Default network correlations analyzed on native surfaces
Highlights
► Automated method using FreeSurfer for analyzing BOLD correlations on native surfaces. ► Surface parcellation allows comparisons of native-surface analyses across subjects. ► Method highlights default network regions in data from individual subjects. ► Results presented for young, elderly, Alzheimer's, Parkinson-related dementias.
Introduction
Characteristic patterns of low-frequency correlations have been repeatedly identified in the blood oxygenation level dependent (BOLD) fMRI signal when subjects are asked to simply lie still in the scanner (Biswal et al., 1995, Greicius et al., 2003, Buckner et al., 2008). The relative consistency of these patterns across studies and analysis methods, as well as the simplicity of the instructions, has led to considerable interest in their potential application as a biomarker in disease (Fox and Raichle, 2007, Rogers et al., 2007, Greicius, 2008, Auer, 2008, van den Heuvel and Hulshoff Pol, 2010). Particular attention has been paid to a collection of regions called the default network1 and the disruption of correlations across these regions in Alzheimer's disease (Greicius et al., 2004, Wang et al., 2006, Allen et al., 2007, Supekar et al., 2008, Koch et al., 2010, Zhang et al., 2010; for review, see Greicius, 2008, Sorg et al., 2009). Disruptions in so-called functional connectivity in the default network have also been reported in conditions believed to precede onset of Alzheimer's disease, including patients with amnestic mild cognitive impairment (Sorg et al., 2007, Pihlajamäki et al., 2009, Bai et al., 2009) and cognitively unimpaired subjects with high amyloid burden (Hedden et al., 2009, Sheline et al., 2010).
Anatomical variability across subjects gives rise to two notable challenges in the analysis of spontaneous BOLD correlations within the default network. First, if analyses are to be extended beyond qualitative assessment in individual subjects, a method of comparing results across subjects is critical. Second, the network of interest has to be identified in each subject. In seed-correlation analyses, this is typically accomplished by choosing an a priori seed region known to lie within the network (e.g., Hedden et al., 2009, Sheline et al., 2010). For independent component analyses (ICA), a template is used to identify the component that best matches the default network (e.g., Greicius et al., 2004, Seeley et al., 2009).
Currently, both of these challenges are addressed by performing analyses in atlas-volume space. Anatomical and functional images from each subject are transformed, or warped, to match a canonical brain (e.g., Talairach-Tournoux or Montreal Neurological Institute template). Once in a standardized, or atlas, volume, seed regions and templates from the literature or other data sets can be applied to the spatially transformed data to identify the default network. The process of transforming data to an atlas volume also permits direct comparison of analysis results across subjects and studies.
Unfortunately, atlas-space results are only valid to the extent that the warping process is valid, a point of particular concern in conditions where participants’ brains differ considerably from the atlas due to disease. Functional correlation analyses are subject to concerns similar to some known issues with voxel-based morphometry, a method for structural MRI analysis which also depends heavily on accurate registration to a template. Improper registration can lead to misleading results in both cases because, for example, a given coordinate represents gray matter in the template but lies in cerebrospinal fluid in a patient's warped brain. Voxel-based morphometry gives varied results depending on the particular warping algorithm used (Senjem et al., 2005), and even algorithms identified as “optimized,” which include multiple steps to improve normalisation, are still prone to errors when atrophy causes gross changes in brain structure (Bookstein, 2001, Ashburner and Friston, 2001, Senjem et al., 2005). Despite the crucial role warping plays in functional correlation analyses and the known pitfalls of common methods in the face of structural brain pathology, accuracy of individual transformations are rarely, if ever, reported or displayed.
Analysis on a subject's native surface offers potential advantages over atlas-volume methods. First, possible ambiguity about precise anatomic locations is reduced. Measuring functional correlations on native surfaces also facilitates accounting for anatomic effects of disease and age. Moreover, by preserving inter-individual anatomic variability, longitudinal patient studies can better avoid confounds due to disease-related structural changes that affect an individual patient's brain over time. Comparison of functional measures to other individual markers is also straightforward on native surfaces, in particular cross-modal imaging markers such as amyloid imaging results and regional cortical thickness. There is also a clinical appeal to obtaining and displaying functional imaging results on the brain surface of an individual patient. Prior studies have pointed to the potential of functional correlations to provide meaningful results in individual patients (Greicius et al., 2004, Buckner and Vincent, 2007, Koch et al., 2010); analyzing functional data on native surfaces may be an important step toward that aim.
