Mining Brain Region Connectivity for Alzheimer's Disease Study via Sparse Inverse Covariance Estimation

1
Mining Brain Region Connectivity for Alzheimer's
Disease Study via Sparse Inverse Covariance
Estimation

• Jieping Ye
• Arizona State University

2
Team Members

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

3
Alzheimers Disease (AD)
Effective diagnosis of Alzheimers 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 Alzheimers 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.

4
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.
5
Neuroimaging PET
Neuroimaging parameters are sensitive and
consistent measures of AD.

• FDG-PET 18F-2-fluoro-2-deoxy-D-glucose
  positron emission tomography is a functional
  imaging technique that measures the cerebral
  metabolic rate for glucose.

6
FDG-PET
7
AD Patient Versus Normal Control
Normal Control
AD Patient
8
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.

9
Our Hypothesis

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

10
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.

11
Sparse Inverse Covariance Estimation

• Given the observations xiN(µ, S), the empirical
  covariance matrix S is
• We can estimate by solving the
  following maximum likelihood problem
• By penalizing the L1-norm, we can obtain the
  sparse inverse covariance matrix

12
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 T 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

13
Related Work

• O. Banerjee, L. El Ghaoui, and A. dAspremont.
  Model selection through sparse maximum likelihood
  estimation for multivariate Gaussian or binary
  data. Journal of Machine Learning Ressearch,
  9485516, 2008.
• J. Friedman, T. Hastie, and R. Tibshirani. Sparse
  inverse covariance estimation with the graphical
  lasso. Biostatistics, 8(1)110, 2008.
• Jianqing Fan, Yang Feng and Yichao Wu. Network
  exploration via the adaptive Lasso and SCAD
  penalties. Annals of Applied Statistics, 2009.

14
Example Gene Network
Rosetta Inpharmatics Compendium of gene
expression profiles described by Hughes et al.
(2000)
15
Example Senate Voting Records Data (2004-06)
Republican senators
Democratic senators
Senator Allen (R, VA) unites two otherwise
separate groups of Republicans and also provides
a connection to the large cluster of Democrats
through Ben Nelson (D, NE), which also supports
media statements made about him prior to his 2006
re-election campaign.
Chafee (R, RI) has only Democrats as his
neighbors, an observation that supports media
statements made by and about Chafee during those
years.
16
Proposed SICE Algorithm

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

17
Block Coordinate Descent








P. Tseng. Convergence of block coordinate descent
method for nondifferentiable maximation. J. Opt.
Theory and Applications, 109(3)474494, 2001.
18
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
19
Brain Regions
20
Experimental result
frontal, parietal, occipital, and temporal lobes
in order
21
Experimental result
22
Experimental result
frontal, parietal, occipital, and temporal lobes
in order
AD MCI
NC
23
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.

24
Key Observations Between-Lobe Connectivity

• In general, human brains tend to have less
  between-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. PLoS Comput Biol 4(6) 1-11.

25
Brain Connectivity for Normal Controls
26
Brain Connectivity of Mild Cognitive Impairment
27
Brain Connectivity for AD Patients
The connectivity between the parietal and
occipital lobes of AD is significantly more than
NC. (help preserve cognitive functions)
28
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 (fMRI) data.

29
Thank you!
30
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.