Home
get_chanlocs: Compute 3-D electrode positions from a 3-D head image ==> Download the get_chanlocs User Guide
The get_chanlocs EEGLAB plug-in is built on functions in FieldTrip to locate 3-D electrode positions from a 3-D scanned head image. Robert Oostenveld, originator of the FieldTrip toolbox, alerted us in 2017 that he and his students in Nijmegen had put functions into FieldTrip to compute positions of scalp electrodes from the recorded 3-D images for one 3-D camera, the Structure scanner mounted to an Apple iPad. (Read Homölle and Oostenveld (2019) and notes on the incorporated FieldTrip functions). We at SCCN have created an EEGLAB plug-in extension, get_chanlocs, to ease the process of digitizing the positions of the electrodes from the acquired 3-D and entering them into the EEG.chanlocs data structure for use with other EEGLAB (plotting and source localization) functions that require electrode position information.
The major advantages of using get_chanlocs to measure electrode positions are that: 1) the 3D image can be recorded quickly (<1 min), thereby saving precious subject time (and attention capacity) better used to record EEG data! The researchers who have been most enthusiastic to hear about get_chanlocs are those collecting data from children and infants -- though even normal adult participants must feel less cognitive capacity for the experimental tasks after sitting, wearing the EEG montage, for 20 min while research assistants record the 3D location of each scalp electrode. 2) The 3D image connects the electrode locations to the head fidicuals in a very concrete and permanent way; future improved head modeling will be able to use the 3D head surface scans to fit to subject MR images or to warp template head models to the actual subject head. 3) Unlike with wand-based electrode localizing (neurologists call this electrode 'digitizing'), retaining the 3D head image allows rechecking the electrode positions (e.g., if some human error occurs on first readout).
In brief, the process is as follows:
Scanning the head surface: A 3-D head image (3-D head ‘scan’) is acquired using the Structure scanner showing the subject wearing the electrode cap; this image acquisition typically requires a minute or less to perform. The resulting 3-D .obj image file is stored along with the EEG data. get_chanlocs also supports use of .obj 3D image files obtained using the itSeez3D scanning app, which we have found to be easier to capture good 3D images with than the Structure scanner's native app (Suggestion: Ask iSeez3D about a non-commercial license).
Localizing the electrodes in the 3D scan: When the data are to be analyzed, the get_chanlocs plug-in, called from the Matlab command line or EEGLAB menu, guides the data analyst through the process of loading the recorded 3-D head image and then clicking on each of the electrodes in the image in a pre-planned order to compute and store their 3-D positions relative to 3 fidicual points on the head (bridge of nose and ears). (Note: in future, this digitizing step may be automated at some point in the future using a machine vision approach). The electrode labels and their 3-D positions relative to the three skull landmarks (‘fiducial points’) are then written directly into the dataset EEG.chanlocs structure. During this process, a montage template created for the montage used in the recorded experiment can be shown by get_chanlocs as a convenient visual reference to speed and minimize human error in the electrode digitization process.
User Guide See the illustrated get_chanlocs User Guide for details.
Uses: Once the digitized electrode positions have been stored in the dataset, further (scalp field plotting and source localization) processes can use the digitized positions.
Ethical considerations: An institutional review board (or equivalent ethics review body) will likely consider head images as personally identifiable information. Here is the IRB-approved UCSD subject Consent form, allowing participants to consent to different degrees of use of their 3D head image, that we use at SCCN.
To achieve high-resolution EEG (effective) source imaging requires (a) an accurate 3-D electrical head model, and (b) accurate co-registration of the 3-D scalp electrode positions to the head model. Several packages are available for fashioning a geometrically accurate head model from an anatomic MR head image. We use Zeynep Akalin Acar's Neuromagnetic Forward problem Toolbox (NFT), which she is now coupling to the first non-invasive, universally applicable method (SCALE) for estimating individual skull conductivity from EEG data (Akalin Acar et al., 2016; more news of this soon!). When a subject MR head image is not available, equivalent dipole models for independent component brain sources can use a template head model. Zeynep has shown that the dipole position fitting process is more accurate when the template head is warped to fit the actual 3-D positions of the electrodes -- IF these are recorded accurately. This kind of warping is performed in Zeynep's NFT toolbox for EEGLAB.
