Intrinsic Variability of GM Density Maps: Its Implications to VBM Studies
1
Methods Methods The T1 weighted whole head MRI scans were The T1 weighted whole head MRI scans were acquired using a GE Sigma 1.5T scanner (SPGR) acquired using a GE Sigma 1.5T scanner (SPGR) and a Philips Achieva 3T scanner (MPRAGE) and a Philips Achieva 3T scanner (MPRAGE) scanners. All subjects were young adults with scanners. All subjects were young adults with no history of neurological disorders. All no history of neurological disorders. All scans from one session were averaged to obtain scans from one session were averaged to obtain a low-noise structural image, and a single a low-noise structural image, and a single ‘best’ normalization transformation was ‘best’ normalization transformation was estimated from it. Therefore, normalized estimated from it. Therefore, normalized results were not directly affected by possible results were not directly affected by possible variability among normalization transforms variability among normalization transforms estimated from individual scans. Identical T1 estimated from individual scans. Identical T1 scans were repeated four to six times during scans were repeated four to six times during the same session. In order to compare and the same session. In order to compare and generalize our results, we also analyzed data generalize our results, we also analyzed data sets available to the scientific community as sets available to the scientific community as part of the OASIS project (Open Access Series part of the OASIS project (Open Access Series of Imaging Studies, of Imaging Studies, http://www.oasis-brains.org http://www.oasis-brains.org / / ). The OASIS project includes a data set ). The OASIS project includes a data set containing 20 normal subjects imaged four times containing 20 normal subjects imaged four times during a session and repeated on a subsequent during a session and repeated on a subsequent visit within 90 days of their initial session. visit within 90 days of their initial session. The data were processed using a unified The data were processed using a unified segmentation/normalization framework (Ashburner segmentation/normalization framework (Ashburner et. al 2005) as implemented in the SPM5 et. al 2005) as implemented in the SPM5 software package ( software package ( http:// http:// www.fil.ion.ucl.ac.uk/spm www.fil.ion.ucl.ac.uk/spm / / ) and VBM5 toolkit ( ) and VBM5 toolkit ( http://dbm.neuro.uni-jena.de/vbm/vbm5-for-spm5/ http://dbm.neuro.uni-jena.de/vbm/vbm5-for-spm5/ ). We used the default software settings and ). We used the default software settings and analysis parameters of SPM5, unless noted analysis parameters of SPM5, unless noted otherwise. Multiple scans obtained within the otherwise. Multiple scans obtained within the same session were coregistered and segmented same session were coregistered and segmented independently in each subject’s native space. independently in each subject’s native space. Following conventional VBM analysis (Ashburner Following conventional VBM analysis (Ashburner and and Friston, Friston, 2000, 2005), the segmented images 2000, 2005), the segmented images were smoothed using 12 mm Gaussian kernel, were smoothed using 12 mm Gaussian kernel, yielding the measure of GM density. The GM yielding the measure of GM density. The GM density maps were normalized and modulated density maps were normalized and modulated using a single normalization transform for using a single normalization transform for multiple scans.. Variability of GM density was multiple scans.. Variability of GM density was assessed by computing the standard deviation assessed by computing the standard deviation (SD) of GM density across individual scans (SD) of GM density across individual scans Figure 3. Between Session Variability Variability between the scan and rescan session (OASIS data set). The GM density variability maps were calculated using averages of four scans obtained during each session to reduce the effect of within session variability. Intrinsic Variability of GM Density Intrinsic Variability of GM Density Maps: Maps: Its Implications to VBM Studies Its Implications to VBM Studies Andrew Poliakov Andrew Poliakov 1 1 , Judy McLaughlin , Judy McLaughlin 2 2 , Geoffrey Valentine , Geoffrey Valentine 2 2 , Ilona Pitkanen , Ilona Pitkanen 2 2 and and Lee Osterhout Lee Osterhout 2 2 1 Department of Biological Structure and Department of Biological