BLINKER: Automated Extraction of Ocular Indices from EEG Enabling Large-Scale Analysis

Abstract

Electroencephalography (EEG) offers a platform for studying the relationships between behavioral measures, such as blink rate and duration, with neural correlates of fatigue and attention, such as theta and alpha band power. Further, the existence of EEG studies covering a variety of subjects and tasks provides opportunities for the community to better characterize variability of these measures across tasks and subjects. We have implemented an automated pipeline (BLINKER) for extracting ocular indices such as blink rate, blink duration, and blink velocity-amplitude ratios from EEG channels, EOG channels, and/or independent components (ICs). To illustrate the use of our approach, we have applied the pipeline to a large corpus of EEG data (comprising more than 2000 datasets acquired at eight different laboratories) in order to characterize variability of certain ocular indicators across subjects. We also investigate dependence of ocular indices on task in a shooter study. We have implemented our algorithms in a freely available MATLAB toolbox called BLINKER. The toolbox, which is easy to use and can be applied to collections of data without user intervention, can automatically discover which channels or ICs capture blinks. The tools extract blinks, calculate common ocular indices, generate a report for each dataset, dump labeled images of the individual blinks, and provide summary statistics across collections. Users can run BLINKER as a script or as a plugin for EEGLAB. The toolbox is available at https://github.com/VisLab/EEG-Blinks. User documentation and examples appear at http://vislab.github.io/EEG-Blinks/.

Keywords: EEG; EEGLAB; artifact; big data; blink duration; eye blinks; human behavior; machine learning.

Figure 1

Steps used by the BLINKER software to extract blink properties from time series. Each step is described in a separate subsection of the Methods and materials.

Figure 2

Output of BLINKER software showing various blink landmarks. The figure overlays three different time-series (an independent component, an EEG channel, and an upper vertical EOG channel, respectively) corresponding to the time of the blink. The green crosses mark the leftBase and rightBase, respectively. The green horizontal line marks the half-zero duration. The dotted black lines represent the best linear fits to the upStroke and downStroke of the blink, respectively. The thin vertical line is the perpendicular from the tentPeak to the zero line.

Figure 3

Distributions of maximum blink amplitudes for the upper vertical EOG channel of three different datasets. The green line shows the histogram for all potential blinks, and the thick light gray line shows the histogram for blinks whose R² values are at least 0.90 (good). The medium gray line shows the histogram for blinks whose R² values are at least 0.95 (better), and the black line shows the histogram of blinks with R² values of at least 0.98 (best). The magenta line shows the histogram of blinks selected by BLINKER. (A) Dataset has a typical distribution of blinks with a few very large amplitude non-blink artifacts. (B) Dataset has a significant number of low-amplitude non-blink eye movements well-separated from normal blink amplitudes. (C) Dataset has a significant number of non-blink eye movements with moderate R² values and amplitude distribution that overlaps with the normal blink amplitude distribution.

Figure 4

BLINKER channel selection for different datasets. Dots show positions of candidate channels. Labeled channels were selected. Channels labeled in red were picked for only 1 dataset. (A) 64-channel ARL-BCIT driving collection (also has 4 EOG channels). (B) 256-channel ARL-BCIT driving collection (also has 4 EOG channels). (C) 34-channel ARL-Shoot data (also has 4 EOG channels). (D) 32-channel NCTU-LK driving data collection.

Figure 5

Distribution of blink durations for the four collections. Top panel shows a quantile plot against the normal distributions of the average half-zero blink duration for each dataset. The red dotted lines represent the best-fit normal distributions for the each of the four collections. Bottom panel shows the histogram of individual half-zero blink durations for each collection.

Figure 6

ANOVA analysis of blink rate for ARL-Shoot. (A) Top panel shows the distribution of average dataset blink rate for each of the 14 subjects (A through N) over the 9 tasks. The bottom panel shows the same distributions when each average is divided by the average of the dataset blink rates for no-math tasks. (B) Left panel compares the distributions of dataset average blink rates, grouping the math and no-math tasks for all subjects. The right panel shows the same distributions after dividing the dataset average blink rates by the subject's average no-math blink rate.

Figure 7

An output by BLINKER of blink 15 of a specific ARL-BCIT dataset over-plotted with landmarks manually extracted from video: First eyelid movement (red circle), eyelid at top of pupil moving downward (red x), at eyelid reversal (red cross), eyelid at top of pupil moving upward (red square), and last detectable eyelid movement (red diamond). The x-axis label shows the frames and times in seconds of the leftBase point and point of blink maximum amplitude, respectively.

Figure 8

Sample output from the BLINKER group display for the first and last datasets from Table 5. (A) A miss with two closely spaced candidates intermixed with a saccade. (B) A miss with a leading saccade. (C) A miss with a leading saccade. (D) A falsely detected blink that was a saccade.