Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cognitive Load Estimation Using Brain Foundation Models and Interpretability for BCIs

Deeksha M. Shama, Dimitra Emmanouilidou, Ivan J. Tashev

TL;DR

The paper tackles real-time cognitive load estimation for BCIs amid cross-subject variability by leveraging Brain Foundation Models to extract high-resolution EEG embeddings. It introduces LaBraM and CBraMod BFMs adapted for long-term EEG monitoring, with a lightweight fine-tuning approach and fixed-size feature representations via anatomically informed pooling. A Partition SHAP-based explainability probe reveals neurophysiologically consistent fronto-parietal patterns and tracks learning progression over days, while longitudinal results show increased prefrontal involvement as cognitive load decreases. Empirical results show LaBraM achieving superior cross-subject accuracy compared to PSD baselines and end-to-end models, with real-time inference on standard hardware. The work advances scalable, interpretable brain-informed cognitive load monitoring for real-world adaptive learning and BCIs.

Abstract

Accurately monitoring cognitive load in real time is critical for Brain-Computer Interfaces (BCIs) that adapt to user engagement and support personalized learning. Electroencephalography (EEG) offers a non-invasive, cost-effective modality for capturing neural activity, though traditional methods often struggle with cross-subject variability and task-specific preprocessing. We propose leveraging Brain Foundation Models (BFMs), large pre-trained neural networks, to extract generalizable EEG features for cognitive load estimation. We adapt BFMs for long-term EEG monitoring and show that fine-tuning a small subset of layers yields improved accuracy over the state-of-the-art. Despite their scale, BFMs allow for real-time inference with a longer context window. To address often-overlooked interpretability challenges, we apply Partition SHAP (SHapley Additive exPlanations) to quantify feature importance. Our findings reveal consistent emphasis on prefrontal regions linked to cognitive control, while longitudinal trends suggest learning progression. These results position BFMs as efficient and interpretable tools for continuous cognitive load monitoring in real-world BCIs.

Cognitive Load Estimation Using Brain Foundation Models and Interpretability for BCIs

TL;DR

The paper tackles real-time cognitive load estimation for BCIs amid cross-subject variability by leveraging Brain Foundation Models to extract high-resolution EEG embeddings. It introduces LaBraM and CBraMod BFMs adapted for long-term EEG monitoring, with a lightweight fine-tuning approach and fixed-size feature representations via anatomically informed pooling. A Partition SHAP-based explainability probe reveals neurophysiologically consistent fronto-parietal patterns and tracks learning progression over days, while longitudinal results show increased prefrontal involvement as cognitive load decreases. Empirical results show LaBraM achieving superior cross-subject accuracy compared to PSD baselines and end-to-end models, with real-time inference on standard hardware. The work advances scalable, interpretable brain-informed cognitive load monitoring for real-world adaptive learning and BCIs.

Abstract

Accurately monitoring cognitive load in real time is critical for Brain-Computer Interfaces (BCIs) that adapt to user engagement and support personalized learning. Electroencephalography (EEG) offers a non-invasive, cost-effective modality for capturing neural activity, though traditional methods often struggle with cross-subject variability and task-specific preprocessing. We propose leveraging Brain Foundation Models (BFMs), large pre-trained neural networks, to extract generalizable EEG features for cognitive load estimation. We adapt BFMs for long-term EEG monitoring and show that fine-tuning a small subset of layers yields improved accuracy over the state-of-the-art. Despite their scale, BFMs allow for real-time inference with a longer context window. To address often-overlooked interpretability challenges, we apply Partition SHAP (SHapley Additive exPlanations) to quantify feature importance. Our findings reveal consistent emphasis on prefrontal regions linked to cognitive control, while longitudinal trends suggest learning progression. These results position BFMs as efficient and interpretable tools for continuous cognitive load monitoring in real-world BCIs.
Paper Structure (13 sections, 1 equation, 4 figures, 1 table)

This paper contains 13 sections, 1 equation, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Data Collection and EEG Preprocessing
  4. Ground-truth Labels from Adaptive Training System
  5. Feature Extraction with Brain Foundation Models
  6. Estimating and Explaining Cognitive Load
  7. Experiments
  8. Results
  9. Model and Feature performance comparison
  10. Ablation Pooling studies
  11. Effect of variable EEG configurations
  12. SHAP explainability results
  13. Conclusion

Figures (4)

  • Figure 1: Overall pipeline of cognitive load estimation with brain foundation model (BFM) in adaptive training systems.
  • Figure 2: Electrodes and groupings. Green fill: part of 32‑chann setup, underline: part of 28‑chann setup, purple outline: part of 26‑chann.
  • Figure 3: Average correlation trends with confidence intervals over training size (points of cohort additions shown with blue verticals). LaBraM embeddings remain robust for the cognitive load task, while a Linear layer helps avoid overfitting to specific data cohorts. The x-axis appears not-rounded as each point reflects the addition of full participant trial sets.
  • Figure 4: Topomap of SHAP feature relevance, averaged globally (left) and per day (right); red indicates high importance. Daily averages of cognitive load (decreasing over time), focus stability (declining), blink duration (increasing over time) are also shown.