Surface-based TBM Boosts Power to Detect Disease Effects on the ...
1
Surface-based TBM Boosts Power to Detect Disease Effects on the Brain: An N=804 ADNI Study Yalin Wang 1 , Yang Song 2 , Priya Rajagopalan 3 , Tuo An 3 , Krystal Liu 3 , Yi‐Yu Chou 3 , Boris Gutman 3 , Arthur W. Toga 3 , Paul M. Thompson 3 1 Computer Science and Engineering, Arizona State University, Tempe, AZ, USA 2 Department of Mathematics, University of Washington, Seattle , WA, USA. 3 Lab. of Neuro Imaging, Brain Mapping Division, Dept. of Neurology, UCLA School of Medicine, Los Angeles, CA, USA [email protected] We combined the power of parametric surface modeling with tensor‐based morphometry to study hippocampal differences associated with Alzheimer’s disease (AD) and mild cognitive impairment (MCI) in 490 subjects (97 AD, 245 MCI, 148 controls) and ventricular differences in 804 subjects scanned as part of the Alzheimer’s Disease Neuroimaging Initiative (ADNI; 184 AD, 391 MCI, 229 controls). We aimed to show that a new analysis of anatomy, based on surface multivariate tensor‐based morphometry (mTBM) and radial distance mapping provides a more powerful way to detect localized anatomical differences than conventional surface‐based analysis methods (Figure 1). Results Group differences: Figure 3 shows regions with significant ventricular shape differences in AD versus MCI (N=575) as an example application. We show the statistical p‐map based on four different local statistics: mTBM + radial distance, mTBM, radial distance and Jacobian determinant. All 4 statistics found consistent morphometric differences, but the new multivariate statistics gave greatest effect sizes (see cumulative distribution plots of p‐values). Figure 3. Statistical p‐map results (left panel) show ventricular surface shape differences b f MCI d AD Introduction between groups of MCI and AD subjects (N=575), using four different surface statistics. Non‐ blue colors show vertices with statistical differences, at the p=0.05 level, uncorrected. The new statistics (mTBM + radial distance)detected anatomical differences more powerfully than the other three statistics. The right panel shows that the new statistic yields results that are highly significant using FDR (as they rise above the reference line y=20x). Figure 4. Statistical p‐map results (left panel) show correlation between Tau/Aβ 1‐42 and various surface statistics in ADNI (Hippocampus: N=271 subjects) Non‐blue colors show vertices Correlation with CSF biomarkers: For the hippocampus, our new combined statistic detected strongest correlations for all CSF biomarkers, outperformed the other statistics. Figure 4 compares different statistics for detecting correlations with Tau/Aβ 1‐42 (N=271). Again, multivariate statistics gave greatest effect sizes. Methods In our recent studies (Wang et al. 2010, 2011), surface mTBM was shown to be more sensitive to detecting group differences than other standard surface‐based statistics, such as the radial distance of the surface to a medial curve. Intuitively, radial distance and mTBM are complementary: radial distance describes anatomical deformations largely parallel to the surface normal, but mTBM measures more tangential expansions or contractions within surfaces Here we propose a combination of radial distance and mTBM (Figure 2) offering Figure 1. A flow chart of the steps in our multivariate statistical analysis of brain morphometry. In our system, the subcortical structures are segmented as in prior work (Morra et al., 2009b; Chou et al., 2010). We then reconstruct the surfaces and compute conformal coordinates for registration. The computed statistics are used for group difference analysis and analysis of correlation with CSF biomarkers. Non blue colors show vertices that correlate to Tau/Aβ 1‐42 , at the p=0.05 level, uncorrected. The right panel shows that the new statistics detected gave greatest effect size using FDR (as they rise above the reference line y=20x). surfaces. Here we propose a combination of radial distance and mTBM (Figure. 