For too long, it has been expensive and/or time consuming (for both experimenter and subject) to record (or 'digitize') the 3-D positions of the scalp electrodes for each subject. In recent years, however, cameras capable of recording images in 3-D have appeared and are now becoming cheaper and more prevalent. Robert Oostenveld, originator of the FieldTrip toolbox, alerted us that he and his students in Nijmegen had added functions to FieldTrip to compute the 3-D positions of scalp electrodes from scanned 3-D images acquired by one such camera, the Structure scanner mounted to an Apple iPad.
Recording the actual electrode positions in a 3-D head image minimizes the time spent by the experimenter and subject on electrode position recording during the recording session to a minute or less, while also minimizing position digitizing system cost (to near $1000) and the space required (to an iPad-sized scanner plus enough space to walk around the seated subject holding the scanner). Digitizing the imaged electrode positions during data preprocessing is made convenient in get_chanlocs by using a montage template. In future, we anticipate an automated template-matching app will reduce time required to simply checking the results of an automated procedure.
The get_chanlocs plug-in has been tested under Matlab 9.1 (R2016b) on Windows 10 as well as OS X 10.10.5. Please provide feedback concerning any incompatibilities, bugs, or feature suggestions using the GitHub issue tracker.
Scanning software: In theory, any combination of 3-D scanning hardware and software that produces a Wavefront OBJ file (.obj) with the corresponding material texture library (.mtl) and JPEG (.jpg) files can be used for the plug-in. get_chanlocs has only been tested with head models produced by the Structure Sensor camera attached to an iPad Air (model A1474). We use the default calibrator app to align the Sensor camera and the tablet camera, and both the default scanning software (Scanner) and a third-party scanning software (itSeez3D).
Scanner vs. itSeez3D: While the default scanning app (Scanner) is free and produces models that are of high enough quality for the plug-in, we find the third-party app (itSeez3D) easier to use. It seems to be more robust, providing better tracking and faster scans while minimizing the effects of adverse lighting conditions. itSeez3D features a user friendly online account system for accessing high-resolution models that are processed on their cloud servers. Users may contact itSeez3D to change processing parameters; for get_chanlocs, we found that increasing the model polygon count beyond 400,000 results in longer processing time without providing an appreciable increase in resolution. Unfortunately, while scanning is free, exporting models (required for get_chanlocs) has a per export or subscription cost. Please contact itSeez3D regarding discounts for educational institutions and other non-commercial purposes.
Incorrect units in resulting electrode locations: 3-D .obj model units are estimated by relating the range of the recorded vertex coordinates to an average-sized head: a captured model that is much larger or smaller than average will cause errors. If your project requires scanning an atypically-sized model (e.g. large bust scan including ECG electrode, arm scan for EMG sleeve, etc.), manually set obj.unit - instead of using ft_determine_units
- to the correct unit used by your scanner {'m','dm','cm','mm'} to avoid complications.
Keyboard settings: Key presses are used to rotate 3-D head models
when selecting electrode locations in get_chanlocs. Key press
parameters should be adjusted per user discretion: macOS and Windows
systems have adjustable Keyboard Properties, where 'Repeat delay' and
'Repeat rate' may be modified. For some versions of macOS, long key
presses will instead bring up an accent selection menu; in such cases,
repeated single key presses can be used to control MATLAB, or users may
disable the accent selection menu and enable repeating keys by typing
(or pasting) the following in the terminal:
defaults write -g ApplePressAndHoldEnabled -bool false
One way to circumvent this issue is to use the 3-D figure rotation tool in MATLAB. First select the rotation tool, then mark electrodes by clicking as normal; to rotate the model, hold the click after selecting an electrode and drag the mouse; else, be sure to press 'r' to remove points as necessary.
Low resolution in head model: Models will have lowered resolution in MATLAB due to how 3-D .obj are imported and handled, even if they have show a reasonable resolution in other 3-D modeling software (e.g. Paint 3D). Increase the polygon count of the model to circumvent this issue (we recommend 400,000 uniform polygons for itSeez3D).
To download get_chanlocs, use the extension manager within EEGLAB. Alternatively, plug-ins are available for manual download from the EEGLAB plug-in list.
Please check the commit history of the plug-in's GitHub repository.
View/download the get_chanlocs User Guide
Creation and documentation by:
Clement Lee, Applications Programmer, SCCN/INC/UCSD, [email protected] Scott Makeig, Director, SCCN/INC/UCSD, [email protected]