MBIC-Lecture; Michael Tangermann
Decoding the ongoing brain state to guide language rehabilitation
Abstract
A quasi-realtime readout of (limited) brain states can be realized with a brain-computer interface (BCI). Using invasive or non-invasive brain signal recordings, machine learning methods can be trained to estimate, if a BCI user wants to elicit a control command, attends a specific stimulus or engages a functional network.
In my talk, I will describe, how a BCI system can support the rehabilitation after stroke, specifically for patients with a chronic language deficit (aphasia). The system utilizes auditory evoked potentials during a language understanding task. Using a classification model that adapts over time to changing signals, it provides immediate feedback to the patient about how well target words were reflected in the patient's EEG. I will show results of a study with chronic stroke patients with aphasia and discuss the chances of this approach as well as its limitations.