Research

Statistical Modeling and Analysis of Medical Imaging Data

BrainRapid advances in medical imaging of the human brain are imposing unprecedented demands for new statistical methods that can be used to detect small differences in brain activity between normal individuals and individuals suffering from brain dysfunction. Stimulated by and in collaboration with Dr. Robert Haley of the University of Texas Southwestern Medical Center at Dallas, Professors Richard F. Gunst, William R. Schucany, and Wayne A. Woodward are leading a group of researchers in the Department of Statistical Science in the development of innovative statistical procedures that will allow more powerful comparisons of brain activity signals from structures deep within the brain than the statistical methods that are currently in use. Their work is focused on combining the notoriously weak signals from individual locations in the brain to produce stronger signals from collections of adjacent locations that lie within deep brain structures. The culmination of this work will provide physicians and medical researchers with more powerful statistical methods for detecting differences between normal and dysfunctional brain activity.

Advances to date include techniques for improved identification of locations within the brain (Carmack et al. 2004), better normalization of the unscaled brain activation signals in regional blood flow using single-photon-emission computed tomography (SPECT) imaging (Spence et al. 2006), and a comprehensive approach to incorporating spatial correlations in the analysis of brain imaging data that greatly improved the detection of brain activity differences between seriously impaired Persian Gulf War veterans and control group veterans (Spence et al. 2007). All of these efforts involved graduate students who co-authored these research articles.

Recent work has focused on the statistical design and analysis of functional Magnetic Resonance Imaging (fMRI) experiments on Persian Gulf War veterans. fMRI imaging is important because it is a noninvasive technology that indirectly measures brain activity through blood oxygen level dependence (BOLD) changes, changes in the magnetic properties of blood flow. This work is critical to the success of the new studies because the statistical analysis methods currently available and in widespread use are relatively new, not yet sufficiently grounded in statistical theory, and often are not powerful enough to identify the types of subtle differences in brain structure and function considered likely to explain the brain illnesses that affect many Gulf War veterans. A number of research activities are currently underway to develop innovative statistical methods both for the design of fMRI experiments and for the analysis of fMRI data, which include hundreds of brain scans on each individual.

Average Residuals from a Quadratic Drift Fit

Investigations of the statistical properties of designs proposed for upcoming fMRI experiments involving Gulf War veterans has led to recommendations for changes in some of the standard protocols for such experiments. Novel applications of wavelets analysis of temporal signals are begin applied to spatiotemporal analysis of fMRI brain imaging data with the intent to produce even more powerful statistical methods for detecting differences in brain activation. Applications of this work have the potential to be used in the detection of brain activation differences in many types of illnesses that result in deep-brain impairment.

Carmack, P.S., Spence, J.S., Gunst, R.F., Schucany, W.R., Woodward, W.A, and Haley, R.W. "Improved Agreement Between Talairach and MNI Coordinate Spaces in Deep Brain Regions," NeuroImage, 22 (2004), 367-371.

Spence, J.S., Carmack, P.S., Gunst, R.F., Schucany, W.R., Woodward, W.A, and Haley, R.W. "Using a White Matter Reference to Remove the Dependency of Global Signal on Experimental Conditions in SPECT Analyses," NeuroImage, 32 (2006), 49-53.

Spence, J.S., Carmack, P.S., Gunst, R.F., Schucany, W.R., Woodward, W.A, and Haley, R.W. "Accounting for Spatial Dependence in the Analysis of SPECT Brain Imaging Data", Journal of the American Statistical Association, 102 (2007), 464-473.