Fusing functional signals by sparse canonical correlation analysis improves network reproducibility

We contribute a novel multivariate strategy for computing the structure of functional networks in the brain from arterial spin labeling (ASL) MRI. Our method fuses and correlates multiple functional signals by employing an interpretable dimensionality reduction method, sparse canonical correlation analysis (SCCA). There are two key aspects of this contribution. First, we show how SCCA may be used to compute a multivariate correlation between different regions of interest (ROI). In contrast to averaging the signal over the ROI, this approach exploits the full information within the ROI. Second, we show how SCCA may simultaneously exploit both the ASL-BOLD and ASL-based cerebral blood flow (CBF) time series to produce network measurements. Our approach to fusing multiple time signals in network studies improves reproducibility over standard approaches while retaining the interpretability afforded by the classic ROI region-averaging methods. We show experimentally in test-retest data that our sparse CCA method extracts biologically plausible and stable functional network structures from ASL. We compare the ROI approach to the CCA approach while using CBF measurements alone. We then compare these results to the joint BOLD-CBF networks in a reproducibility study and in a study of functional network structure in traumatic brain injury (TBI). Our results show that the SCCA approach provides significantly more reproducible results compared to region-averaging, and in TBI the SCCA approach reveals connectivity differences not seen with the region averaging approach.

For each metric, using both region averaging (orange) and SCCA (yellow), connectivity matrices were calculated from ASL data acquired in separate acquisitions in the same day and for data acquired one week apart. Whole network correlations were then calculated to examine reliability for the daily (left) and weekly (right) data for each subject. Here we illustrate results using sparsity values of s=t=0.05. A range of sparsity values (s=t) up to 0.25 were examined and these higher values did not produce qualitatively different results.

For each metric, using both region averaging (orange) and SCCA (yellow), connectivity matrices were calculated from ASL data acquired in separate acquisitions in the same day and for data acquired one week apart. Whole network correlations were then calculated to examine reliability for the daily (left) and weekly (right) data for each subject. Here we illustrate results using sparsity values of s=t=0.05. A range of sparsity values (s=t) up to 0.25 were examined and these higher values did not produce qualitatively different results.

J. T. Duda, J. A. Detre, J. Kim, J. Gee, and B. B. Avants, “Fusing functional signals by sparse canonical correlation analysis improves network reproducibility,” in Medical image computing and computer-assisted intervention–miccai 2013, Springer, 2013, pp. 635-642.

Advertisement

Structural and functional connectivity have network-wide influences upon cognitive performance

In this paper functional subnetworks in the brain were examined using MRI to measure both structural connectivity and functional connectivity. Additionally, the influence on behavior of both types of connectivity examined to determine the degree to which each provides unique information as well as how this information may be used to identify the parts of a network that are most influential on behavioral performance. Functional connectivity involves co-activation of brain regions during performance of a task while brain recruitment is monitored with fMRI. Structural connectivity is related to the long tract white matter projections that may integrate recruited brain regions biologically. Here we demonstrate how structural and functional connectivity may be used to examine small, functionally defined subnetworks in the brain during performance of a common language task. Functionally defined cortical regions are used along with a population-averaged diffusion tensor atlas to identify the white matter pathways that provide the basis for biological connectivity. A centerline-based method is used to provide a geometric model that facilitates the equidimensional comparison of functional and structural connectivity within a network. Behavioral data are used to identify the relative contributions of function and structure, and the degree to which each provides unique insight into behavior.

Duda, Jeffrey T., “Characterizing Connectivity In Brain Networks Using Magnetic Resonance Imaging” (2010). Publicly accessible Penn Dissertations. Paper 191.

Structural connectivity disruptions after traumatic brain injury

In each hemisphere of the brain, the thalamus and three cortical subregions in the prefrontal cortex were identified and used along with diffusion tensor based fiber tractography to model the white matter fiber bundles that connect the thalamus to each cortical region.

Traumatic brain injury (TBI) is one of the most common causes of long-term disability. Each year, approximately 1.5 million people sustain TBI in the United States alone, causing billions of dollars of economic cost. Among the survivors, many individuals are left with significant long-term cognitive and motor disabilities. However, efforts to identify the neuropathologic correlates of these deficits have gained only limited success to date. The use of more sensitive and reliable in vivo neuroimaging protocols may facilitate the identification of specific brain-behavior relationships in the TBI population. Here we present a study that explores novel methodologies for examining neuroimaging data to gain further insight into TBI.

Two different types of Magnetic Resonance Images (MRI) are used: diffusion tensor (DT) images quantify connectivity patterns in the brain while the T1 modality provides high-resolution images of tissue interfaces. Our objective is to use both modalities to build subject-specific, quantitative models of fiber connections in order to discover effects specific to a neural system. We first use a population-specific average T1 and DT template to label the thalamus and cortical regions of interest. We then build an expected connection model (illustrated above) within this template space that is transferred to subject space in order to provide a prior restriction on probabilistic tracking performed in subject space. This allows for the comparison of properties such as fractional anisotropy (FA) within a common framework along fiber pathways.

Students t-test results after FDR correction at p<0.02 indicate that the left hemisphere connection between thalamus and Brodmann area 10 is affected by TBI. Arrows indicate regions where TBI survivors show reduced FA compared to controls. A sagittal slice from the T1 component of the template is shown for anatomical reference.

J. T. Duda, B. B. Avants, J. Kim, H. Zhang, S. Patel, J. Whyte, and J. C. Gee, “Multivariate Analysis of Thalamo-Cortical Connectivity Loss in TBI,” in Proc. Computer Vision and Pattern Recognition, Ninth IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis (MMBIA), Anchorage, AK, 2008.