Structure and 2 Department of Psychology, University of Washington, Department of Psychology, University of Washington, Seattle, WA Seattle, WA Introduction Introduction Voxel-based morphometry (VBM) is an unbiased, Voxel-based morphometry (VBM) is an unbiased, objective neuroimaging technique for objective neuroimaging technique for identifying structural changes in the brain. identifying structural changes in the brain. VBM involves a voxel-wise comparison of the VBM involves a voxel-wise comparison of the local concentration of gray matter (GM) in local concentration of gray matter (GM) in whole brain MRI scans. Recent VBM studies have whole brain MRI scans. Recent VBM studies have investigated changes in GM density due to investigated changes in GM density due to learning and practice. Such changes are learning and practice. Such changes are expected to be small, thereby imposing expected to be small, thereby imposing stringent requirements on VBM sensitivity. stringent requirements on VBM sensitivity. However, VBM sensitivity is not uniform However, VBM sensitivity is not uniform throughout the brain. Consequently, GM density throughout the brain. Consequently, GM density changes might be more observable in some changes might be more observable in some regions of the brain than in others. Here, we regions of the brain than in others. Here, we explore three sources of variability in VBM explore three sources of variability in VBM sensitivity: sensitivity: (1) variability introduced by choices in (1) variability introduced by choices in imaging protocols; imaging protocols; (2) within session variability; (3) between (2) within session variability; (3) between session variability. session variability. Figure 1. Variability of GM Density: Single Subject Data An example of single subject data acquired on Philips Achieva 3T scanner (scans were repeated four times during each session). The two columns correspond to two different versions of MPRAGE protocol (Left : TR=7.5 ms, TE = 3.5, ms Flip angle 8 0 ; Sagital, 0.86 x 0.86 x 1 mm 3 ; Right: TR=7.4 ms TE = 3.44 ms Flip angle 8 0 ; Coronal, 0.92 x 0.92 x 1 mm 3 ). A. The average of four T1 scans (normalized). B. GM segmentation of this image (normalized and modulated). C. The mean GM density map with variability map overlaid (red). Analysis shows noticeable difference in both pattern and magnitude of variability values. Discussion Discussion VBM can be a powerful tool for studying subtle VBM can be a powerful tool for studying subtle differences in GM density distributions that differences in GM density distributions that reflect anatomical correlates of cognitive reflect anatomical correlates of cognitive parameters (Maguire et al., 2000, Gaser et al., parameters (Maguire et al., 2000, Gaser et al., 2003). The objective of such VBM studies is to 2003). The objective of such VBM studies is to detect statistically significant changes in GM detect statistically significant changes in GM density. However, segmentation accuracy is density. However, segmentation accuracy is affected by a number of factors which can lead affected by a number of factors which can lead to variability in the estimation of GM density to variability in the estimation of GM density obtained under identical conditions (i.e. same obtained under identical conditions (i.e. same subject, scanner and protocol). subject, scanner and protocol). Some of the factors affecting segmentation Some of the factors affecting segmentation accuracy result from noise and imaging accuracy result from noise and imaging artifacts; these factors largely depend on the artifacts; these factors largely depend on the scanner and the data acquisition protocol. Our scanner and the data acquisition protocol. Our analysis indicates that even slightly different analysis indicates that even slightly different data acquisition protocols on the same scanner data acquisition protocols on the same scanner can produce noticeably different patterns in can produce noticeably different patterns in the magnitude and distribution of GM the magnitude and distribution of GM variability maps (see Fig.1). These differing variability maps (see Fig.1). These differing distributions could potentially affect the distributions could potentially affect the results and the conclusions of VBM studies, results and the conclusions of VBM studies, leading to significantly different findings for leading to significantly different findings for the