2), offering a powerful set of surface statistics for subcortical surface morphometry. We hypothesized that this combination might boost statistical power to detect disease effects on subcortical structures. Estimated sample sizes necessary to detect disease effects: We randomly selected N subjects from the original dataset. Subjects from each diagnostic group were chosen to preserve the relative proportions of control, MCI and AD subjects. We repeated this sampling process 100 ti Fig re 5 h CDF f th l f dt ti diff bt MCI dC t l Experiments We analyzed baseline T1‐weighted MRI scans from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset (http://www.loni.ucla.edu/ADNI). The segmented data were from our prior work (Morra et al., 2009; Chou et al., 2010). Using a method from exterior calculus based on holomorphic differentials, we induced conformal grids on these meshes (Wang et al., 2011). We left two holes at the front and back of the hippocampus, representing its t i j ti ith th dl d it t i li it it t it th hit tt times. Figure 5 shows CDFsof the p‐values for detecting differences between MCI andControl groups, for both hippocampus and lateral ventricle. When detecting group differences, the minimal effective sample size for the hippocampus was lower than those we reported previously (Morra et al. 2009), where radial distance was used as the statistic. Figure 5. Effects of varying the sampel size: CDFs of p‐values measuring the effect sizes for diagnostic group differences (among MCI and CTL) for both hippocampal atrophy and ventricular enlargement. The numbers are the critical p‐values Figure 2. The proposed multivariate surface statist consists of two popular surface statistics: radial distance (Thompson et al. 2004) and multivariate tensor‐based morphometry (mTBM) (Wang et al. 2010). anterior junction with the amygdala and its posterior limit as it turns into thewhite matter of the fornix. The hippocampus then represented as an open boundary genus one surface, i.e., a cylinder, and was conformally mapped to a rectangle. To model the lateral ventricular surface, we automatically located and introduced three cuts, based on the topology of the lateral ventricles, in which several horns are joined together at the ventricular “atrium” or “trigone”. With holomorphic flow segmentation (Wang et al., 2011), each lateral ventricular surface was automatically partitioned into 3 pieces . Although surface geometry is widely variable across subjects, these zero point locations are stable and intrinsically determined by the surface conformal structures, and the resulting surface partition into component meshes is stable. We registered the hippocampal and lateral ventricular surfaces across subjects using a constrained harmonic map. Conclusion: Our method consists of automated surface extraction, conformal parameterization and partitioning, harmonic mapping, and tensor statistics. In this large subcortical morphometry for each experiment. The minimal effective sample sizes for the hippocampus are comparable to, or lower than, those reported in our prior work (Morra et al., 2009) in the same dataset when radial distance was used as the statistic. Acknowledgement: This work was funded by the National Institutes of Health through the NIH Roadmap for Medical Research, Grant U54 RR021813. Additional support was provided by the National Institute on Aging (AG016570 to PMT), the National Library of Medicine, the National Institute for Biomedical Imaging and Bioengineering, and the National Center for Research Resources (LM05639, EB01651, RR019771 to PMT). As well as examining shape differences using the eigenvalues of the surface deformation tensor, we also used a log‐Euclidean metrics, which allows multivariate statistics on the tensors to be computed easily using standard Euclidean formulae. Together with radial distance, on each surface point, we compute a 4x1 feature vector. We applied Hotelling’s T 2 test to study group difference and multiple linear regression to study statistical shape correlation with CSF biomarkers. study, greatest effect sizes were obtained when using our novel surface‐based analysis method. The complete software package is available at http://gsl.lab.asu.edu/conformal.htm. Reference: Chou, Y., et al., Ventricular maps in 804 ADNI subjects: correlations with CSF biomarkers and clinical decline. Neurobiology of Aging, 2010. 31(8): p. 1386-1400. Morra, J.H., et al., Automated mapping of hippocampal atrophy in 1-year repeat MRI data from 490 subjects with Alzheimer's disease, mild cognitive impairment, and elderly controls. NeuroImage, 2009. 45(1, Supplement 1): p. S3 - S15. Thompson, P. M., et al. 2004. Mapping hippocampal and ventricular change in Alzheimer's disease. NeuroImage, 22(4), 1754 - 1766. Wang Y., et al. 2010. Multivariate tensor-based morphometry on surfaces: Application to mapping ventricular abnormalities in HIV/AIDS. NeuroImage, 49(3), February 2010, p. 2141-2157. Wang, Y., et al. 2011. Surface-based TBM Boosts Power to Detect Disease Effects on the Brain: An N=804 ADNI Study, NeuroImage, 56(4), June 2011, pp. 1993-2010
Transcript of Surface-based TBM Boosts Power to Detect Disease Effects on the ...
