brain-computer interface based on motor imagery:

12
Brain-computer Interface Based on Motor Imagery: the Most Relevant Sources of Electrical Brain Activity Alexander A. Frolov 1,2 , Dusan Husek 3 , Vaclav Snasel 1 , Pavel Bobrov 1,2 , Olesya Mokienko 2 , Jaroslav Tintera 4 , and Jan Rydlo 4 1. VŠB Technical University of Ostrava, 17 listopadu 15/2172, 708 33 Ostrava, Czech Republic 2. Institute for Higher Nervous Activity and Neurophysiology of Russian Academy of Sciences, Butlerova 5a, Moscow, Russian Federation 3. Institute of Computer Science, Academy of Sciences of the Czech Republic, Pod Vodárenskou Veži 2, Prague 8, Czech Republic 4. Institute for Clinical and Experimental Medicine, Videnska 1958/9,Praha, Czech Republic

Upload: menora

Post on 23-Feb-2016

220 views

Category:

Documents


0 download

DESCRIPTION

Brain-computer Interface Based on Motor Imagery: the Most Relevant Sources of Electrical Brain Activity. Alexander A. Frolov 1,2 , Dusan Husek 3 , Vaclav Snasel 1 , Pavel Bobrov 1,2 , Olesya Mokienko 2 , Jaroslav Tintera 4 , and Jan Rydlo 4. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Brain-computer Interface Based on Motor Imagery:

Brain-computer Interface Based on Motor Imagery: the Most Relevant Sources of Electrical Brain Activity

Alexander A. Frolov1,2, Dusan Husek3, Vaclav Snasel1, Pavel Bobrov1,2, Olesya Mokienko2, Jaroslav Tintera4, and Jan Rydlo4

1. VŠB Technical University of Ostrava, 17 listopadu 15/2172, 708 33 Ostrava, Czech Republic

2. Institute for Higher Nervous Activity and Neurophysiology of Russian Academy of Sciences, Butlerova 5a, Moscow, Russian Federation

3. Institute of Computer Science, Academy of Sciences of the Czech Republic, Pod Vodárenskou Veži 2, Prague 8, Czech Republic

4. Institute for Clinical and Experimental Medicine, Videnska 1958/9,Praha, Czech Republic

Page 2: Brain-computer Interface Based on Motor Imagery:

Aim of the work

• Localizing the sources of electrical brain activity the most relevant for performance of motor imagery based BCI using individual head model.

• Verifying the results of localization with clusters of fMRI activity.

Page 3: Brain-computer Interface Based on Motor Imagery:

Procedure

Eight subjects have been training to control BCI for 10 days (1 session a day).

Relaxation

4 10

Left hand MI

4 10

Right hand MI

4 10

Foot MI

4 10

Block: 4 tasks are permuted randomly

3 blocks, no feedback 7 blocks, feedback is on

Session

On the 10-th day fMRI session was carried out for each subject, with the same instructions presented

Relaxation Foot MI Left hand MI Right hand MI Correct classifier guess (feedback)

Page 4: Brain-computer Interface Based on Motor Imagery:

Extraction of patterns of EEG activity. Procedure

Experimental day data, X Independent componentsX=Wξ

ICA decomposition

Components, relevant to the BCI performance

IC selectionusing cross-validation

Source localization using the weights of the

optimal components

Source locations

fMRI and anatomic MR scans

Anatomic scans for head model

Task-relevant activations

Step1

Step2

Step3

Page 5: Brain-computer Interface Based on Motor Imagery:

ξn

ξ are activities of the independent components in time

Extraction of patterns of EEG activity. Step 1. Independent Component Analysis

Source2. Source1.

Source3. Source n.

EEG = ×W1

Column of weights Wi is a contribution of thei-th independent component into the signal at all channels

ξ1 +…+ ×Wn

Bell-Sejnowski algorithm was used

Page 6: Brain-computer Interface Based on Motor Imagery:

Extraction of patterns of EEG activity. Step 2. Independent Component Selection

1. Check all triples of independent components using Kohen`s Kappa, κ,obtained by cross-validation (7 blocks testing set, 3 blocks training set, 100 trials)

2. Add a component to the previously obtained set so that κ is maximal3. Repeat until all components are selected

Dependence of κ on the number of IC (Ncmp)

Individual maximum (subject & session dependent): artifact elimination

Optimal triple:the most relevant sources

All ICs: equiv. toEEG channels used

κ

Ncmp

Page 7: Brain-computer Interface Based on Motor Imagery:

Extraction of patterns of EEG activity. Step 2. The most relevant components

These components appeared in optimal triples almost for each session

Hz

Hz

Hz

Hz

Left hand MIRight hand MIFoot MIRelaxation

mu-rhythm ERD in left hand area

mu-rhythm ERD in foot area

mu-rhythm ERD in right hand area

supplementary motor area activity

Page 8: Brain-computer Interface Based on Motor Imagery:

Scalp, 0.35 Sm/m

Bone (skull), 0.0132 Sm/m

Cerebrospinal fluid, 1.79 Sm/m

Gray matter, 0.33 Sm/m

White matter, 0.14 Sm/m

Localization of sources the most relevant to the BCI performance. Step 3.

Finite element model

Page 9: Brain-computer Interface Based on Motor Imagery:

Experiment Approximation

Subject 1 (Examples of potential distribution approximation)

Experiment Approximation

Left hand MI Right hand MI

Residual variance average over all subjects was less than 1%Distance to the closest focus of fMRI activity averaged 9 mm

Localization of sources the most relevant to the BCI performance. Step 3.

Results

Page 10: Brain-computer Interface Based on Motor Imagery:

Localization of sources the most relevant to the BCI performance. Step 3.Mu-rhythm ERD in hand areas

Page 11: Brain-computer Interface Based on Motor Imagery:

Localization of sources the most relevant to the BCI performance. Step 3.

Mu-rhythm ERD in foot area and SMA activity

Page 12: Brain-computer Interface Based on Motor Imagery:

Conclusions and future work

Conclusions

1. The method allows for identification of sources of the brain electrical activity the mostrelevant to motor imagery based BCI performance

2. The relevant sources were localized at the bottom of the central sulcus, i.e. close to the hand representation areas, close to the foot representation area, and in supplementary motor cortex

Future research plans

3. Introduce anisotropy into the model

4. Create fast precise localization instrument for each subject using reciprocal approach. The instrument can be then used as a base for creation of source location-based BCI which idea and implementation has attracted researchers` attention recently.