MEG data analysis overview
Data analysis in LHBD is based on the single trial tomographic source reconstruction of MEG signals, covering the entire brain including cerebral cortex, cerebellum, and brainstem, with sub-millisecond time resolution. We use a range of software, much of it developed in-house. Commercial software (CURRY, BESA, BrainVoyager, SPSS, and VMS and Elekta software) has also been used in some studies. The in-house software allowed standard methods for statistical analysis and novel ones based on information and graph theory concepts to be applied post source reconstruction and thus make sense of the huge volumes of data generated by the unaveraged (single trial) tomographic solutions.
For the tomographic source reconstructions we use magnetic field tomography (MFT), which is a non-linear method for solving the ill-posed biomagnetic inverse problem, under minimal assumptions. MFT has over 30 year history of use in MEG. It was originally developed at the Open University, UK in 1988 and extensively used since then, at the Open University, Juelich Research Center, Germany and RIKEN BSI, Japan. The most up-to-date version of this software is currently at LHBD in AAI Scientific Cultural Services Ltd. and was integrated with the signal processing and the post-MFT statistical analysis tools, creating an easy-to-use data analysis environment.
Standard MEG signal processing (filtering, converting to 3rd order synthetic gradient, etc.) is performed before the MFT source reconstruction. We then use independent component analysis (ICA), implemented in our laboratory, for removing eye and heart artifacts, in addition to any other strong artifacts from the MEG signal.
We have conducted substantial theoretical work and developed practical tools for biomagnetic forward and inverse problem modeling. Much emphasis was put on post source reconstruction single trial data analysis. In the course of several years we have developed theoretical concepts and designed software that now allow us to statistically compare conditions just as in PET and fMRI data analysis (but with a time dependence measured in milliseconds rather than seconds). In addition, it allows us to study interactions between activated areas (functional connectivity) and efficiently mine and extract the necessary information from the massive volumes of data (a typical single trial MFT analysis produces about half a million tomographic images for each subject).
Key findings related to MEG source analysis ...
Our emphasis now is shifting towards adaptation of our methods, including MFT for EEG signals as well, and the development of dedicated hardware and software to record (and analyze) in detail and with millisecond resolution stimulus details, subject state and behavioral output. We expect that these new tools, together with our tomographic solutions, will enable us to explore the concepts of hierarchy, top-down and bottom-up processing, and test their validity and utility in reasoning about how the brain is organized and functions.
More recently, we have developed methods that bypass the computationally heavy single trial tomographic source analysis and directly capture from the MEG signal spatial and temporal patterns of interest using techniques like spherical harmonic expansion (SHE) and ICA. These methods allow us to review fast processes at different temporal scales, for example the accumulation of events with a characteristic timescale measured in milliseconds over minutes or even hours.
A general overview of a framework for an exploratory single trial data analysis ...