Surface-based TBM Boosts Power to Detect Disease Effects on the Brain: An N=804 ADNI Study
Yalin Wang1, Yang Song2, Priya Rajagopalan3, Tuo An3, Krystal Liu3, Yi‐Yu Chou3, Boris Gutman3, Arthur W. Toga3, Paul M. Thompson3
1Computer Science and Engineering, Arizona State University, Tempe, AZ, USA2Department of Mathematics, University of Washington, Seattle , WA, USA.
3Lab. of Neuro Imaging, Brain Mapping Division, Dept. of Neurology, UCLA School of Medicine, Los Angeles, CA, [email protected]
We combined the power of parametric surface modeling with tensor‐based morphometryto study hippocampal differences associated with Alzheimer’s disease (AD) and mildcognitive impairment (MCI) in 490 subjects (97 AD, 245 MCI, 148 controls) and ventriculardifferences in 804 subjects scanned as part of the Alzheimer’s Disease NeuroimagingInitiative (ADNI; 184 AD, 391 MCI, 229 controls). We aimed to show that a new analysis ofanatomy, based on surface multivariate tensor‐based morphometry (mTBM) and radialdistance mapping provides a more powerful way to detect localized anatomical differencesthan conventional surface‐based analysis methods (Figure 1).
ResultsGroup differences: Figure 3 shows regions with significant ventricular shape differences in AD versus MCI (N=575) as an example application. We show the statistical p‐map based on four different local statistics: mTBM + radial distance, mTBM, radial distance and Jacobiandeterminant. All 4 statistics found consistent morphometric differences, but the new multivariate statistics gave greatest effect sizes (see cumulative distribution plots of p‐values).
Figure 3. Statistical p‐map results (left panel) show ventricular surface shape differences b f MCI d AD
Introduction
y ( g )between groups of MCI and AD subjects (N=575), using four different surface statistics. Non‐blue colors show vertices with statistical differences, at the p=0.05 level, uncorrected. The new statistics (mTBM + radial distance)detected anatomical differences more powerfully than the other three statistics. The right panel shows that the new statistic yields results that are highly significant using FDR (as they rise above the reference line y=20x).
Figure 4. Statistical p‐map results (left panel) show correlation between Tau/Aβ1‐42 and various surface statistics in ADNI (Hippocampus: N=271 subjects) Non‐blue colors show vertices
Correlation with CSF biomarkers: For the hippocampus, our new combined statistic detected strongest correlations for all CSF biomarkers, outperformed the other statistics. Figure 4compares different statistics for detecting correlations with Tau/Aβ1‐42 (N=271). Again, multivariate statistics gave greatest effect sizes.
MethodsIn our recent studies (Wang et al. 2010, 2011), surface mTBM was shown to be more sensitive to detecting group differences than other standard surface‐based statistics, such as the radial distance of the surface to a medial curve. Intuitively, radial distance and mTBMare complementary: radial distance describes anatomical deformations largely parallel to the surface normal, but mTBMmeasures more tangential expansions or contractions within surfaces Here we propose a combination of radial distance and mTBM (Figure 2) offering
Figure 1. A flow chart of the steps in our multivariate statistical analysis of brain morphometry. In our system, the subcortical structures are segmented as in prior work (Morra et al., 2009b; Chou et al., 2010). We then reconstruct the surfaces and compute conformal coordinates for registration. The computed statistics are used for group difference analysis and analysis of correlation with CSF biomarkers.
Non blue colors show vertices that correlate to Tau/Aβ1‐42, at the p=0.05 level, uncorrected. The right panel shows that the new statistics detected gave greatest effect size using FDR (as they rise above the reference line y=20x).
surfaces. Here we propose a combination of radial distance and mTBM (Figure. 2), offering a powerful set of surface statistics for subcortical surface morphometry. We hypothesized that this combination might boost statistical power to detect disease effects on subcorticalstructures.
