A Brain-Computer Interface (BCI) acquires brain signals, extracts informative features, and translates these features to commands to control an external device. This work investigates the application of a non-invasive Electroencephalography (EEG)-based BCI to identify brain signal features in regard to actual hand movement . This provides a more refined control for a BCI system in terms of movement parameters. An experiment was performed to collect EEG data from subjects while they performed right and left hand movement. The informative features from the data was obtained using the Wavelet-Common Spatial Pattern (W-CSP) algorithm that provided high temporal-spatial-spectral resolution. The applicability of these features to classify the two movement and to reconstruct the movement profile was studied. SVM classifier is used to classify the two class of hand movement. The spatial patterns of the W-CSP features obtained showed activations in parietal and motor areas of the brain. This work promises to provide a more refined control in BCI by including control of movement speed.

1. EEG-basedClassification of Fast and
Slow Hand Movements Using
Wavelet-CSPAlgorithm
1DEPT. OF ECE, KUET

2. OUTLINE
 Introduction
 Background
 Objectives
 Methodology
 Experiment
 Result Analysis
 Conclusion
DEPT. OF ECE, KUET 2

3. INTRODUCTION
 Brain computer Interface
 Patients with neuro-muscular disorders
 Movement related features use brain signals
 Presence of movement information in the very low frequency
bands of the EEG data
DEPT. OF ECE, KUET 3

4. BACKGROUND
 MEG signal in the low frequency (2-5 Hz) found to contain movement-speed related data
 Event Related Potential of Sensory Motor Rhythm (SMR) to analyze speed-related data
OBJECTIVES
 To investigate the presence of movement-related parameters in lowband
 To classify the speed of movement using LF components
 To study how these are affected if the movement is performed in 4 different directions.
 To develop a BCI with a more refined control
DEPT. OF ECE, KUET 4

5. METHODOLOGY
Fig 1: Block diagram of BCI System
DEPT. OF ECE, KUET 5

6. DEPT. OF ECE, KUET 6
1. Feature extraction
1.1 Common Spatial Pattern (CSP)
 Optimally discriminate between two classes of EEG data
Z = WX (1) Where, W = CSP Projection matrix
X = EEG data in a single trial of size N x T,
N = the number of channels used
T = the number of samples recorded in each trial
Covariance matrix , C= 𝑿𝑿`
𝒕𝒓 𝑿𝑿 `
(𝟐)
FeatureVector , fp = (
𝑣𝑎𝑟 𝑍𝑝
𝑖=1
2𝑗
𝑣𝑎𝑟(𝑍𝑖)
) (3)

7. 1.2 Wavelet-CSP Algorithm
DEPT. OF ECE, KUET 7
 Daubechies wavelets for the creating filter banks
 Half band lowpass filter and half band highpass filter
 Reconstructed from subspaces using reverse process
 Each subband are filtered using CSP
 W-CSP is obtained by 𝒛 𝒘
𝑳
= 𝒘 𝑳 𝒙 𝝎
𝑳
 Distinctive feature obtained by feature vector
equation
Fig 1.2 :
(a) Signal decomposition and reconstruction using
filters and up/down sampling,
(b) Signal decomposition into subspaces to produce
similar results as in DWT at each level.
2. Fisher Linear Discriminant (FLD) Classifier
 Maximizes the ratio of between class scatter to within
𝐹 =
𝐹′ܵB𝐹
𝐹′ܹܵ𝐹
where, SB = between class scatter matrix
Sw = within class scatter matrix obtained from the feature

8. 3. Experiment Protocol
Four direction – North, South, East, West
Slow Movement – 1200ms
Fast Movement – 400ms
Fig 1.3: direction and speed studied
8

9. 4.1 Comparisons
4. Result Analysis:
DEPT. OF ECE, KUET 9
 Low Frequency < 7
 High Frequency 7-100
 All frequency 1-100
 Low frequency performs
better
 W-CSP has greater accuracy
Fig 1.4 : Comparison among various methods

10. 4.2 Effect of muscular activation
DEPT. OF ECE, KUET 10
Fig 1.5 : Better performance using cross validation

11. DEPT. OF ECE, KUET 11
Fig. 7. (i-v) Spatial patterns obtained at five lower frequency subbands
by W-CSP method for subject 1
E. Discriminating features as shown by CSP
 Activation in contra
lateral motor
 parietal cortex.

12. Conclusion
 Low frequency EEG band related to movement information
 Wavelet-CSP algorithm has classification accuracy of 83.71%
 Spatial patterns showed the activation in contra lateral motor area and parietal
regions
 Showed the possibility of introducing a refined control command set to BCI system
DEPT. OF ECE, KUET 12

13. T H A N K YO U !
13