MEGAN: A Software Package for MEG and EEG Analysis & Visualization

Neural ElectroMagnetic (NEM) methods, (MEG and EEG) allow noninvasive study of neuronal activity in the brain by measuring the magnetic field outside the head or the electric potential on the scalp.  For decades EEG (and more recently MEG) have been employed for investigating the temporal dynamics of neural population activity.  The more recent use of NEM techniques for localizing current sources within the brain creates additional technical and analytical  requirements. We have developed the software package MEGAN in response to the requirements for reliable signal processing, data visualization, source localization and temporal analysis capabilities, and a convenient user interface, with the intent to make it available to the brain mapping community.

System Capabilities:  MEG/EEG data comes in a variety of formats, depending upon the sensor system that was used, on the form in which the data was saved, and on the type of experiment that was done.  MEGAN currently handles MEG data from several sensor systems and is being extended to accept user-written modules to read data from other systems and to provide full support for EEG data. It supports both continuous and averaged evoked response data.  Continuous data may consist of spontaneous activity or may contain embedded sensory or behavioral responses;  MEGAN provides capability for retrospective averaging relative to a stimulus or response record.  This capability facilitates experimental paradigms that employ rapid or complex designs that produce temporally overlapping responses. MEGAN provides a rich variety of visualization options for selecting epochs of spontaneous data, data conditioning, and viewing of the data in a variety of forms, for example as field distributions or as waveform displays.  All of the forms of data that MEGAN handles can be written to our standard  netMEG file, a flexible, extensible, self-documenting, and highly portable file, written using the netCDF format.  We have developed code to read the netMEG file into programs written in C, Fortran, MATLAB and IDL.  MEGAN is written in IDL, which has many advantages as a development and interactive runtime environment.

MEGAN emphasizes signal processing because NEM data is especially vulnerable to noise contamination, both from the subject's own body and from environmental sources, such as the 60 Hz AC power signal.  A variety of sensor designs, hardware filters, and shielding are used to reduce recorded noise, but digital filtering and artifact rejection in post-processing are also useful.  MEGAN offers a great deal of flexibility and user control in signal processing, with options tailored to each type of data, as well as general, highly configurable signal processing routines, and the ability to call user-supplied routines.  Capabilities for Fourier analyses are provided, both for analysis of oscillatory activity and as tools for digital filtering.  MEGAN is an event driven program, with the user controlling the flow of processing by means of a point and click graphical user interface.  It can also be run in batch mode using an automatically developed script.  NEM experimental designs often employ many resolvable variations of a stimulus or behavioral response during a single experimental session.  MEGAN allows these cases to be analyzed separately using the same processing stream, in order to facilitate comparisons between experimental conditions.

Localization of the sources of neuronal activity using NEM methods requires explicit computational models of the sources and of  head volume conductivity; even with such models, results are ambiguous.  Geometrical models of head and cortical anatomy provided by another software tool developed in our laboratory, MRIVIEW, can improve the accuracy and reliability of such estimates.  Widely used analytical strategies such as multi-dipole spatial temporal modeling are powerful but are highly interactive. Newer strategies including automatic multi-start methods, constrained linear inverse procedures and Bayesian analysis can proceed automatically after setup but may require considerable computing time. MEGAN handles such computationally intensive tasks by spawning independent processes that can proceed while the interactive process remains available for use.  Results from such analyses may be retrieved by MEGAN for visualization and further analysis.  MEGAN currently supports two multi-start routines, and provides mechanisms to accommodate new methods as they mature. A few examples of MEGAN screens follow, illustrating the three major areas of MEGAN functionality: signal processing, data visualization, and data analysis.

(click to enlarge)

The Channel Selector is a user interface for finding and inactivating bad channels.  The data shown here is from an MEG visual evoked response study done on the Neuromag whole head system. Each subplot displays the signal for one stimulus condition from all channels.  The user has clicked on a questionable waveform in one plot, and the corresponding waveforms in all of the subplots have been highlighted. The "Selected Channel" box indicates that the selected waveform is from Channel 106,  and the "Channel Passes" list shows that, on average, only 28 epochs from this channel survived the artifact rejection process for each stimulus condition.  The "Stimulus, Averaged Passes Used" list to the left, shows that there were originally at least 77 epochs of data for each stimulus condition.  The data from channel 106 can be inactivated by clicking the button next to its name on the lower right.

(click to enlarge)

(click to enlarge)

The Waveform Location and Waveform Overplots windows are shown above.  In the Waveform Location window, each waveform is plotted starting from the location of its sensor, projected onto a plane.  A variety of projection methods are available; in this case, the user has selected the azimuthal equidistant projection, which shows the entire surface, flattened out.  We are looking down upon the head; the subject's nose would be at the top of the plot.  The data is from the Neuromag system, which has a pair of planar gradiometers at each location.  The user has chosen the option to use a vertical offset to separate the pairs of waveforms from collocated sensors.  The Waveform Location window is a graphical user interface which enables the user to click on waveforms to inactivate them or to select channels of data to display in the Waveform Overplots window.The Waveform Overplots window, which shows selected waveforms plotted on the same axis,  is useful for determining the onset and duration of the response or spike, and for seeing the maximum and minimum amplitudes.  Clicking in the Overplots window will select a waveform or, as shown in this example, a time slice. A vertical marker appears at the selected time on each waveform in the Waveform Location window, and the contour plot for that time slice can be displayed.  If waveform mode were selected, the selected waveform would be highlighted in both the Overplot and Waveform Location windows.

MEGAN supports source localization both by launching calculations, and by supplying visualizations of the results.  In this example, a source localization calculation on somatosensory data was done using MSST, a multistart downhill simplex method for fitting dipoles.  MEGAN offers a variety of results plots, depending upon the information available from the analysis code. In the series of contour plots at selected time slices shown above, the forward solution, the field due to the calculated dipoles, is compared to the original data.  The difference between these two is the residual, the remaining signal that is not explained by the solution dipoles.  The contour plots provide information about the magnitude of the residual, as well as spatial and temporal variations in goodness of fit.  The projection used for these contour plots is the same as the one described for the Waveform Location plot.