Home
Novel Methods for Brain Connectivity Analysis

Understanding the organization of brain networks necessitates development of sophisticated statistical tools that harness power of network analysis and machine learning. We develop novel computational neuroimaging methods to better characterize brain connectivity.


Contributions of multiple factors in explaining observed structural connectivity. (a) Illustration of connectivity matrices encoding observed streamline counts, physical closeness of regions, and genetic similarity between region. The matrices depict the strength of connectivity between regions, as measured by different factors. Rows and columns correspond to regions in the same order. Connectivity matrices are averaged across multiple participants and normalized to the range 0-1 for visualization purposes. (b) Coefficients of different factors in the generative model. Box plots summarize values across all participants. (c) The proportion of explained connectivity by models including different sets of factors. The models including modules and hubs also include closeness and genetic similarity factors. (d) Comparisons of mean values between three models, namely the base model (closeness and genes), the modular model, and the modular yet integrative model.
Meso-scale structures of the human brain network. (a) The connectivity matrix of the brain that defined the network. Edges between nodes were weighted by the number of streamlines (normalized so as to have values between 0 and 100). (b) Model fits with different candidate meso-scale structures; three structures were compared. The proposed inference algorithm was run with different number of communities (x-axis) and the log-likelihood (y-axis) was calculated for each. The upper panel gives the modularity measure (Q) for different number of communities. Comparison with Q shows that the change in the likelihood value as we increase the number of communities, is similar to the change in the traditionally used modularity measure.
The application of Adaptive Clustering for group-wise consistent TOI extraction. The first row shows an atlas subject with all 327 clusters and selected WM tracts. Results for two test subjects are shown in the second and the third rows. The bundles (all from left hemisphere) corresponding to the internal capsule, the inferior fronto occipital fasciculus, the inferior longitudinal fasciculus, the arcuate fasciculus, and the uncinate are shown. It can be seen that while the fiber bundles are comparable, the individual variability is maintained.

Source Codes
Unified community and core-periphery detection in networks

Publications
Tunç B, Verma R, Unifying Inference of Meso-Scale Structures in Networks, PloS One, 10:11, e0143133, 2015
Tunç B, Parker W A, Ingalhalikar M, Verma R, Automated tract extraction via atlas based Adaptive Clustering, NeuroImage, 102:2, 596-607, 2014
Tunç B, Ingalhalikar M, Parker D, Lecoeur J, Singh N, Wolf R L, Macyszyn L, Brem S, Verma R, Individualized Map of White Matter Pathways: Connectivity-Based Paradigm for Neurosurgical Planning, Neurosurgery, 79:4, 568-77, 2016
Tunç B, Smith A R, Wasserman D, Pennec X, Wells W M, Verma R, Pohl K M, Multinomial probabilistic fiber representation for connectivity driven clustering, Information Processing in Medical Imaging (IPMI), vol:23, 730-741, 2013

Home