Handwriting decoding as a challenging motor task for EEG Foundation Models

2026-05-15

AI Summary

This study addresses the limited evaluation and adaptability of existing EEG foundation models in fine-grained handwriting decoding tasks, despite their strong performance in coarse-grained motor imagery. For the first time, handwriting decoding is incorporated into the evaluation framework for EEG foundation models, using a multi-subject, single-trial dataset to systematically compare foundation models against specialized lightweight architectures. The analysis further investigates the impact of movement onset timing and signal quality. Results reveal that current foundation models significantly underperform compared to task-specific models; removing movement onset information reduces average accuracy from 41.3% to 32.4%; and enhancing test signal quality boosts peak subject accuracy from 45% to 78%, underscoring the critical roles of temporal alignment and data fidelity in decoding performance.

Abstract

Recent attempts at creating Foundation Models (FMs) for Electroencephalography (EEG) have achieved state-of-the-art performance on multiple tasks including Motor Imagery (MI). These MI tasks have typically involved coarse classification between imagined limb movements. However, the development of foundation models necessitates diverse datasets, both for pretraining and evaluating the progress of these models. In this work, we propose handwriting decoding as a challenging motor task for FMs. We show that several existing datasets are potentially confounded, and introduce a dataset that more rigorously evaluates models. On this dataset, we find that current FMs, despite showing SOTA performance in multiple MI datasets are outperformed by smaller task-specific models. We also highlight challenges specific to EEG-based handwriting decoding to inform future work. In our 4-letter classification task, we show that (a) Knowledge of movement-onset is crucial to reported decoding performance in prior works, with average performance across subjects dropping from $41.3\%$ to $32.4\%$. (b) Increasing test-time signal quality provides significant performance improvements ($45\%$ to $78\%$ in our best subject) compared to scaling training data with single-trial EEG. (c) While scaling training data steadily improves decoding performance, existing FMs do not outperform specialist models in handwriting decoding. We make our code available at https://anonymous.4open.science/r/EEG-Handwriting-BCI-DFCD/

Problem

• handwriting decoding
• EEG Foundation Models
• motor task
• brain-computer interface
• fine motor decoding

Innovation

• handwriting decoding
• EEG Foundation Models
• motor imagery
• BCI benchmarking
• signal quality