We assessed the utility of the FreeSurfer (http://surfer.nmr.mgh.harvard.edu) cortical parcellation to analyze functional correlations on the native surfaces of individual subjects. Automated processes are employed to anatomically parcellate each subject's cortical surface into distinct regions of cortex (subcortical gray matter structures are included after a similar automated volume segmentation). One cortical region, the isthmus cingulate, is proposed as a suitable native-surface seed for identification and analysis of the default network. Parcellation and segmentation regions are then used for group-level analyses by comparing interregional correlations between equivalent regions in different subjects. Additionally, registration of native sulcal and gyral patterns to an average surface allows display of group-level results after quantitative parcellation analysis on native surfaces.
Here we present results from the application of this method to BOLD data from young, healthy subjects as a proof of concept. The primary findings were reproduced in preliminary data from multiple disease populations, including Alzheimer's disease (AD), Parkinson's disease dementia (PDD), Dementia with Lewy bodies (DLB), and cognitively unimpaired elderly controls.
Section snippets
Subjects
Subject demographics are provided in Table 1. Patients designated ‘Alzheimer's disease’ had a clinical diagnosis of probable AD based on the NINCDS/ADRDA criteria (McKhann et al., 1984); diagnoses for dementia with Lewy bodies and Parkinson's disease dementia were based on the criteria established by the Movement Disorders Society Task Force (Geser et al., 2005, McKeith, 2007). Diagnosis for all patients was made by consensus of two or more neurologists in the UCSD Shiley-Marcos Alzheimer's
Results
Both linear and nonlinear algorithms successfully aligned the atrophied brain to the MNI template (Fig. 1). The nonlinear methods (FSL, SPM2, SPM8) appear to have reduced ventricular spaces and stretched the brain tissue to fill portions of the adjacent CSF space (distortions to the skull and other tissues in the nonlinear examples should be ignored, as the methods are optimized for registration of the brain, not extraparenchymal tissues). None of the transformations, however, fully accounted
Discussion
Spontaneous BOLD correlation studies may afford opportunities to increase our understanding of how regions of the brain interact and to develop clinical tools for diagnosis or measurement of disease progression. Already, intriguing results have been reported in various diseases, including mild cognitive impairment and Alzheimer's disease. Analysis in native space may improve accuracy, allow more rigorous investigation into resting-state correlation phenomena, and otherwise facilitate transition
Acknowledgements
We are grateful for the assistance of Donald J. Hagler, Jr. and Anders Dale for helpful suggestions for data analysis. We would also like to thank Elizabeth A. Murphy and Erik J. Kaestner for their help in recruiting and scanning the elderly subjects and patients, as well as the staff at the Center for Functional MRI and the Radiology Imaging Laboratory at UCSD for technical support during data acquisition. Funding for this work was generously provided by NIA 2P50AG005131, NINDS K02 NS067427,
References (60)
- et al.
Why voxel-based morphometry should be used
NeuroImage
(2001) Spontaneous low-frequency blood oxygenation level-dependent fluctuations and functional connectivity analysis of the ‘resting’ brain
Magn Reson Imaging
(2008)- et al.
Abnormal resting-state functional connectivity of posterior cingulate cortex in amnestic type mild cognitive impairment
Brain Res
(2009) “Voxel-based morphometry” should not be used with imperfectly registered images
NeuroImage
(2001)- et al.
Unrest at rest: default activity and spontaneous network correlations
NeuroImage
(2007) - et al.
Time–frequency dynamics of resting-state brain connectivity measured with fMRI
Neuroimage
(2010) - et al.
Cortical surface-based analysis. I. Segmentation and surface reconstruction
Neuroimage
(1999) - et al.
An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest
Neuroimage
(2006) - et al.
Automatic parcellation of human cortical gyri and sulci using standard anatomical nomenclature
Neuroimage
(2010) - et al.
Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain
Neuron
(2002)