The Illusion of Accuracy in Rehabilitation BCI: Ensuring Robust Neural Feature Learning

Research Contribution:

This paper investigates the robustness of deep learning models in clinical settings. We found that while CNNs achieve high “apparent” accuracy, they are often sensitive to artifacts. By using Explainable AI (XAI) and RSA, we show that our Transformer-based approach is more robust and truly decodes the patient’s intent, making it a reliable solution for neuro-rehabilitation.