An Expectation-naximization Algorithm for Analyzing Multicenter Repeated Functional Magnetic Resonance Imaging Data

• Activity Number: 27
• Type: Contributed
• Date/Time: Sunday, August 8, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Kelly H. Zou*+ and Steven D. Pieper and Meng Wang and Douglas N. Greve and Simon K. Warfield and William M. Wells, III and Ron Kikinis and First Birn
• Companies: Harvard Medical School and Brigham and Women's Hospital and Brigham and Women's Hospital and Massachusetts General Hospital and Brigham and Women's Hospital and Brigham and Women's Hospital and Brigham and Women's Hospital and Harvard University
• Keywords: image rrocessing ; expectation-maximization algorithm ; sensitivity ; specificity ; functional magnetic resonance imaging ; multicenter study

Abstract

Functional magnetic resonance imaging (fMRI) has significantly contributed to understanding both normal and diseased human brains. A large amount of variability often exists in the magnitude, spatial distribution, and statistical significance of the resulting fMRI maps due to differences in equipment and other site-specific differences. Therefore, understanding the effect of these differences in a multicenter functional imaging trial, an efficient pooling and comparison mechanism is desirable, particularly due to the costly imaging, demanding tasks, and analytical burden. We have applied and extended a recently developed expectation-maximization (EM) algorithm, namely Simultaneous Truth and Performance Level Estimation (STAPLE), for evaluating multicenter repeated fMRI data derived from the same set of study subjects across all centers. Over repeated runs per visit at each study center, at each given activation threshold, we used STAPLE to calculate a three-dimensional best estimate brain activation map. Based on such estimated voxel-wise "gold standard," sensitivity, specificity, and predictive values were then estimated using voxel counts.