Incremental Full Correlation Matrix Analysis for Real-Time fMRI Studies
Report ID: TR-985-16Author: Li, Kai / Sundaram, Narayanan / Wang, Yida / Keller, Bryn / Capotă, Mihai / Anderson, Michael / Cohen, Jonathan / Turk-Browne, Nicholas / Willke, Theodore
Date: 2016-05-31
Pages: 10
Download Formats: |PDF|
Abstract:
Real-time functional magnetic resonance imaging (rtfMRI) is an emerging approach both for studying the function of the human brain, and for neural feedback-based training. To date, rtfMRI has relied exclusively on the real-time analyses that treat activity in different regions as independent of one another. However, critical aspects of brain function may depend on, and be revealed more sensitively by correlations among regions. An exhaustive analysis of such correlations is substantially more demanding computationally, so full correlation matrix analysis (FCMA) of the entire brain has not been carried out in real-time before. This paper presents the algorithms and an implementation of the first real-time system to perform whole brain FCMA in real-time. Our system includes incremental voxel selection, model training, and real-time classification. We have implemented this system on an fMRI scanner connected to a computer cluster over HTTP. Experiments show that our system is able to achieve real-time FCMA analysis of a stream of brain volumes with neurofeedback with less than 200 ms of lag with very few exceptions. The incremental FCMA algorithm running on our real-time fMRI system performs about 1.8x-6.2x faster than using an offline FCMA toolbox in the real-time context while getting comparable neurofeedback results