EEG Forward Source Computation performance benchmark single GPU

A varying number of electric dipoles are assumed inside the brain and their combined effect projected onto the electrodes placed on the scalp is computed. The computation is based on an analytic 4-shell spherical head model.

This operation is an essential step in the EEG source localization problem, where we wish to find the activation sources in the brain based on the measured scalp potential values. The benchmarks used a single V100 GPU and varied the number of electrodes, the number of dipoles, the length of the measurement data (time points where the potential maps is needed) and the parallel thread count. The benchmark programs were written in CUDA 10.2.