Contents ● Web surfing with BCI ● Auditory-controlled BCI ● Visual and auditory feedback comparison ● BCI using ECoG ● Comparison of non-invasive BCI approaches
Brain Controlled Web Surfing ● Allow patients to surf the web by concious changes of brain activity ● Enables a completely paralyzed patient to participate in the broad portion of life reflected by the WWW. ● History of providing WWW access to ALS patients dates back to 1999 TTD was used to operate a standard web browser, i.e. Descartes ● Descartes was controlled by binary decisions ● Services provided Writing letters, writing emails, and surfing the web.
Web surfing with “Descartes“ - A ● Patient views a list of predefined WebPages. ● Each webpage is offered successively at the bottom of the screen for selection. ● Page selection through positive SCPs whereas page rejection by negative SCPs.
Web surfing with “Descartes“ - B ● Page loaded after its selection and shown for a predefined period of time.
Web surfing with “Descartes“ - C ● The links on the previous page are offered alphabetically as a dichotomous tree. ● Subject will select or reject each item by regulating SCPs
“Nessi“ – An Improved Graphical Brain-Controllable Browser
Comparison between Visual and Auditory Feedback
Functional MRI and BCI ● BCI combined with FMRI to uncover relevant areas of brain activation during regulation of SCPs. ● EEG from 12 healthy subjects was recorded inside an MRI scanner while they regulate their SCPs. ● Successful positive SCP shift was related to an increase of blood oxygen level dependent (BOLD) in the anterior basal ganglia. ● While negativity was related to an increased BOLD in the thalamus.
SVM Classification of Autoregressive Coefficients: ● In contrast to SCPs: Frequency range below 1Hz Classified according to their time domain representation ● EEG correlates of an imagined-movement as best represented by oscillatory features of higher frequencies, i.e. 8-15 and 20-30 Hz Desynchronization of μ–rhythm over motor areas. ● Coefficients of a fitted autoregressive (AR) model were used to realize this phenomena. ● SVM was them employed for the classification of these AR coefficients.
SVM Classification of Autoregressive Coefficients:
BCI using ECoG signals: ● EEG: Limited signal-to-noise ratio Low frequency range ● Invasive ECoG signals: Broader frequency range (0.016 to 300 Hz) Increased signal-to-noise ratio 3 out of 5 epilepsy patients were able to spell their names within only one or two training sessions. ● ECoG signals were derived from a 64-electrode grid placed over motor-related areas. ● Imagery of finger or tongue movements was classified with SVM classification of AR coefficients.
Comparison of Noninvasive Input Signals for BCI ● Noninvasive BCI: Sensorimotor rhythms (SMR) Slow cortical potentials (SCPs) P300 ● Extensively studied in healthy participants and to a lesser extent in patients. ● For this reason SCP-, SMR-, and P300-based BCIs were compared for free spelling.
Comparison Study ● SCPs: None of the seven patients showed sufficient performance after 20 sessions. ● SMR Half the patients showed an accuracy ranging from 71 to 81 %. ● P300 Performance ranged from 31.7 to 86.3 %