EyeBrain: Left and Right Brain Lateralization Activity Classification Through Pupil Diameter and Fixation Duration

Technical Summary: EyeBrain

Title: EyeBrain: Left and Right Brain Lateralization Activity Classification Through Pupil Diameter and Fixation Duration
Authors: Ko Watanabe, Pooja Pol, Nicolas Großmann, Shoya Ishimaru, and Andreas Dengel

1. Problem Statement

The research addresses the challenge of estimating brain lateralization—the functional specialization of the left and right hemispheres—using non-invasive ocular metrics. While it is well-established that the left hemisphere handles language and arithmetic, and the right hemisphere manages creativity (drawing) and music perception, there is limited research on whether eye-tracking data can distinguish these hemispheric activities. The authors aim to move away from features highly dependent on user interfaces (like gaze coordinates or saccade direction) and instead focus on physiological ocular metrics that reflect internal cognitive states.

2. Methodology

The study employs a supervised machine-learning approach to classify brain activity based on eye-tracking data.

• Data Collection: 34 participants performed four distinct tasks: Arithmetic (Left), Essay Writing (Left), Drawing (Right), and Listening to Music (Right). Data was captured using a Tobii Pro eye-tracker (90Hz) in a controlled environment.
• Preprocessing: A robust pipeline was implemented to remove noise, outliers (e.g., blinks, edge artifacts), and invalid samples. The pupil diameter signal was interpolated to 1000Hz, smoothed via a zero-phase low-pass filter, and normalized using subtractive baseline correction.
• Feature Engineering: The authors used a sliding window algorithm to extract two categories of features:
  • Pupil Diameter (PD): 25 time-domain features (e.g., mean, std, entropy) and 30 frequency-domain features (via FFT) for both left and right eyes, plus Pearson correlation.
  • Fixation (Fx): 11 time-domain features derived from fixation duration and the frequency of fixations (fx_count).
• Classification Models: The study compared SVM (with RBF kernel), XGBoost, and Random Forest. They utilized Leave-One-Participant-Out Cross-Validation (LOPOCV) to ensure the model's ability to generalize to new users, and Leave-One-Activity-Out Cross-Validation (LOAOCV) for individual task recognition.

3. Key Contributions

• Novel Feature Set: Demonstrated that combining pupillometry (temporal/spectral) with fixation duration significantly improves classification accuracy over using pupil diameter alone.
• Parameter Optimization: Identified optimal experimental parameters, specifically determining that a total activity duration of ~7.5 minutes and a sliding window size of approximately 90 seconds yield the highest performance.
• Identification of Critical Metrics: Discovered that fixation-based features (specifically the minimum fixation duration and fixation count) are among the most influential predictors of brain lateralization.
• Conceptual Framework: Introduced the "EyeBrain" concept—a vision for a real-time cognitive monitoring application that provides users with "life logs" of their brain activity balance.

4. Results

• High Classification Accuracy: The methodology achieved a high classification performance, reaching an F1 score of 0.894 for the best-performing configuration.
• Impact of Fixation: Adding fixation features to pupil diameter data resulted in an average F1 score improvement of over 5%.
• Fixation Duration Threshold: The model performed best when the average fixation duration was within the 500–750 ms range; performance declined when fixations were extremely short or excessively long.
• Task-Specific Performance:
  • Music (Right) and Arithmetic (Left) were the most accurately recognized tasks.
  • Drawing (Right) and Essay Writing (Left) showed lower accuracy, likely due to physical movement (head/eye obstructions) and the inherent complexity of the cognitive processes (e.g., the left brain's role in artistic thinking).

5. Significance

This research proves that ocular metrics are robust, non-invasive indicators of lateralized brain activity. By successfully classifying hemispheric dominance through pupil and fixation data, the study paves the way for:

• Cognitive Monitoring: Real-time applications to help individuals balance logical and creative tasks in daily life.
• Neurorehabilitation: Tools to monitor cognitive recovery or health in clinical settings.
• Human-Computer Interaction (HCI): Developing more responsive systems that adapt to a user's current cognitive load or mental state.