The Billion Byte Brain: Combining Physiological Data and Gigabytes of Images to Improve Maps of Brain Activity

• Activity Number: 262
• Type: Topic Contributed
• Date/Time: Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Kary L. Myers*+
• Companies: Carnegie Mellon University
• Keywords: brain imaging ; optical imaging ; image analysis ; large datasets ; physiological noise ; video data

Abstract

Scientists use optical imaging to make maps of stimulus-induced brain activity. However, physiological fluctuations unrelated to the stimulus can mask the changes due to activation. I will present a method for overcoming the physiological noise to reveal the signal of interest. Two features distinguish this work from that reported by others. First, the size of the data I consider is much larger; while a typical article reports experiments with about 12 megabytes of raw data, the experiment I analyze produced over six gigabytes of raw data (almost 25 gigabytes of floating point numbers). As I write this, my working directory contains 750 gigabytes of intermediate results. Second, I present a scientifically appealing way of reducing the noise masking the signal: instead of using decompositions like PCA to separate signal from noise, I augment the images with simultaneously recorded physiological measurements. My current procedure, multiple linear regression, is simple statistically but challenging computationally: I perform almost 20 million individual regressions with 20,000 observations each. By cleaning the data of physiological noise, I hope to make better maps of brain activity.