Multimodal Fusion Analysis On EEG And FMRI

During the last decade, the fusion of different imaging modalities such as EEG and fMRI becomes a hotspot in cognitive and neural science researches. This idea is a promising direction for activation localization and brain region connectivity detection. EEG and fMRI are complementary modalities with repect to spatiotemporal resolution. Through fusion of EEG and fMRI, high spacial and temporal resolution information about underlying neuro-activity can be achieved, and the limitations in uni-modal analysis of EEG and fMRI are overcomed effectively.In this paper, several essential aspects for EEG/fMRI fusion are investigated. Firstly, the problems, algorithm and application in unified modality analysis are introduced and assessed. Secondly, as the main part of this paper, the whole procedure for EEG/fMRI fusion is illustrated, such as datasets collection, artifacts reduction, head model reconstruction, hemodynamic response function, EEG-fMRI co-registration, and the most promising fusion algorithm. Among these details, head model reconstruction and fusion algorithm are discussed. The principle and equations to potential between cortex and scalp are expressed for spherical head models, realistic head models and complex head models. The main fusion algorithm is separated into two classes, constrained fusion methods and direct data fusion methods. The former includes Kalman filtering, Twomey regularization, Linear inverse estimation and the latter, Spatiotemporal optimiztion, Multiway partial least squares, Symmetrical Bayesian Model. Then, taking the dataset at resting state as an example, NPLS method is practised, and meaningful results are achieved. In the end, several critical problems confronted are addressed, and the future research is represented.