CONCLUSIONS  We developed a new procedure for awFC that combines information derived from fMRI and DTI data to evaluate FC, accounting for the probabilities of structural connections between regions.  We derived a novel distance metric, which can be incorporated into well-established hierarchical (or partitioning) clustering techniques.  We proposed an objective function for our awFC procedure that can be used to simultaneously optimize the number of clusters and a parameter governing the strength of anatomical weighting.  Our awFC method yields both individual-level and group-level clustering results.  Our awFC technique can be executed very quickly (less than 1 minute) in MATLAB. Method: Optimization  We regard a good clustering solution as one whose intracluster regions (a) exhibit a high degree of correlation between the associated fMRI time series and (b) reveal a relatively high degree of underlying SC. Thus, we develop an objective function to optimize our clustering solution with respect to G and . o Define FCwi as the average of the correlations (each maximized over lag-u, e.g. u in [-3,3]) for all within cluster region pairs and FCtot as the average over all region pairs. Similarly, SCwi and SCtot are the averaged probabilities of SC for within cluster region pairs and throughout the entire brain, respectively. We optimize our clustering solution according to the following objective function: RESULTS Group-level awFC Resting-State Networks (λ=1): Results Summary:  The awFC analysis isolates a parsimonious network of clusters, which contain several previously reported regions in the default mode (DFM) network [3,4].  Several correlations between intracluster region pairs were statistically significantly greater than 0.3 (α=0.01 familywise error rate) and almost all were significantly greater than 0 (not shown above).  awFC exhibited a more parsimonious system of DFM network regions than the corresponding clustering solution with no anatomical weighting.  The featured clusters include regions found to be involved in autobiographical recall, visualization (including mental imagery), awareness, and response selection. EXPERIMENTAL DATA  Both resting-state fMRI and DTI data were acquired from 6 healthy female subjects on a 3T Siemens scanner. A series of 210 scans were acquired with TR=2 sec, 20 slices, and 3.4 x 3.4 x 4 mm 3 voxel resolution. The functional runs were collected with a Z-saga sequence to avoid orbitofrontal signal ablation. Diffusion weighting was isotropically distributed along 13 directions (12 directions + baseline). Both fMRI and DTI data were preprocessed in FSL (FMRIB's Software Library). Probabilistic diffusion tensor tractography (DTT) was performed in FDT (FMRIB's Diffusion Toolbox). [1]  We summarized both the fMRI and DTI data over 86 regions for each subject, with regions defined according to the Automated Anatomical Labeling (AAL) system [2]. fMRI Regional Summaries: We first transformed the data to the frequency domain using the fast Fourier transform, averaged the power spectral density estimates across subjects, and applied a singular value decomposition (SVD) for each region separately to obtain the dominant frequency pattern in the region. Next, we identified the voxel that exhibited frequency characteristics most closely resembling the regional summary and selected roughly 150 neighboring voxels based on a 6 mm sphere. To summarize over these voxels, producing a single time series for each region and for each subject, we applied a second SVD to the time-domain data within each regional sphere. We evaluated our final selections to ensure that our objectives (homogeneity and representative of the region) were met. The final summaries for each subject can be viewed as a 210x86 matrix. DTI-Based Regional Summaries: We applied probabilistic DTT for each AAL region, yielding voxel-level counts indicating the likelihood of having fiber tracts extending from the AAL region to each voxel outside the region. Superimposing the AAL map for regions outside the one under consideration, we obtained region-to-region summaries by summing over all of the voxels in the target regions and then appropriately scaling to form probabilities of connections. We imposed symmetry by calculating the maximum of the two directional probabilities for each region. Objective: To evaluate resting-state FC in a group of healthy subjects, incorporating both functional characteristics from fMRI data and structural information derived from DTI. METHODS  We propose an anatomically-weighted FC (awFC) method, which combines fMRI and DTI data, to explore resting-state FC networks.  We use cluster analysis to implement our new approach, with a novel distance metric that contains contributions from the fMRI and DTI data.  