same experimental paradigm. the same experimental paradigm. On the other hand, some cortical areas may On the other hand, some cortical areas may consistently show increased variability in GM consistently show increased variability in GM density due to relatively complex anatomy density due to relatively complex anatomy and/or the persistence of imaging artifacts. and/or the persistence of imaging artifacts. For example, the tip of the temporal lobe For example, the tip of the temporal lobe consistently showed higher variability in all consistently showed higher variability in all data sets we analyzed (see Figs. 2 and 3). data sets we analyzed (see Figs. 2 and 3). This is not unexpected considering the This is not unexpected considering the proximity of non-brain tissues of similar proximity of non-brain tissues of similar intensities, as well as the possibility of intensities, as well as the possibility of blood flow, eye movement and susceptibility blood flow, eye movement and susceptibility artifacts. We noticed that very few VBM artifacts. We noticed that very few VBM studies reported GM density changes it this studies reported GM density changes it this area. In contrast, the parietal and occipital area. In contrast, the parietal and occipital cortex are not subject to these or other common cortex are not subject to these or other common imaging artifacts. As a result, these areas imaging artifacts. As a result, these areas show little variability in GM density. show little variability in GM density. Interestingly, the parietal and occipital Interestingly, the parietal and occipital cortices are the areas that have been cortices are the areas that have been implicated in a number of VBM studies. implicated in a number of VBM studies. We suggest that intrinsic variability of GM We suggest that intrinsic variability of GM density should be taken into consideration when density should be taken into consideration when interpreting the results of VBM studies. interpreting the results of VBM studies. Variability analysis may be a useful tool for Variability analysis may be a useful tool for designing and planning a VBM study. It may designing and planning a VBM study. It may help identify problematic areas, detect subtle help identify problematic areas, detect subtle imaging artifacts and help refine data imaging artifacts and help refine data acquisition and analysis. acquisition and analysis. References References • Ashburner J. and Friston K.J. Unified segmentation. Ashburner J. and Friston K.J. Unified segmentation. NeuroImage, 26:839-851, 2005. NeuroImage, 26:839-851, 2005. • Ashburner J. and Friston. K.J. Voxel-Based Ashburner J. and Friston. K.J. Voxel-Based Morphometry -- The Methods. NeuroImage, 11:805-821, Morphometry -- The Methods. NeuroImage, 11:805-821, 2000. 2000. • Maguire EA, Gadian DG, Johnsrude IS, et al. Maguire EA, Gadian DG, Johnsrude IS, et al. Navigation-related structural change in the hippocampi Navigation-related structural change in the hippocampi of taxi drivers. Proc Natl Acad Sci USA. 2000; 97: of taxi drivers. Proc Natl Acad Sci USA. 2000; 97: 4398-4403. 4398-4403. • Gaser C, Schlaug G. Brain structures differ between Gaser C, Schlaug G. Brain structures differ between musicians and non-musicians. J Neurosci 2003; 23: musicians and non-musicians. J Neurosci 2003; 23: 9240-9245. 9240-9245. • Draganski B, Gaser C, Busch V, Schuierer G, Bogdahn Draganski B, Gaser C, Busch V, Schuierer G, Bogdahn U, May A. Neuroplasticity: changes in grey matter U, May A. Neuroplasticity: changes in grey matter induced by training. Nature 2004; 427: 311-2. induced by training. Nature 2004; 427: 311-2. • Mechelli A, Crinion JT, Noppeney U, et al. Mechelli A, Crinion JT, Noppeney U, et al. Neurolinguistics: structural plasticity in the Neurolinguistics: structural plasticity in the bilingual brain. Nature 2004; 431: 757. bilingual brain. Nature 2004; 431: 757. Acknowledgements Supported by NIH grants R01DC01947 and P30DC04661 Figure 2. Within Session Variability of GM Density: Averaged Across Subject Groups The variability maps of GM Density averaged across subject groups are color coded and superimposed onto GM probability density maps. Both cases show that that variability was strongly non-homogenious throughout the brain. The patterns appear to be quite different reflecting the differences in the data acquisition protocol, artifact correction techniques etc. A. Data set obtained using GE Signa 1.5T (four subjects, 2 sessions of 6 scans, SPGR). B. OASIS data set (15 subjects, two session of four scans, MPRAGE). Note that GE Signa was an older scanner, and modern systems and data acquisition protocols generally produce better image quality . A B C A B
