Leveraging Foundation Models for Calibration-Free c-VEP BCIs

TL;DR

This paper tackles the calibration burden of code-modulated Visual Evoked Potential (c-VEP) BCIs by leveraging a pre-trained foundation model (FM) as a frozen EEG encoder and a lightweight two-layer task head. It evaluates three training regimes—Calibration-Free, Limited Calibration, and Within-Subject Calibration—on two diverse datasets (Fast-Stim and Group-Mod) to demonstrate cross-subject generalization and data efficiency. Results show that calibration-free performance reaches around $70$–$72\%$ accuracy on unseen subjects, with substantial gains from limited calibration using only a small fraction of subject data (e.g., $10$–$20\%$) yielding $89$–$93\%$ accuracy, closely matching or exceeding full-calibration baselines. The study demonstrates the feasibility of FM-based, calibration-free c-VEP BCIs across different stimulus designs, suggesting practical applicability for end users including children and individuals with complex disabilities, while noting limitations such as adult-only data and computational demands of the FM.

Abstract

Foundation Models (FMs) have surged in popularity over the past five years, with applications spanning fields from computer vision to natural language processing. Brain-Computer Interfaces (BCIs) have also gained momentum due to their potential to support individuals with complex disabilities. Among BCI paradigms, code-modulated Visual Evoked Potentials (c-VEPs) remain relatively understudied, despite offering high information transfer rates and large selection target capacities. However, c-VEP systems require lengthy calibration sessions, limiting their practicality outside of laboratory settings. In this study, we use a FM for the first time to eliminate the need for lengthy calibration in c-VEP BCI systems. We evaluated two approaches: (1) a truly calibration-free approach requiring no subject-specific data, and (2) a limited calibration approach, where we assessed the benefit of incorporating incremental amounts of calibration data. In both cases, a classification head is trained on data from other subjects. For a new subject, no calibration data is required in the calibration-free setup, making the c-VEP system effectively plug-and-play. The proposed method was tested on two c-VEP datasets. For the calibration-free approach, the average accuracy on the first dataset (n = 17) was 68.8% +/- 17.6%, comparable to the full-calibration performance reported in the original study (66.2% +/- 13.8%), which required approximately 11 minutes of calibration. On the second dataset (n = 12), the calibration-free accuracy was 71.8% +/- 20.2%, versus 93.7% +/- 5.5% from the original study, which required around 3.5 minutes. A limited-calibration approach using only 20% of the subject's data (approximately 43 seconds) yielded 92% +/- 5.2% accuracy. These results indicate that our FM-based approach can effectively eliminate or significantly reduce the need for lengthy calibration in c-VEP BCIs.

Figures (2)

• Figure 1: The signal processing pipeline for applying the EEGPT foundation model to c-VEP datasets under three scenarios: (a) Calibration-Free, (b) Limited Calibration, and (c) Within-Subject Calibration. For each case, the left panel illustrates task head training, while the right panel shows the deployment stage, which was evaluated offline to simulate real-time classification.
• Figure 2: Box plots comparing the results of the Calibration-Free, Limited Calibration, and Within-Subject Calibration approaches for (a) the Fast-Stim dataset and (b) the Group-Mod dataset. For the Limited and Within-Subject Calibration approaches, classification accuracy is shown across varying percentages of calibration data used to train the task head. For reference, the results from the original study for each dataset are also included.