Project start: 1.10.2025.

Motor imagery (MI) based brain-computer interfaces (BCIs) hold great promise for restoring communication and control capabilities in individuals with motor impairments. Building upon the results of my previous research, this project aims to deepen and expand the investigation into the classification of MI electroencephalographic (EEG) signals, with a focus on methods that enhance application generalization, reduce complexity of problems, and increase accessibility of BCI systems. While my earlier work demonstrated the viability of using time-frequency representations (TFRs) of EEG signals as image inputs to deep neural networks, the proposed research introduces several new directions. First, one-dimensional (1D) approach to signal representation will be explored, where each EEG channel is analyzed individually before being integrated—or alternatively treated separately—for classification. This approach may allow more efficient and interpretable feature extraction. Additionally, a systematic channel importance analysis will be conducted to assess the contribution of each electrode to classification performance. This could enable the development of less time-consuming, less resource consuming BCI devices by reducing the number of necessary electrodes without compromising accuracy—an important step toward practical, user-friendly BCI applications. Another major aim is to apply transfer learning techniques to improve the generalization of developed models. Unlike traditional pipelines that depend heavily on manual feature selection and preprocessing techniques such as Independent Component Analysis (ICA), the aim is to create a more automated and robust system that minimizes reliance on domain expertise. Finally, the research will be extended to include the data collected using the IDUN Guardian headset—an innovative in-ear EEG device acquired with funding from my previous project.

Together, these components aim to advance the development of adaptable, interpretable, and practical EEG-based BCI systems.