description
1 Department of Biological Structure and 2 Department of Psychology, University of Washington, Seattle, WA. Introduction - PowerPoint PPT Presentation
Transcript of Intrinsic Variability of GM Density Maps: Its Implications to VBM Studies
MethodsMethods
The T1 weighted whole head MRI scans were acquired using a The T1 weighted whole head MRI scans were acquired using a GE Sigma 1.5T scanner (SPGR) and a Philips Achieva 3T GE Sigma 1.5T scanner (SPGR) and a Philips Achieva 3T scanner (MPRAGE) scanners. All subjects were young adults scanner (MPRAGE) scanners. All subjects were young adults with no history of neurological disorders. All scans from one with no history of neurological disorders. All scans from one session were averaged to obtain a low-noise structural image, session were averaged to obtain a low-noise structural image, and a single ‘best’ normalization transformation was estimated and a single ‘best’ normalization transformation was estimated from it. Therefore, normalized results were not directly affected from it. Therefore, normalized results were not directly affected by possible variability among normalization transforms by possible variability among normalization transforms estimated from individual scans. Identical T1 scans were estimated from individual scans. Identical T1 scans were repeated four to six times during the same session. In order to repeated four to six times during the same session. In order to compare and generalize our results, we also analyzed data sets compare and generalize our results, we also analyzed data sets available to the scientific community as part of the OASIS available to the scientific community as part of the OASIS project (Open Access Series of Imaging Studies, project (Open Access Series of Imaging Studies, http://www.oasis-brains.orghttp://www.oasis-brains.org//). The OASIS project includes a ). The OASIS project includes a data set containing 20 normal subjects imaged four times during data set containing 20 normal subjects imaged four times during a session and repeated on a subsequent visit within 90 days of a session and repeated on a subsequent visit within 90 days of their initial session. their initial session.
The data were processed using a unified The data were processed using a unified segmentation/normalization framework (Ashburner et. al 2005) segmentation/normalization framework (Ashburner et. al 2005) as implemented in the SPM5 software package (as implemented in the SPM5 software package (http://http://www.fil.ion.ucl.ac.uk/spmwww.fil.ion.ucl.ac.uk/spm//) and VBM5 toolkit () and VBM5 toolkit (http://dbm.neuro.uni-jena.de/vbm/vbm5-for-spm5/http://dbm.neuro.uni-jena.de/vbm/vbm5-for-spm5/). We used ). We used the default software settings and analysis parameters of SPM5, the default software settings and analysis parameters of SPM5, unless noted otherwise. Multiple scans obtained within the unless noted otherwise. Multiple scans obtained within the same session were coregistered and segmented independently same session were coregistered and segmented independently in each subject’s native space. Following conventional VBM in each subject’s native space. Following conventional VBM analysis (Ashburner and analysis (Ashburner and Friston,Friston, 2000, 2005), the segmented 2000, 2005), the segmented images were smoothed using 12 mm Gaussian kernel, yielding images were smoothed using 12 mm Gaussian kernel, yielding the measure of GM density. The GM density maps were the measure of GM density. The GM density maps were normalized and modulated using a single normalization normalized and modulated using a single normalization transform for multiple scans.. Variability of GM density was transform for multiple scans.. Variability of GM density was assessed by computing the standard deviation (SD) of GM assessed by computing the standard deviation (SD) of GM density across individual scansdensity across individual scans
Figure 3. Between Session Variability
Variability between the scan and rescan session (OASIS data set). The GM density variability maps were calculated using averages of four scans obtained during each session to reduce the effect of within session variability.
