EEG-fMRI fusion of non-triggered data using Kalman filtering

We present a method to combine simultaneous acquisitions of EEG and fMRI measures for estimating ongoing cortical activity. We do not assume that the activity is linked to any repeated stimulation. Rather we solve a very large inverse problem, where EEG and fMRI are noisy measures of the unknown underlying neural activity, and are related to it via realistic physiological models. An extended Kaiman filter is used to estimate neural activity from the combined measurements at all instants and all cortical locations. Actually it is already an interesting tool for analyzing EEG or fMRI acquisition in isolation since it is able to handle common difficulties with the temporal aspect of both modalities (temporal smoothness in the EEG inverse problem, and hemodynamic deconvolution in fMRI). Its application to simulated data shows how it takes advantage of EEG high temporal resolution and fMRI spatial precision when reconstructing sources activity