The Journal of Machine Learning Research
Arabica: Robust ICA in a Pipeline
ICA '09 Proceedings of the 8th International Conference on Independent Component Analysis and Signal Separation
Fast and robust fixed-point algorithms for independent component analysis
IEEE Transactions on Neural Networks
| Hi-index | 0.00 |
Functional brain mapping is often performed by analysing neuronal responses evoked by external stimulation. Assuming constant brain responses to repeated identical stimuli, averaging across trials is usually applied to improve the typically poor signal-to-noise ratio. However, since wave shape and latency vary from trial to trial, information is lost when averaging. In this work, trial-to-trial jitter in visually evoked magnetoencephalograms (MEG) was estimated and compensated for, improving the characterisation of neuronal responses. A denoising source separation (DSS) algorithm including a template based denoising strategy was applied. Independent component analysis (ICA) was used to compute a seed necessary for the template construction. The results are physiologically plausible and indicate a clear improvement compared to the classical averaging method.