Intrinsic Variability of GM Density Maps: Intrinsic Variability of GM Density Maps: Its Implications to VBM StudiesIts Implications to VBM Studies
Andrew PoliakovAndrew Poliakov11, Judy McLaughlin, Judy McLaughlin22, Geoffrey Valentine, Geoffrey Valentine22, Ilona Pitkanen, Ilona Pitkanen22 and Lee Osterhout and Lee Osterhout22
11Department of Biological Structure and Department of Biological Structure and 22Department of Psychology, University of Washington, Seattle, WADepartment of Psychology, University of Washington, Seattle, WA
IntroductionIntroduction
Voxel-based morphometry (VBM) is an unbiased, objective Voxel-based morphometry (VBM) is an unbiased, objective neuroimaging technique for identifying structural changes in the neuroimaging technique for identifying structural changes in the brain. VBM involves a voxel-wise comparison of the local brain. VBM involves a voxel-wise comparison of the local concentration of gray matter (GM) in whole brain MRI scans. concentration of gray matter (GM) in whole brain MRI scans. Recent VBM studies have investigated changes in GM density Recent VBM studies have investigated changes in GM density due to learning and practice. Such changes are expected to be due to learning and practice. Such changes are expected to be small, thereby imposing stringent requirements on VBM small, thereby imposing stringent requirements on VBM sensitivity. However, VBM sensitivity is not uniform throughout sensitivity. However, VBM sensitivity is not uniform throughout the brain. Consequently, GM density changes might be more the brain. Consequently, GM density changes might be more observable in some regions of the brain than in others. Here, we observable in some regions of the brain than in others. Here, we explore three sources of variability in VBM sensitivity: explore three sources of variability in VBM sensitivity:
(1) variability introduced by choices in imaging protocols; (1) variability introduced by choices in imaging protocols; (2) within session variability; (3) between session variability.(2) within session variability; (3) between session variability.
Figure 1. Variability of GM Density: Single Subject Data
An example of single subject data acquired on Philips Achieva 3T scanner (scans were repeated four times during each session). The two columns correspond to two different versions of MPRAGE protocol (Left : TR=7.5 ms, TE = 3.5, ms Flip angle 80 ; Sagital, 0.86 x 0.86 x 1 mm3; Right: TR=7.4 ms TE = 3.44 ms Flip angle 80 ; Coronal, 0.92 x 0.92 x 1 mm3). A. The average of four T1 scans (normalized). B. GM segmentation of this image (normalized and modulated). C. The mean GM density map with variability map overlaid (red). Analysis shows noticeable difference in both pattern and magnitude of variability values.
DiscussionDiscussionVBM can be a powerful tool for studying subtle differences in VBM can be a powerful tool for studying subtle differences in GM density distributions that reflect anatomical correlates of GM density distributions that reflect anatomical correlates of cognitive parameters (Maguire et al., 2000, Gaser et al., 2003). cognitive parameters (Maguire et al., 2000, Gaser et al., 2003). The objective of such VBM studies is to detect statistically The objective of such VBM studies is to detect statistically significant changes in GM density. However, segmentation significant changes in GM density. However, segmentation accuracy is affected by a number of factors which can lead to accuracy is affected by a number of factors which can lead to variability in the estimation of GM density obtained under variability in the estimation of GM density obtained under identical conditions (i.e. same subject, scanner and protocol). identical conditions (i.e. same subject, scanner and protocol).
Some of the factors affecting segmentation accuracy result from Some of the factors affecting segmentation accuracy result from noise and imaging artifacts; these factors largely depend on the noise and imaging artifacts; these factors largely depend on the scanner and the data acquisition protocol. Our analysis scanner and the data acquisition protocol. Our analysis indicates that even slightly different data acquisition protocols indicates that even slightly different data acquisition protocols on the same scanner can produce noticeably different patterns on the same scanner can produce noticeably different patterns in the magnitude and distribution of GM variability maps (see in the magnitude and distribution of GM variability maps (see Fig.1). These differing distributions could potentially affect the Fig.1). These differing distributions could potentially affect the results and the conclusions of VBM studies, leading to results and the conclusions of VBM studies, leading to significantly different findings for the same experimental significantly different findings for the same experimental paradigm.paradigm.
