Applying deep learning in brain computer interface to classify motor imagery.

The article focuses on a novel deep learning (DL) architecture designed to address challenges in Brain Computer Interface (BCI) systems, particularly the motor imagery problem using electroencephalogram (EEG) signals. This architecture integrates Adaptive Resonance Theory (ART) and Fuzzy ART modules within a neuro-fuzzy framework called S-dFasArt, enabling self-organized learning directly from raw EEG data without prior preprocessing. The system emphasizes interpretability through rule-based fuzzy logic, distinguishing it as "Transparent Deep Learning," and demonstrates improved classification accuracy and adaptability on motor imagery tasks using both professional and low-cost EEG devices like Emotiv EPOC. Experimental results with multiple users show the architecture’s effectiveness in detecting motor intentions and non-motor thoughts, while also addressing intra- and inter-user variability through multi-model learning, rule pruning, and dimensionality reduction techniques.