Decoding of motor task difficulty and execution speed from EEG data with application to stroke rehabilitation

Official URL: http://risc01.sabanciuniv.edu/record=b1640321 (Table of Contents)

Brain-computer interfaces (BCIs) which provide an alternative channel between human brain and computer world are hope for many patients who su er from neurological diseases such as ALS and stroke. Scientists bene t from BCIs not only for communication but also for rehabilitation and e ective use of robotic exoskeletons. While healthy people can decide the speed level and the amount of force applied for daily activities, stroke patients are not able to transfer their intended movement speed and force magnitude to robotic limbs and mechanical devices. Even though brain signals potentially contain such information about task execution, existing BCI systems do not exploit that information. In this thesis, the possibility of decoding intended human activity speed and task di culty levels from an electroencephalography (EEG) based BCI system is investigated. In particular, two experimental setups are designed to collect data while subjects are performing two di erent tasks, and di erent protocols are proposed with their advantages and drawbacks to extract accurate information from these setups. Moreover, the problem of intention level detection is analyzed in response to task difficulty and speed level. As it is not possible to access the true intention level of a human being to execute a particular task, we use the di culty and speed of the executed task as proxies for the true intention level and then analyze the corresponding neural correlates. We have applied several classi cation protocols and our results indicate that some classi cation protocols are able to detect task speed and di culty from EEG signals.