electrocorticography signal processing – clinical and research

electrocorticography signal processing
– clinical and research applications
Erik Aarnoutse
ECoG electrodes implanted subdurally to find epileptic focus and mapping with stimulation
functional mapping compared to electrocortical
stimulation and functional MRI brain computer interface based on ECoG:"
communication for people with locked-in-syndrome
"
Results
"
next generation: intuitive BCI"
Four
decoding
gestures and phonemes with
fMRIphoneme
and ECoGECoG cl
"
oG) has been shown to be an important platform for Brain-Computer Interfaces (BCIs)
gestures ECoG
g an improved communication channel for profoundly paralyzed individuals.
ults for a project in which we would like to use a multi-channel ECoG grid implant over
s the platform for an intuitive multi-dimensional BCI communication signal.
work of Guenther and Kennedy [1], who was able to synthesize two phonemic sounds
several neurons in the mouth motor area in a human subject.
gestures fMRI
dimensionality of BCI control that can be gained from high"
al responses using distinct mouth motor movements.
Phoneme decoding using high-density ECoG signal from the mouth region of motor cor
Figure!2.!Hand!gestures!that!had!to!be!executed.!The!gestures!differ!in!the!combinations!of!the!fingers!that!
Z.V. Freudenburg1,2, M.J. Vansteensel1, M.G. Bleichner1, E.J. Aarnoutse1, F.S. S. Leijten1, N.F. Ramsey1"
had!to!be!flexed!or!(kept)!extended,!and!in!their!similarity!with!each!other.!‘D’!and!‘V’!are!the!most!alike,!as!
1
Rudolf Magnus Institute of Neuroscience, Dept. of Neurology and Neurosurgery, University Medical Center Utrecht, The Netherlands
2Dept. of Computer Science, Washington University, St. Louis, MO, USA
they!differ!only!in!the!flexion!of!the!middle!finger.!‘D’!and!‘F’!are!‘inverted’!in!terms!of!the!fingers!that!had!to!
Figure!1.!Position!of!the!electrode!grid!(black)!shown!on!the!individual!anatomy.!The!white!lines!indicate!the!
onIntroduction
•
Results
phonemes
ECoG
be!flexed.!‘Y’!is!different!from!all!other!fingers!as!it!is!the!only!gesture!that!does!not!require!a!flexion!of!the!
central!sulcus.!The!red!lines!indicate!the!location!of!the!hand!knob!area,!as!defined!on!the!axial!slices!and!
Four phoneme ECoG classification
thumb.!!
projected!to!the!surface.!For!subject!1!the!grid!was!located!on!the!left!hemisphere,!for!subject!2!the!grid!was!
Electrocorticography (ECoG) has been shown to be an important platform for Brain-Computer Interfaces (BCIs)
for
application in providing an improved communication channel for profoundly paralyzed individuals.
ilepsy
patient
• These are preliminary results for a project in which we would like to use a multi-channel ECoG grid implant over
the
mouth motor cortex as the platform
signal.
d two• We
sessions
" for an intuitive multi-dimensional BCI communication
located!on!the!right!hemisphere.!!
were inspired by the work of Guenther and Kennedy [1], who was able to synthesize
two phonemic sounds
sk . from the spiking rates of several neurons in the mouth motor area in a human subject.
•
•
of
•
Correlation betwee
class activations an
Goal: Determine the dimensionality of BCI control that can be gained from highdensity ECoG
ECoG
grid spectral responses using distinct mouth motor movements.
ty
sensoy cortex.
"
There are similar but d
phoneme
There is an important
the patterns
/p/ and /k/ give the str
"
Methods
hannel
4x8signal
grid
• ECoG
acquisition
"
• Data subject: intractable epilepsy patient
ace electrodes
who consented to performed two sessions
a overt spoken phoneme task .
ng.
"
with
of
Learned classifier spectral response patterns
•
• Implant location: High density ECoG grid
over right mouth motor and sensoy cortex.
• Implant specifications: 32 channel 4x8 grid
with a 1.3mm exposed surface electrodes
3mm center-to-center spacing.
• Task
• Stimuli: visual cues for trails of 4 phoneme
classes (‘p’, ‘k’, ‘ah’, and ‘oh’) and trials
of no mouth movement (‘+’).
s of 4 phoneme
•
with
Learned classifier spectral response patterns
Four phoneme speech c
•
Four phoneme speech classification
•
There are distinct formant
patterns for each
Each speech classifie
negatively correlated
The responses of the
patterns are highly co
The responses of the
patterns are highly co
Correlation between speech classifier
class activations and target labels
Conclusions
nick-ramsey.eu"
neuroprosthesis.eu Utrecht BCI Group: Nick Ramsey (PI) Erik Aarnoutse Mariska VanSteensel Zac Freudenberg Mark Bruurmijn Elmar Pels Mariana Branco Max van den Boom Dept. Clinical Neurophysiology: Frans Leijten Cyrille Ferrier Geertjan Huiskamp Tineke Gebbink Nico Teunissen Dept. Neurosurgery: Peter Van Rijen Bon Verweij Peter Gosselaar