Our awFC method operates on the basis of the following principles: Collections of regions are deemed to exhibit FC if they have similar temporal brain activity characteristics. ｾ Our method regards FC as more likely if there is evidence of associated SC. ｾ In the absence (or with weak evidence) of SC, we regard FC as possible, but our method requires stronger evidence by the functional data in this case. Functional Distance: The lag-u functional distance between regions i and j is: where is the fMRI measure of brain activity in region i at time t. Structural Distance: is the DTI-based probability of SC between regions. Once the are obtained, we allow for indirect SC between regions by defining Anatomically-Weighted Functional Distance: Clustering:  Use average linkage (AL) to cluster data (group- or individual-level). ｾ AL begins with each region representing its own cluster. ｾ At each iteration, two clusters merge on the basis of d ij (small distances (on average) are preferred). ｾ The iterative process ceases when only a single cluster remains. ｾ We optimize the number of clusters G and the parameter . , )] ( ) ( [ ) ( 1 2 ∑ − = − + = u T t j i ij t y u t y u f ) (t y i Evaluating Functional Connectivity using fMRI Data with Diffusion-Based Anatomical Weighting F. DuBois Bowman, Gordana Derado, and Shuo Chen Department of Biostatistics and Bioinformatics, Emory University, Atlanta, GA 30322, U.S.A. Evaluating Functional Connectivity using fMRI Data with Diffusion-Based Anatomical Weighting F. DuBois Bowman, Gordana Derado, and Shuo Chen Department of Biostatistics and Bioinformatics, Emory University, Atlanta, GA 30322, U.S.A. INTRODUCTION Neural processing relies on functional connectivity (FC) or associations between brain activity measures from different regions. Given the intricate system of inter-communicating neuronal networks via white matter fiber tracts, it is likely that functional connections between two brain regions are mediated by anatomical or structural connectivity (SC). Diffusion tensor imaging (DTI) enables the reconstruction and probabilistic quantification of major fiber tracts in the brain, and this SC information provides an opportunity to improve our understanding of FC. We propose a new statistical framework that combines fMRI and DTI data to help describe the functional organization within the human brain. ACKNOWLEDGEMENTS  Many thanks to Dr. Ying Guo of CBIS at Emory, to Dr. Clint Kilts, Dr. Helen Mayberg, and Mr. Tim Ely from the Dept. of Psychiatry and Behavioral Sciences at Emory, and to Dr. Andrew James from the BITC at Emory for providing us with the data employed in our analysis and for valuable discussions about the data.  Supported by R01-MH079251 and NIH predoctoral training grant T32 GM074909-01. Reprints of this poster presentation are available online at http://www.sph.emory.edu/bios/CBIS/presentations.html References 1. T.E.J. Behrens, M.W. Woolrich, M. Jenkinson, H. Johansen-Berg, R.G. Nunes, S. Clare, P.M Matthews, J.M. Brady and S.M. Smith (2003). Characterization and Propagation of Uncertainty in Diffusion-Weighted MR Imaging, Mag Res in Med 50:1077–1088. 2. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, Mazoyer B, Joliot M.(2002). Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. NeuroImage 15, 273- 289. 3. Buckner, R.L., Andrews-Hanna, J.R., Schacter, D.L., The Brain's Default Network: Anatomy, Function, and Relevance to Disease, Ann NY Acad Sci, Vol. 1124, No. 1. (1 March 2008), pp. 1-38. 4. Michael D. Fox, Abraham Z. Snyder, Justin L. Vincent, Maurizio Corbetta, David C. Van Essen, and Marcus E. Raichle (2005). The human brain is intrinsically organized into dynamic, anticorrelated functional networks. PNAS, vol. 102, no. 27, pp. 9673–9678. ij π 1 0 < π ≤ ij [ ) ∞ ∈ λ , 1 ( ) [ ], ) ( min exp 1 1 u f d ij U u ij ij ∈ − ⎥ ⎦ ⎤ ⎢ ⎣ ⎡ λ π − = λ ij π . )] ( max , max[ kj ik k ij ij π π π = π λ ∑ = ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ + ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ = λ n k k k k k SCtot SCwi FCtot FCwi n G h 1 log log 1 ) , ( fMRI: DTI: 1 2 84 85 210 86 1 Scan Region Data Summaries Distances Anatomically-weighted Functional Distance ) , ( G h λ Strength of inter-regional correlations are statistically significant at family-wise error rate of α=0.01. 30 . 0 : vs. 30 . 0 : 1 0 > ρ = ρ ij ij H H No anatomical weighting (λ=∞): 0 20 40 60 0 10 20 30 40 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 Num. Clusters (G) λ log h(λ, G) (lag G) 0 10 20 30 40 50 60 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 G (lag 1) Σ log h(λ,G) G=13 0 5 10 15 20 25 30 35 40 1.02 1.04 1.06 1.08 1.1 1.12 1.14 1.16 1.18 λ Σ log h(λ,G) λ=1 Optimization: