Brain Connectivity

An Interactive Platform for Brain Connectivity Learning, Analysis, and Visualization

About Brain Connectivity

Machine Learning

State-of-the-art machine learning algorithms are implemented to convert your neuroimaging data into large-scale brain connectivity networks

Network Analysis

Intuition-oriented brain connectivity network analysis is provided to characterize the brain connectivity network with a small number of neurobiologically meaningful measures

Interactive Visualization

Visualization techniques are applied to unveil/visualize the underlying structures between brain regions using interactive interfaces

Tutorial

Data Structure

The platform currently ONLY supports CSV files. An example of the expected CSV file is shown in below: each row represents one subject, and each column represents one region of interest (ROI). An example of the CSV file that you could use to test the platform can be downloaded here.

Precentral_L Precentral_R Frontal_Sup_L Frontal_Sup_R Frontal_Sup_Orb_L
Tom 0.25 0.23 0.29 0.29 0.37
John 0.28 0.25 0.29 0.27 0.38
Katty 0.21 0.19 0.24 0.22 0.33
Sally 0.28 0.24 0.30 0.28 0.39

How SICE works

SICE, which stands for Sparse Inverse Covariance Estimation, is a machine learning learning technique that was developed in this study to learn the brain connectivity network from high-dimensional neuroimaging data. SICE uses the L1-regularization technique to enhance the robustness of the learning results, and uses convex optimization tools to achieve stable and efficient computation.

By using the L1-regularization, it allows control of the complexity of the network that can either be specified by user of SICE or by model selection criteria such as Bayesian Information Criterion (BIC).

More specifically, the complexity of the network is controlled by the regularization parameter, commonly referred as lambda. For any given lambda, the brain connectivity network that fits the data best (out of all the network models whose complexity is not beyond the complexity specified by lambda) will be identified by SICE.


SICE for Single Group Analysis

SICE can be used to convert your high-dimensional neuroimaging data into a brain connectivity network. Complex network analysis (here is a comprehensive survey) is to characterize the brain connectivity network with a small number of neurobiologically meaningful measures. Network characteristics will be automatically calculated on your brain connectivity network such as the diameter, degree distribution, characteristic path length, global efficiency, clustering coefficient, etc.


Example

SICE for multiple group comparison

SICE was used in this study to compare the brain connectivity networks across three groups, the Alzheimer's disease, Mild Cognitive Impairment, and Normal Aging. A particularly visualization form, a "checkbox" figure is used to visualize each network. For instance, in this checkbox figure, each row and each column represents a brain region, and a check in the figure represents that the two corresponding brain regions are connected. By making the rows and the columns correspond to the same set of brain regions following the same order, it makes the network comparison across groups easier.

Publications

  1. Huang, S., Li, J., Sun, L., Ye, J., Chen, K. and Wu, T., 2009, “Learning Brain Connectivity of Azheimer's Disease from Neuroimaging Data,” Proceedings of Neural Information Processing Systems Conference (NIPS) (paper acceptance rate 8%), Dec. 7-9, 2009, Vancouver, B.C., Canada.
  2. Huang, S., Li, J., Sun, Li., Ye, J., Fleisher, A., Wu, T., Chen, K., and Reiman, E., (2011), “Learning Brain Connectivity of Alzheimer’s Disease by Sparse Inverse Covariance Estimation”, NeuroImage, 50, 935-949.
  3. Huang, S., Li J., Ye, J., Chen, L., Wu, T., Fleisher, A. and Reiman, E., 2011, “Identifying Alzheimer’s Disease-Related Brain Regions from Multi-Modality Neuroimaging Data using Sparse Composite Linear Discrimination Analysis,” Proceedings of Neural Information Processing Systems Conference (NIPS) (paper acceptance rate 4.8%), Dec. 12-17, 2011, Granada, Spain.
  4. Huang, S., Li, J., Ye, J., Fleisher, A., Chen, K. and Wu, T., 2011, “Brain Effective Connectivity Modeling for Alzheimer’s Disease by Sparse Bayesian Network,” The 17th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2011) (paper acceptance rate 17.5%), Aug. 21-24, 2011, San Diego, USA.
  5. Huang, S., Ye, J., Fleisher, A., Chen, K., Reiman, E., Wu, T., and Li, J., 2013, “A Sparse Structure Learning Algorithm for Bayesian Network Identification from High-dimensional Data,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(6), 1328-1342.
  6. Huang, S., Li, J., Chen, K., Wu, T., Ye, J., Wu, X., and Li, Y., 2013, “A Transfer Learning Approach for Network Modeling,” IIE Transactions, 44, 915-931.
  7. Liu, Y., Fabri, P., Zayas-Castro, J. and Huang, S., 2014 “Learning High-dimensional Networks with Nonlinear Interactions by a Novel Tree-Embedded Graphical Model”, Pattern Recognition Letters, 49 (1), 207-213.

Team Members

Dr. Shuai Huang's Group at University of Washington

Principal Investigator

Human brain is the most complex system known to human being. How the brain system efficiently and robustly support the neural computation and information flow is still an open question. Modern neuroimaging techniques have enabled high-throughput measurements and characterization of the function and structure of the brain. Neuroimaging has been found to be a powerful method with enormous implications on both scientific discovery and clinical applications, such as understanding how the brain structure support the cognitive functions, how to identify brain regions disrupted by neurodegenerative diseases, how the disease processes such as Alzheimer’s Disease disrupt the functions, how to monitor the disease progression, and how to evaluate the treatment effect as a more sensitive and reliable index than conventional subjective cognitive measurements, etc. We are interested on the modeling and analysis of the brain systems using multi-modality neuroimaging data, focusing on system-level understanding and innovations in machine learning and statistical analysis, to create an analytic framework that can convert the high-dimensional and noisy neuroimaging data into scientific knowledge and better clinical practices.

Shuai Huang
Assistant Professor

Graduate Research Assistant and Webmaster

Yan's research interest is development of novel network learning algorithms and application of these network models in different applications including healthcare and manufacturing. Yan is also the webmaster who maintains this website. More information about Yan can be found in here.

Yan Jin
PhD Student

Get Connected