Catalyzing Inquiry at the Interface of Computing and Biology

236 CATALYZING INQUIRY

and age-matched controls. Some of the depression patients go on to develop Alzheimer’s disease (AD)
and the goal of the MIRIAD project is to measure the changes in brain images, specifically volume
changes in cortical and subcortical gray matter, that correlate with clinical outcome.
Of particular significance from the standpoint of cyberinfrastructure, the MIRIAD project is distrib-
uted among four separate sites: Duke University Neuropsychiatric Imaging Research Laboratory,
Brigham and Women’s Hospital Surgical Planning Laboratory, University of California, Los Angeles
Laboratory of Neuro Imaging, and University of California, San Diego BIRN. Each of these sites has
responsibility for some substantive part of the work, and the work would not be possible without the
BIRN infrastructure to coordinate it.

Duke

Archives

UCLA

AIR Registration

and Lobar Analysis

BWH

Intensity Normalization

and EM Segmentation

Duke

Clinical Analysis

1

2

3

4

BWH Probabilistic

Atlas

(one-time transfer)

UCSD

Supercomputing

MIRIAD Data Flow

1. Uploading of
   retrospective date from
   Duke study


2. Lobar analysis and
   registration of atlas
   to subjects


3. Anatomical segmentation


4. Comparison to clinical
   history

FIGURE 7.1 Steps in data processing in the BIRN MIRIAD project.

1. T2-weighted and proton density (PD) MRI scans from the Duke University longitudinal study are loaded into
   the BIRN data archive (data grid), accessible by members of the MIRIAD group for analysis using the computer
   resources at the University of California, San Diego (UCSD) and the San Diego Supercomputer Center (compute
   grid).


2. The Laboratory of Neuro Imaging at the University of California, Los Angeles (UCLA) performs a nonlinear
   registration to define the three-dimensional geometric mapping between each subject and a standard brain atlas
   that encodes the probabilities of each tissue class at each location in the brain.


3. The Surgical Planning Laboratory at Brigham and Women’s Hospital (BWH) then applies an intensity nor-
   malization and expectation-maximization algorithm to combine the original image pixel intensities (T2 and PD)
   and the tissue probabilities to label each point in the images and to calculate the overall volumes of tissue classes.


4. Duke performs statistical tests on the image-processing results to assess the predictive value of the brain
   morphometry measurements with respect to clinical outcome.