerney, Joanne Kenney, John McFarland, Mairead Waldron, Srinath Ambati, Anna Fullard, Sam Logan, Brian Hallahan, Gareth J. Barker, Mark A. Elliott, Peter McCarthy, Dara M. Cannon, Colm McDonald
    Schizophrenia Research 2014 159 1
  • Regional brain atrophy and cognitive decline depend on definition of subjective cognitive decline
    Cassandra Morrison, Mahsa Dadar, Neda Shafiee, Sylvia Villeneuve, D. Louis Collins
    NeuroImage: Clinical 2022 33

More in this TOC Section

  • Optimal MRI Sequence for Identifying Occlusion Location in Acute Stroke: Which Value of Time-Resolved Contrast-Enhanced MRA?
  • SWI or T2*: Which MRI Sequence to Use in the Detection of Cerebral Microbleeds? The Karolinska Imaging Dementia Study
  • Progression of Microstructural Damage in Spinocerebellar Ataxia Type 2: A Longitudinal DTI Study
Show more Brain

Similar Articles

Advertisement

Indexed Content

  • Current Issue
  • Accepted Manuscripts
  • Article Preview
  • Past Issues
  • Editorials
  • Editors Choice
  • Fellow Journal Club
  • Letters to the Editor

Cases

  • Case Collection
  • Archive - Case of the Week
  • Archive - Case of the Month
  • Archive - Classic Case

Special Collections

  • Special Collections

Resources

  • News and Updates
  • Turn around Times
  • Submit a Manuscript
  • Author Policies
  • Manuscript Submission Guidelines
  • Evidence-Based Medicine Level Guide
  • Publishing Checklists
  • Graphical Abstract Preparation
  • Imaging Protocol Submission
  • Submit a Case
  • Become a Reviewer/Academy of Reviewers
  • Get Peer Review Credit from Publons

Multimedia

  • AJNR Podcast
  • AJNR SCANtastic
  • Video Articles

About Us

  • About AJNR
  • Editorial Board
  • Not an AJNR Subscriber? Join Now
  • Alerts
  • Feedback
  • Advertise with us
  • Librarian Resources
  • Permissions
  • Terms and Conditions

American Society of Neuroradiology

  • Not an ASNR Member? Join Now

© 2025 by the American Society of Neuroradiology All rights, including for text and data mining, AI training, and similar technologies, are reserved.
Print ISSN: 0195-6108 Online ISSN: 1936-959X

Powered by HighWire