Neural Engineering
Margaret Leland
Student
East Carolina University
Matthew Walenski
Associate Professor
East Carolina University, United States
Visualization is vital for the identification and interpretation of patterns in data, particularly for rich and complex, multi-dimensional data sets. The event-related potential (ERP) technique presents a good challenge for data visualization. In this technique, voltage data is collected over time at a high sampling rate from multiple electrodes on the scalp, creating an electroencephalogram. The electroencephalogram is time-locked to stimulus presentation (from one or more conditions) and averaged across trials. Typically difference waves are created by computing the voltage difference at each sample point, to identify large voltage differences across stimulus conditions. Data sets therefore include the spatial coordinates for the position of each electrode on the scalp (typically in 3-D Cartesian coordinates), and the magnitude of voltage over time (“waveforms”), for individual stimulus conditions or for their difference waves for 2 or more stimulus conditions. Thus, there could be as many as five dimensions to plot: the x, y, and z coordinates, voltage, and time. There is currently a variety of methods to visualize these data sets, which each simplifies this multidimensional space in different ways. Here we create a novel 3D visualization method to allow for easy visualization of the distribution of ERP effects (voltage or voltage differences) in space (over the scalp) and time.
For our visualization, we envisioned a 3D plot, with 2 dimensions for the spatial distribution of the electrodes, and a third dimension for time. As electrode coordinates are generally specified in 3 dimensions, we used a projection into a 2-dimensional plane of the standard 3D position of each electrode in the International 10-20 system of electrode placement. Our x-axis captured the anterior/posterior direction, and the y-axis the left/right direction. Time is represented in milliseconds along the z-axis. Each data point was represented as a cube in this 3D space (‘voxel’), centered on its space and time coordinates. The voxels are also assigned a transparency value ranging from 0 (fully transparent) to 1 (fully opaque). Voltage (positive or negative) amplitude is represented as a color assigned to each voxel. We created the graphs in MATLAB, to take advantage of built-in color maps that can be dynamically assigned and the ability to rotate 3D graphics in real-time. As a test case, we created a plot for previously collected data examining N400 effects in a word vs. non-word lexical decision task. In this task, the N400 effect is seen as the voltage for non-words being more negative than the voltage for words over centro-parietal electrode positions, with the difference peaking about 400ms post-stimulus onset). The data set included ERPs from 28 scalp electrodes sampled every 2ms in a 900ms epoch with a 100ms baseline (plotted from -100 to 898).
Figure 1 illustrates the initial plot for the N400 effect. Transparency was set to 0.3. The color scale uses the ‘jet’ color map, corresponding to the magnitude of the voltage at that voxel. Though this plot shows all of the data, it is difficult to see the N400 effect that is present – it’s buried in the middle of the figure, surrounded by less negative voxels. To improve visibility, we added a feature to allow the user to filter out any data they don’t want to see. The user can select a range of voltage values to exclude. In Figure 2, we show the same effect from 2 different angles (created by taking still images of the figure after rotating in), with voltages from -1.5 to 2 removed, avoiding all positive voltage values in the data, and removing small negative values close to zero. The filter settings can be determined more objectively based on data quality measures, effect size measures, or other statistical information about the effect. With these changes, the time course and spatial extent of the N400 effect can be more clearly seen. Overall, this novel method has proved to be very useful in visualizing ERP data - it is adaptable and flexibly allows the user to visualize different aspects of the data.
This material is based upon work supported by the National Science Foundation under Grant No. 1950507. Any opinions, findings, and conclusions, or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
The data utilized here was part of a larger, multi-site project examining the neurobiology of language recovery in people with aphasia (NIH P50-DC012283, PI: Thompson), and was also supported by NIH-DC001948 (PI: Thompson).