Decoding mental concept identities using electrocorticography

PROJECT SUMMARY/ABSTRACT Aphasia is a common and disabling outcome following stroke. Although some treatments are available in the acute phase, people with chronic, severe deficits rarely have meaningful recovery. Frequently, these patients have phonological or articulatory planning deficits, while their semantic functions are preserved. Because of this, a novel treatment modality in these patients is a speech brain-computer interface (BCI) designed to decode semantic activity. In this project we are developing a machine learning model to decode brain activity to concept identities, to be used in such a device. We will first develop the model in patients with no language deficits using invasive electrical recordings. During awake brain surgeries, we will place high-density electrocorticography (ECoG) grids on prespecified brain locations corresponding to high-level semantic areas. Patients will perform a semantic decision task, and the neural network model will be trained to predict concept identities from the recorded ECoG activity using a semantic model developed by our lab. We will then demonstrate the application of this model to people with aphasia by performing the same task using the noninvasive magnetoencephalography (MEG) in people with severe aphasia. Demonstrating that this model can be used to decode concept identities from brain activity, and that it is applicable to people with severe aphasia, will open up a new avenue of treatment for this population.