From Unipolar to Bipolar- EEGLAB Weakness?

Originally posted on the EEGLAB Listserv.

Question: “Hi, I’m newuser of EEGLab. Please, could you give me information about to visualize through Plot Channel Data (scroll) the EEG in unipolar montage to bipolar(example, longitudinal montage). I tried to look for extensions andplugin in EEGLab without success. Thanks and regards.”

Response by Michael Villanueva: “ I had the very same question: Back In 2016 I went to my first EEGLAB workshop, and asked: “How do you rearrange the channel order so to look at the EEG in different montages?” The next 5–10 minutes of conversation with an SCCN engineer were nearly nonsensical: He was utterly flummoxed, asking several times over “Why would you want to do that?” The more I explained the why and how we (in neurofeedback land) routinely reordered montages against various references, the more non-sensical we became to each other. In hindsight it was singularly the most illuminating conversation I have ever had with a neuroscientist.

EEGLAB is designed for source space, not scalp space. Whereas tools like WinEEG are great for visual clinical analysis of EEG in scalp space, statistical and numerical analysis are the correct (and only viable) tools in source space. EEGLAB is a clinical research tool that relies on statistical and numerical computation to understand brain function; as a research tool rooted in source space, it does not readily lend itself to forming visually-based clinical diagnostic impressions extracted from the raw channel EEG. Using learning to pilot an aircraft as a metaphor might be helpful to you: When you learn to pilot an aircraft, you are trained to fly by visual rules, commonly called “VFR” (Visual Flight Rules); however, as you advance, you are advanced, you learn to fly the plane only by use of your cockpit instruments, commonly called IFR (Instrument Flight Rules). You can think of neuroscientists who use EEGLAB are using IFR to statistically and numerically navigate brain sources; while clinicians like electro-neurophysiologists, neurologists, and neurofeedback providers using VFR to connect observable pathology to the individual EEG. Both groups operate in “EEG Space” but one specializes in Source Space and the other in Scalp Space.

Even if you took the time using MATLAB to code various montages as suggested, EEGLAB would still be less than adequate. First, the resultant re-montaged raw EEG would not have the channel EEG visual resolution a commercial program like WinEEG already has. Secondly, using Plot > Channel data (scroll) would be painfully slow, and, as such, scanning for transient discharges or watching shifting vigilance states in different montages would be hellish. Thirdly, there is no clinically suitable channel based Power Spectrum Density function in EEGLAB. While Tools > Channel spectra and maps is ok, the frequency resolution combined with the default log power scale occults the variances in power-squared clinical frequency information most useful to us. Could you make a better power spectra graph and maps visual tool in MATLAB? Given enough time and training, yes; however, if you are just starting in MATLAB as you say you are, it would be a rough ride.

In her native source space, EEGLAB is nimble, agile, and powerful; in scalp space, she is hobbled and painfully slow.

For 4 years now, we use WinEEG and ICA in our clinical practice for rapid visual-based clinical examination of the raw in accordance with standard electroencephalography. We do think EEGLAB has untapped and unexplored clinical strengths in several areas, notably multiple model AMICA, however (for us) at this time there is no easy co-existence between source and scalp space. There is simply too much we do not know about the brain.

My advice to you after this lengthy discussion is to forget about forcing EEGLAB into visual clinical work: such efforts are a distraction and a Red Herring. Instead, combine EEGLAB source space strengths with free or commercial programs designed for clinical electrophysiology analysis and simply observe. Personally, I think the hours spent learning how to code and develop work-arounds in MATLAB would be far, far better spent learning the clinical applications of AMICA and ERPs and how to create peer survivable research study designs.

Once you intuitively grasp EEGLAB is all about source space and not scalp space, once you understand (or believe) the time series data recorded at each channel location is a linear sum of all biological and environmental electrical activity, your learning will accelerate: you will make intuitive use of ICA and ERPs in your future clinical formulations, your questions will be more intelligible to the engineers who will help guide you, and your research contributions will be firmly rooted in neuroscience.

Best of luck.”