On the other hand, some cortical areas may consistently show On the other hand, some cortical areas may consistently show increased variability in GM density due to relatively complex increased variability in GM density due to relatively complex anatomy and/or the persistence of imaging artifacts. For anatomy and/or the persistence of imaging artifacts. For example, the tip of the temporal lobe consistently showed example, the tip of the temporal lobe consistently showed higher variability in all data sets we analyzed (see Figs. 2 and higher variability in all data sets we analyzed (see Figs. 2 and 3). This is not unexpected considering the proximity of non-3). This is not unexpected considering the proximity of non-brain tissues of similar intensities, as well as the possibility of brain tissues of similar intensities, as well as the possibility of blood flow, eye movement and susceptibility artifacts. We blood flow, eye movement and susceptibility artifacts. We noticed that very few VBM studies reported GM density noticed that very few VBM studies reported GM density changes it this area. In contrast, the parietal and occipital changes it this area. In contrast, the parietal and occipital cortex are not subject to these or other common imaging cortex are not subject to these or other common imaging artifacts. As a result, these areas show little variability in GM artifacts. As a result, these areas show little variability in GM density. Interestingly, the parietal and occipital cortices are the density. Interestingly, the parietal and occipital cortices are the areas that have been implicated in a number of VBM studies. areas that have been implicated in a number of VBM studies.
We suggest that intrinsic variability of GM density should be We suggest that intrinsic variability of GM density should be taken into consideration when interpreting the results of VBM taken into consideration when interpreting the results of VBM studies. Variability analysis may be a useful tool for designing studies. Variability analysis may be a useful tool for designing and planning a VBM study. It may help identify problematic and planning a VBM study. It may help identify problematic areas, detect subtle imaging artifacts and help refine data areas, detect subtle imaging artifacts and help refine data acquisition and analysis. acquisition and analysis.
ReferencesReferences• Ashburner J. and Friston K.J. Unified segmentation. NeuroImage, Ashburner J. and Friston K.J. Unified segmentation. NeuroImage, 26:839-851, 2005. 26:839-851, 2005.
• Ashburner J. and Friston. K.J. Voxel-Based Morphometry -- The Ashburner J. and Friston. K.J. Voxel-Based Morphometry -- The Methods. NeuroImage, 11:805-821, 2000.Methods. NeuroImage, 11:805-821, 2000.
• Maguire EA, Gadian DG, Johnsrude IS, et al. Navigation-related Maguire EA, Gadian DG, Johnsrude IS, et al. Navigation-related structural change in the hippocampi of taxi drivers. Proc Natl Acad Sci structural change in the hippocampi of taxi drivers. Proc Natl Acad Sci USA. 2000; 97: 4398-4403.USA. 2000; 97: 4398-4403.
• Gaser C, Schlaug G. Brain structures differ between musicians and Gaser C, Schlaug G. Brain structures differ between musicians and non-musicians. J Neurosci 2003; 23: 9240-9245.non-musicians. J Neurosci 2003; 23: 9240-9245.
• Draganski B, Gaser C, Busch V, Schuierer G, Bogdahn U, May A. Draganski B, Gaser C, Busch V, Schuierer G, Bogdahn U, May A. Neuroplasticity: changes in grey matter induced by training. Nature 2004; Neuroplasticity: changes in grey matter induced by training. Nature 2004; 427: 311-2.427: 311-2.
• Mechelli A, Crinion JT, Noppeney U, et al. Neurolinguistics: structural Mechelli A, Crinion JT, Noppeney U, et al. Neurolinguistics: structural plasticity in the bilingual brain. Nature 2004; 431: 757.plasticity in the bilingual brain. Nature 2004; 431: 757.
Acknowledgements
Supported by NIH grants R01DC01947 and P30DC04661
Figure 2. Within Session Variability of GM Density: Averaged Across Subject Groups
The variability maps of GM Density averaged across subject groups are color coded and superimposed onto GM probability density maps. Both cases show that that variability was strongly non-homogenious throughout the brain. The patterns appear to be quite different reflecting the differences in the data acquisition protocol, artifact correction techniques etc.
A. Data set obtained using GE Signa 1.5T (four subjects, 2 sessions of 6 scans, SPGR).
B. OASIS data set (15 subjects, two session of four scans, MPRAGE).
Note that GE Signa was an older scanner, and modern systems and data acquisition protocols generally produce better image quality .
A
B
C
A
B
