Benjamin Blankertz Guido Dornhege Matthias Krauledat Klaus-Robert Müller Gabriel Curio

Transcription

1 Benjamin Blankertz Guido Dornhege Matthias Krauledat Klaus-Robert Müller Gabriel Curio

2

3 During resting wakefulness distinct idle rhythms located over various brain areas: around 10 Hz (alpha range) over the pericentral sensorimotor cortices. at about 20 Hz over the motor cortex. The μ and β rhythms are blocked by movements of the corresponding body part, (active, passive, or reflexive).

4 Since changes in brain activity are observed over the motor and/or sensory cortex even when a subject is only imagining a movement/sensation in a specific limb -> This can be used for effective BCI control!

5 Ten subjects (all male; 1 left handed; age years, all staff members at the two involved institutions) None of the subjects had extensive training with BCI feedback Recorded with multi-channel EEG amplifiers using 128 channels band-pass filtered between 0.05 and 200 Hz and sampled at 1000 Hz. Surface EMG at both forearms and the right leg, as well as horizontal and vertical EOG signals, were recorded.

6 CALIBRATION SESSIONS POSITION CONTROLLED CURSOR RATE CONTROLLED CURSOR BASKET GAME

7 Visual stimuli indicated which of the following 3 motor imageries the subject should perform: (L) left hand, (R) right hand, or (F) right foot. Target cues were visible on the screen for a duration of 3.5 s, interleaved by periods of random length, 1.75 to 2.25 s, in which the subject could relax. (280 TRIALS for each subject) Data was then screened to adjust subject-specific parameters of the data processing methods Identified the two classes that gave best discrimination and trained a binary classifier (VERTICAL OR HORIZONTAL PARADIGMS ASSIGNED)

8 The initial classifier was behaving suboptimal. So a calibration phase was introduced at the beginning of the feedback sessions in which the subject controlled the cursor freely and the experimenter adjusted the bias and the scaling of the classifier

9 POSITION CONTROLLED CURSOR 25 fps Each trial begins with a black dot that has to be moved into the central area of the screen where the shape of the cursor changed to a cross. Next, the task was to steer the cursor into the highlighted target by imagining the corresponding unilateral hand movements. RATE CONTROLLED CURSOR Similar to position controlled but always starts in middle and position decided by old position, shifted by an amount proportional to the classifier output (more controlled: speed and direction)

10 Here the scene consisted of three targets, gray rectangles at the bottom of the screen and a magenta ball (cursor) that could be directed as it fell. The two outer rectangles were smaller than the middle one to account for the fact that they were easier to hit. In each trial, one of the targets was highlighted in blue, and the subject was trying to direct a cursor in the form of a magenta ball into this target.

11

12 Able to achieve good discrimination between two motor imagery tasks. Includes for all nine subjects the μ-band around 10 Hz, and extends up to the higher β- band for 6/9 subjects. Successfully implemented an individually optimized passband for every subject In this figure: distinguished left and right imagined movement in the first subject

13 One of the ten study participants was unable to produce distinguishable results with the BCI. Experimenters suggest this is probably due to BCI Illiteracy or an incompatibility of a specific person s brain with the Brain-Computer-Interface.

14 (1) robustly transfers the discrimination of mental states from the calibration to the feedback sessions (2) allows a very fast switching between mental states (3) provides reliable feedback directly after a short calibration measurement and machine learning without the need that the subject adapts to the system In 2005, Kübler demonstrated that ALS patients can operate a BCI by the voluntary control of sensorimotor rhythms.

15