Real‐time functional MRI using a PC cluster

A system for the real-time analysis of functional magnetic resonance imaging (fMRI) time series is evaluated. The system exploits the advantages of parallel computing, coupled with an efficient general linear model (GLM) coefficient estimation algorithm, to overcome several issues constraining the analysis of the whole-brain fMRI data in real time. The highly parallel, voxel-wise processing of fMRI data motivated the use of a cluster of personal computers for parallel computation. Aside from gaining a significant increase in computational speed, the PC cluster provides a versatile way to handle the computational requirements of the system. The use of GLM in the supporting software allows substantial parametric analysis to be performed. Results of the real-time analysis of the whole-brain fMRI data of a normal subject performing a simple finger-tapping task demonstrated the capabilities of the system. For a real-time statistical analysis including real-time image reconstruction, realignment for motion correction, smoothing, GLM coefficient estimation, statistical analysis, and update of the displayed activation map, the time required to process the data for each image volume is about 1.034 s for a 64 × 64 × 30 image volume and 2.561 s for a 128 × 128 × 20 image volume, less than the TR set to 3 s. © 2003 Wiley Periodicals, Inc. Concepts Magn Reson Part B (Magn Reson Engineering) 19B: 14–25, 2003.