Operationalization of conceptual imagery for BCIs

We present a Brain Computer Interface (BCI) system in an asynchronous setting that allows classifying objects in their semantic categories (e.g. a hammer is a tool). For training, we use visual cues that are representative of the concepts (e.g. a hammer image for the concept of hammer). We evaluate the system in an offline synchronous setting and in an online asynchronous setting. We consider two scenarios: the first one, where concepts are in close semantic families (10 subjects) and the second where concepts are from distinctly different categories (10 subjects). We find that both have classification accuracies of 70% and above, although more distant conceptual categories lead to 5% more in classification accuracy.