Estimated sample sizes necessary to detect disease effects: We randomly selected N subjects from the original dataset. Subjects from each diagnostic group were chosen to preserve the relative proportions of control, MCI and AD subjects. We repeated this sampling process 100 ti Fig re 5 h CDF f th l f d t ti diff b t MCI d C t l
ExperimentsWe analyzed baseline T1‐weighted MRI scans from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset (http://www.loni.ucla.edu/ADNI). The segmented data were from our prior work (Morra et al., 2009; Chou et al., 2010). Using a method from exterior calculus based on holomorphic differentials, we induced conformal grids on these meshes (Wang et al., 2011). We left two holes at the front and back of the hippocampus, representing its t i j ti ith th d l d it t i li it it t i t th hit tt
times. Figure 5 shows CDFs of the p‐values for detecting differences between MCI and Control groups, for both hippocampus and lateral ventricle. When detecting group differences, the minimal effective sample size for the hippocampus was lower than those we reported previously (Morra et al. 2009), where radial distance was used as the statistic.
Figure 5. Effects of varying the sampel size: CDFs of p‐values measuring the effect sizes for diagnostic group differences (among MCI and CTL) for both hippocampal atrophy and ventricular enlargement. The numbers are the critical p‐values
Figure 2. The proposed multivariate surface statist consists of two popular surface statistics: radial distance (Thompson et al. 2004) and multivariate tensor‐based morphometry (mTBM) (Wang et al. 2010).
anterior junction with the amygdala and its posterior limit as it turns into the white matter of the fornix. The hippocampus then represented as an open boundary genus one surface, i.e., a cylinder, and was conformally mapped to a rectangle. To model the lateral ventricular surface, we automatically located and introduced three cuts, based on the topology of the lateral ventricles, in which several horns are joined together at the ventricular “atrium” or “trigone”. With holomorphic flow segmentation (Wang et al., 2011), each lateral ventricular surface was automatically partitioned into 3 pieces . Although surface geometry is widely variable across subjects, these zero point locations are stable and intrinsically determined by the surface conformal structures, and the resulting surface partition into component meshes is stable. We registered the hippocampal and lateral ventricular surfaces across subjects using a constrained harmonic map.
Conclusion:Our method consists of automated surface extraction, conformal parameterization and partitioning, harmonic mapping, and tensor statistics. In this large subcortical morphometry
pfor each experiment. The minimal effective sample sizes for the hippocampus are comparable to, or lower than, those reported in our prior work (Morra et al., 2009) in the same dataset when radial distance was used as the statistic.
Acknowledgement: This work was funded by the National Institutes of Health through the NIH Roadmap for Medical Research, Grant U54 RR021813. Additional support was provided by the National Institute on Aging (AG016570 to PMT), the National Library of Medicine, the National Institute for Biomedical Imaging and Bioengineering, and the National Center for Research Resources (LM05639, EB01651, RR019771 to PMT).
As well as examining shape differences using the eigenvalues of the surface deformation tensor, we also used a log‐Euclidean metrics, which allows multivariate statistics on the tensors to be computed easily using standard Euclidean formulae. Together with radial distance, on each surface point, we compute a 4x1 feature vector.
We applied Hotelling’s T2 test to study group difference and multiple linear regression to study statistical shape correlation with CSF biomarkers.
study, greatest effect sizes were obtained when using our novel surface‐based analysis method. The complete software package is available at http://gsl.lab.asu.edu/conformal.htm.
Reference:Chou, Y., et al., Ventricular maps in 804 ADNI subjects: correlations with CSF biomarkers and clinical decline. Neurobiology of Aging, 2010. 31(8): p. 1386-1400.Morra, J.H., et al., Automated mapping of hippocampal atrophy in 1-year repeat MRI data from 490 subjects with Alzheimer's disease, mild cognitive impairment, and elderly controls. NeuroImage, 2009. 45(1, Supplement 1): p. S3 - S15.Thompson, P. M., et al. 2004. Mapping hippocampal and ventricular change in Alzheimer's disease. NeuroImage, 22(4), 1754 - 1766.Wang Y., et al. 2010. Multivariate tensor-based morphometry on surfaces: Application to mapping ventricular abnormalities in HIV/AIDS.NeuroImage, 49(3), February 2010, p. 2141-2157.Wang, Y., et al. 2011. Surface-based TBM Boosts Power to Detect Disease Effects on the Brain: An N=804 ADNI Study, NeuroImage, 56(4), June 2011, pp. 1993-2010
