Mining Brain Region Connectivity for Alzheimers Disease Study

Mining Brain Region Connectivity for Alzheimer's Disease Study via Sparse Inverse Covariance Estimation Jieping Ye Arizona State University

Team Members • Arizona State University – – – Jieping Ye (CSE) Liang Sun (CSE) Jun Liu (CSE) Teresa Wu (IE) Jing Li (IE) Rinkal Patel (CSE) • Banner Alzheimer’s Institute and Banner PET Center – Kewei Chen – Eric Reiman

Alzheimer’s Disease (AD) Effective diagnosis of Alzheimer’s disease (AD) is of primary importance in biomedical research. • Currently, approximately 5 million people in the US – about 10% of the population over 60 are afflicted by Alzheimer’s disease (AD). • The direct cost to care the patients by family members or health care professional is estimated to be over $100 billion per year. • As the population ages over the next several decades, it is expected that the AD cases and the associated costs will go up dramatically.

Neuroimaging: MRI Neuroimaging parameters are sensitive and consistent measures of AD. MRI is a high-resolution structural imaging technique that allows for the visualization of brain anatomy with a high degree of contrast between brain tissue types. Reduced gray matter volume (colored areas) detected by MRI voxel-based morphometry in AD patients compared to normal healthy controls.

Neuroimaging: PET Neuroimaging parameters are sensitive and consistent measures of AD. FDG-PET: [18 F]-2 -fluoro-2 deoxy-D-glucose positron emission tomography is a functional imaging technique that measures the cerebral metabolic rate for glucose.

FDG-PET

AD Patient Versus Normal Control AD Patient

Connectivity Study for AD • Recent studies have demonstrated that AD is closely related to the alternations of the brain network, i. e. , the connectivity among different brain regions – AD patients have decreased hippocampus connectivity with prefrontal cortex (Grady et al. 2001) and cingulate cortex (Heun et al. 2006). • Brain regions are moderately or less inter-connected for AD patients, and cognitive decline in AD patients is associated with disrupted functional connectivity in the brain – Celone et al. 2006, Rombouts et al. 2005, Lustig et al. 2006.

Our Hypothesis • There is significant, quantifiable difference in brain connectivity between AD and normal brains.

Our Main Contributions • Employ sparse inverse covariance estimation for brain region connectivity identification. • Develop a novel algorithm for sparse inverse covariance estimation that facilitates the use of domain knowledge. • Our empirical evaluation on neuroimaging PET data reveals several interesting connectivity patterns consistent with literature findings, and also some new patterns that can help the knowledge discovery of AD.

Sparse Inverse Covariance Estimation • Given the observations xi~N(μ, Σ), the empirical covariance matrix S is • We can estimate by solving the following maximum likelihood problem • By penalizing the L 1 -norm, we can obtain the sparse inverse covariance matrix

Why Sparse Inverse Covariance? • The covariance matrix can be estimated robustly when many entries of the inverse covariance matrix are zero. • The sparse inverse covariance matrix can be interpreted from the perspective of undirected graphical model. – If the ijth component of Θ is zero, then variables i and j are conditionally independent, given the other variables in the multivariate Gaussian distribution. • Many real-world networks are sparse. – Gene interaction network

Related Work • O. Banerjee, L. El Ghaoui, and A. d’Aspremont. Model selection through sparse maximum likelihood estimation for multivariate Gaussian or binary data. Journal of Machine Learning Ressearch, 9: 485– 516, 2008. • J. Friedman, T. Hastie, and R. Tibshirani. Sparse inverse covariance estimation with the graphical lasso. Biostatistics, 8(1): 1– 10, 2008. • Jianqing Fan, Yang Feng and Yichao Wu. Network exploration via the adaptive Lasso and SCAD penalties. Annals of Applied Statistics, 2009.

Example: Gene Network Rosetta Inpharmatics Compendium of gene expression profiles described by Hughes et al. (2000)

Example: Senate Voting Records Data (2004 -06) Republican senators Democratic senators Senator Allen (R, VA) unites two otherwise separate groups of Republicans Chafee (R, RI) has only Democrats as hiscluster neighbors, an observation that. Ben and also provides a connection to the large of Democrats through supports statements made bymedia and about Chafeemade during thosehim years. Nelson (D, media NE), which also supports statements about prior to his 2006 re-election campaign.

Proposed SICE Algorithm • It estimates the matrix Θ directly. • User feedback can be incorporated by adding constraints. • It is based on the block coordinate descent. – Friedman et al. 2008

Block Coordinate Descent P. Tseng. Convergence of block coordinate descent method for nondifferentiable maximation. J. Opt. Theory and Applications, 109(3): 474– 494, 2001.

Data Collected • We used FDG-PET images (49 AD, 116 MCI, 67 NC) • The data were acquired under the support of ADNI – http: //www. loni. ucla. edu/Research/Databases/ Subject AD Male AD Female MCI Male MCI Female NC Male NC Female Number 27 22 76 40 43 24 Mean Age 76. 74 75. 59 75. 84 75. 85 76. 74 75. 33 Std deviation Age 6. 75 6. 11 6. 12 6. 08 6. 26 5. 82

Brain Regions

Experimental result frontal, parietal, occipital, and temporal lobes in order AD MCI occipital parietal frontal temporal NC

Experimental result frontal, parietal, occipital, and temporal lobes in order AD MCI NC

Key Observations: Within-Lobe Connectivity • The temporal lobe of AD has significantly less connectivity than NC. – The decrease in connectivity in the temporal lobe of AD, especially between the Hippocampus and other regions, has been extensively reported in the literature. • The temporal lobe of MCI does not show a significant decrease in connectivity, compared with NC. • The frontal lobe of AD has significantly more connectivity than NC. – Because the regions in the frontal lobe are typically affected later in the course of AD, the increased connectivity in the frontal lobe may help preserve some cognitive functions in AD patients.

Key Observations: Between-Lobe Connectivity • In general, human brains tend to have less betweenfrontal, parietal, occipital, and temporal lobes in order lobe connectivity than within-lobe connectivity. • The connectivity between the parietal and occipital lobes of AD is significantly more than NC which is true especially for mild and weak connectivity. – Compensatory effect – K. Supekar, V. Menon, D. Rubin, M. Musen, M. D. Greicius. (2008) Network Analysis of Intrinsic Functional Brain Connectivity in Alzheimer's Disease. PLo. S Comput Biol 4(6) 1 -11. AD MCI NC

Brain Connectivity for Normal Controls

Brain Connectivity of Mild Cognitive Impairment

Brain Connectivity for AD Patients The connectivity between the parietal and occipital lobes of AD is significantly more than NC. (help preserve cognitive functions)

Conclusion and Future Work • Conclusion – Apply sparse inverse covariance estimation to model functional brain connectivity of AD, MCI, and NC based on PET neuroimaging data. – Our findings are consistent with the previous literature and also show some new aspects that may suggest further investigation in brain connectivity research in the future. • Future work – Investigate the connectivity patterns. – Investigate the connectivity of different brain regions using functional magnetic resonance imaging (f. MRI) data.

Thank you!

PET • Positron emission tomography (PET) is a test that uses a special type of camera and a tracer (radioactive chemical) to look at organs in the body. During the test, the tracer liquid is put into a vein in the arm. The most commonly used for this purpose is a sugar called fluorodeoxyglucose (FDG). The tracer moves through your body, where much of it collects in the specific organ or tissues. The tracer gives off tiny positively charged particles (positrons). The camera records the positrons and turns the recording into pictures on a computer.