Independent component analysis: comparison of algorithms for the investigation of surface electrical brain activity

We compared the performance of 22 algorithms for independent component analysis with the aim to find suitable algorithms for applications in the field of surface electrical brain activity analysis. The quality of the separation is assessed with four performance measures: a correlation coefficient based index, a signal-to-interference ratio, a signal-to-distortion-ratio and the computational demand. Artificial data are used consisting of typical electroencephalogram and evoked potentials signal patterns, e.g. spikes, polyspikes, sharp waves and spindles. We evaluate different noise scenarios and the influence of pre-whitening. The comparisons reveal considerable differences between the algorithms, especially concerning the computational load. Algorithms based on the time structure of the data set seem to have advantages in separation quality especially for sine-shaped signals. Derivates of FastICA and Infomax also attain good results. Our results can serve as a reference for selecting a task-specific algorithm to analyze a large number of signal patterns occurring in the surface electrical brain activity.