Note
Go to the end to download the full example code.
Importing Data from Eyetracking devices#
Eyetracking devices record a persons point of gaze, usually in relation to a screen. Typically, gaze position (also referred to as eye or pupil position) and pupil size are recorded as separate channels. This section describes how to read data from supported eyetracking manufacturers.
MNE-Python provides functions for reading eyetracking data. When possible, MNE-Python will internally convert and store eyetracking data according to an SI unit (for example radians for position data, and meters for pupil size).
Note
If you have eye tracking data in a format that MNE does not support
yet, you can try reading it using other tools and create an MNE
object from a numpy array. Then you can use
mne.preprocessing.eyetracking.set_channel_types_eyetrack()
to assign the correct eyetrack channel types.
See also
Some MNE functions may not be available to eyetracking and other physiological data, because MNE does not consider them to be data channels. See the Glossary of Common Terms and API Elements for more information.
SR Research (Eyelink) (.asc)#
Note
MNE-Python currently only supports reading Eyelink eyetracking data stored in the ASCII (.asc) format.
Eyelink recordings are stored in the Eyelink Data Format (EDF; .edf), which are
binary files and thus relatively complex to support. To make the data in EDF
files accessible, Eyelink provides the application EDF2ASC, which converts EDF
files to a plain text ASCII format (.asc). These files can be imported
into MNE using mne.io.read_raw_eyelink()
.
Note
The Eyelink Data Format (EDF), should not be confused with the European Data Format, the common EEG data format that also uses the .edf extension.
Supported measurement types from Eyelink files include eye position, pupil
size, saccadic velocity, resolution, and head position (for recordings
collected in remote mode). Eyelink files often report ocular events (blinks,
saccades, and fixations), MNE will store these events as mne.Annotations
.
Blink annotation descriptions will be 'BAD_blink'
. For more information
on the various measurement types that can be present in Eyelink files. read below.
Eye Position Data#
Eyelink samples can report eye position data in pixels, units of visual degrees, or as raw pupil coordinates. Samples are written as (x, y) coordinate pairs (or two pairs for binocular data). The type of position data present in an ASCII file will be detected automatically by MNE. The three types of position data are explained below.
Gaze#
Gaze position data report the estimated (x, y) pixel coordinates of the participants’s gaze on the stimulus screen, compensating for head position changes and distance from the screen. This datatype may be preferable if you are interested in knowing where the participant was looking at on the stimulus screen. The default (0, 0) location for Eyelink systems is at the top left of the screen.
This may be best demonstrated with an example. In the file plotted below,
eyetracking data was recorded while the participant read text on a display.
In this file, as the participant read the each line from left to right, the
x-coordinate increased. When the participant moved their gaze down to read a
new line, the y-coordinate increased, which is why the ypos_right
channel
in the plot below increases over time (for example, at about 4-seconds, and
at about 8-seconds).
import mne
fpath = mne.datasets.misc.data_path() / "eyetracking" / "eyelink"
fname = fpath / "px_textpage_ws.asc"
raw = mne.io.read_raw_eyelink(fname, create_annotations=["blinks"])
cal = mne.preprocessing.eyetracking.read_eyelink_calibration(
fname,
screen_distance=0.7,
screen_size=(0.53, 0.3),
screen_resolution=(1920, 1080),
)[0]
mne.preprocessing.eyetracking.convert_units(raw, calibration=cal, to="radians")
Loading /home/circleci/mne_data/MNE-misc-data/eyetracking/eyelink/px_textpage_ws.asc
Pixel coordinate data detected.Pass `scalings=dict(eyegaze=1e3)` when using plot method to make traces more legible.
Pupil-size area detected.
There are 4 recording blocks in this file. Times between blocks will be annotated with BAD_ACQ_SKIP.
Reading calibration data from /home/circleci/mne_data/MNE-misc-data/eyetracking/eyelink/px_textpage_ws.asc
Converted ['xpos_right', 'ypos_right'] to radians.
Visualizing the data#
cal.plot()
custom_scalings = dict(pupil=1e3)
raw.pick(picks="eyetrack").plot(scalings=custom_scalings)
Using qt as 2D backend.
Note that we passed a custom dict
to the 'scalings'
argument of
mne.io.Raw.plot
. This is because MNE expects the data to be in SI units
(radians for eyegaze data, and meters for pupil size data), but we did not convert
the pupil size data in this example.
Head-Referenced Eye Angle (HREF)#
HREF position data measures eye rotation angles relative to the head. It does not take into account changes in subject head position and angle, or distance from the stimulus screen. This datatype might be preferable for analyses that are interested in eye movement velocities and amplitudes, or for simultaneous and EEG/MEG eyetracking recordings where eye position data are used to identify EOG artifacts.
HREF coordinates are stored in the ASCII file as integer values, with 260 or more units per visual degree, however MNE will convert and store these coordinates in radians. The (0, 0) point of HREF data is arbitrary, as the relationship between the screen position and the coordinates changes as the subject’s head moves.
Below is the same text reading recording that we plotted above, except a new ASCII file was generated, this time using HREF eye position data.
fpath = mne.datasets.misc.data_path() / "eyetracking" / "eyelink"
fname_href = fpath / "HREF_textpage_ws.asc"
raw = mne.io.read_raw_eyelink(fname_href, create_annotations=["blinks"])
custom_scalings = dict(pupil=1e3)
raw.pick(picks="eyetrack").plot(scalings=custom_scalings)
Loading /home/circleci/mne_data/MNE-misc-data/eyetracking/eyelink/HREF_textpage_ws.asc
Head-referenced eye-angle (HREF) data detected.
Pupil-size area detected.
There are 4 recording blocks in this file. Times between blocks will be annotated with BAD_ACQ_SKIP.
Pupil Position#
Pupil position data contains (x, y) coordinate pairs from the eye camera.
It has not been converted to pixels (gaze) or eye angles (HREF). Most use
cases do not require this data type, and caution should be taken when
analyzing raw pupil positions. Note that when plotting data from a
Raw
object containing raw pupil position data, the plot scalings
will likely be incorrect. You can pass custom scalings into the scalings
parameter of mne.io.Raw.plot
so that the signals are legible when plotting.
Warning
If a calibration was not performed prior to data collection, the EyeLink system cannot convert raw pupil position data to pixels (gaze) or eye angle (HREF).
Pupil Size Data#
Pupil size is measured by the EyeLink system at up to 500 samples per second. It may be reported as pupil area, or pupil diameter (i.e. the diameter of a circle/ellipse model fit to the pupil area). Which of these datatypes you get is specified by your recording- and/or your EDF2ASC settings. The pupil size data is not calibrated and reported in arbitrary units. Typical pupil area data range between 800 to 2000 units, with a precision of 1 unit, while pupil diameter data range between 1800-3000 units.
Velocity, resolution, and head position data#
Eyelink files can produce data on saccadic velocity, resolution, and head
position for each sample in the file. MNE will read in these data if they are
present in the file, but will label their channel types as 'misc'
.
Warning
Eyelink’s EDF2ASC API allows for modification of the data and format that is converted to ASCII. However, MNE-Python assumes a specific structure, which the default parameters of EDF2ASC follow. ASCII files should be tab-deliminted, and both Samples and Events should be output. If the data were recorded at 2000Hz, timestamps should be floating point numbers. Manual modification of ASCII conversion via EDF2ASC is not recommended.
Total running time of the script: (0 minutes 4.342 seconds)