Professional Documents
Culture Documents
A hybrid BCI based on EEG and fNIRS signals improves the performance of decoding motor
This content has been downloaded from IOPscience. Please scroll down to see the full text.
(http://iopscience.iop.org/1741-2552/12/3/036004)
View the table of contents for this issue, or go to the journal homepage for more
Download details:
IP Address: 169.230.243.252
This content was downloaded on 07/04/2015 at 02:18
E-mail: blxu@sina.cn
Abstract
Objective. In order to increase the number of states classified by a brain–computer interface
(BCI), we utilized a motor imagery task where subjects imagined both force and speed of hand
clenching. Approach. The BCI utilized simultaneously recorded electroencephalographic (EEG)
and functional near-infrared spectroscopy (fNIRS) signals. The time-phase-frequency feature
was extracted from EEG, whereas the HbD [the difference of oxy-hemoglobin (HbO) and deoxy-
hemoglobin (Hb)] feature was used to improve the classification accuracy of fNIRS. The EEG
and fNIRS features were combined and optimized using the joint mutual information (JMI)
feature selection criterion; then the extracted features were classified with the extreme learning
machines (ELMs). Main results. In this study, the averaged classification accuracy of EEG
signals achieved by the time-phase-frequency feature improved by 7%, to 18%, more than the
single-type feature, and improved by 15% more than common spatial pattern (CSP) feature. The
HbD feature of fNIRS signals improved the accuracy by 1%, to 4%, more than Hb, HbO, or HbT
(total hemoglobin). The EEG–fNIRS feature for decoding motor imagery of both force and
speed of hand clenching achieved an accuracy of 89% ± 2%, and improved the accuracy by 1%
to 5% more than the sole EEG or fNIRS feature. Significance. Our novel motor imagery
paradigm improves BCI performance by increasing the number of extracted commands. Both the
time-phase-frequency and the HbD feature improve the classification accuracy of EEG and
fNIRS signals, respectively, and the hybrid EEG–fNIRS technique achieves a higher decoding
accuracy for two-class motor imagery, which may provide the framework for future multi-modal
online BCI systems.
Keywords: EEG-fNIRS, hand clenching force and speed, motor imagery, joint mutual
information (JMI), extreme learning machines (ELMs)
7
These authors contributed equally to the study.
2
J. Neural Eng. 12 (2015) 036004 X Yin et al
3
J. Neural Eng. 12 (2015) 036004 X Yin et al
4
J. Neural Eng. 12 (2015) 036004 X Yin et al
Figure 4. The time course of the four preprocessed fNIRS concentration signals activated during hand clenching force tasks.
5
J. Neural Eng. 12 (2015) 036004 X Yin et al
Table 1. Decoding accuracies of different EEG feature types and their combinations.
Where Xk is the feature to evaluate, Xj is the feature that has ⎡ βT⎤ ⎡tT⎤
been selected already, S is the selected feature set, and Y is the ⎢ 1 ⎥ ⎢ 1 ⎥
β=⎢ ⋮ ⎥ and T = ⎢ ⋮ ⎥ .
feature label. This equation stands for the information ⎢ β T⎥ ⎢⎣ tN T ⎥⎦
between a joint random variable XkXj and the targets. The ⎣ N ⎦N × m N ×m
details of the algorithm can be found in [37].
When the input weight wi and the bias bi are given, the
2.6. Classification hidden layer output matrix H can be calculated. The output
weight β can be calculated using the equation given by
In this study, the extreme learning machines (ELMs) were
used to classify between hand clenching force and speed β = H †T, (9)
motor imageries of the right hand. ELMs are types of single-
hidden layer feedforward neural networks (SLFNs) [38], where H+ is the Moore–Penrose generalized inverse of matrix
which can be used for regression and multiclass classifica- H [41].
tions [39]. Unlike other feedforward neural networks that use Liang [42] showed that although the accuracy of ELMs is
gradient-based learning algorithms to tune all the network comparable with that of support vector machines, ELMs
parameters iteratively, ELMs use an analytical approach by require less computational cost and shorter training time.
choosing the input weights randomly and then determining
the output weights of SLFNs. The advantages of ELMs
include extremely fast learning speed, small training error, 3. Results
and good generalization performance. Huang et al [40] rig-
orously proved that the input weights and hidden layer biases
3.1. EEG
of SLFNs with infinitely differentiable activation functions
can be randomly assigned. The classification accuracies between hand clenching force
{
For a training set ℵ = (xi , ti ) xi ∈ R n , ti ∈ R m , i = and speed motor imageries of the right hand are shown in
1, … , N , standard SLFNs with activation function g (x ) and a table 1 and figure 5. When the power, IA, IP, and IF features
hidden node number N are mathematically modeled next are used alone, their decoding accuracies are 0.70 ± 0.04,
N N 0.72 ± 0.02, 0.81 ± 0.04, and 0.79 ± 0.02, respectively. The
∑βi gi ( x j ) = ∑βi g ( wi phase feature (IP feature) and its derivative feature (IF fea-
)
⋅ x j + bi = t j , j = 1, … , N ,
ture) outperform the power and amplitude features. When the
i=1 i=1
four features are used simultaneously, the classification
(7) accuracy can be increased to 0.88 ± 0.01. The CSP method
where wi is the input weight vector that connects the input performs a weighting of the electrodes to maximize the dif-
nodes and ith hidden node, β is the output weight vector that ference between two different motor tasks, and the channel
connects ith hidden node and the output nodes, and bi is the variance of the filtered signal is used for the classification
bias of ith hidden node. [43]. In this study, we find that the averaged classification
Equation (7) can be written in a compact format as accuracy using CSP is 0.73 ± 0.03, which is worse than that of
the time-phase-frequency feature. This result demonstrates
Hβ = T, (8) that using motor parameter imagery for BCI applications is
possible and the time-phase-frequency method outperforms
where the traditional CSP method.
When the power and IA features are used, the average
⎡ g ( w1 ⋅ x1 + b1) ⋯ g ( w1 ⋅ x1 + bN ) ⎤
⎢ ⎥ accuracy for the trained group is better than that for the non-
H=⎢ ⋮ ⋯ ⋮ ⎥ , trained group. However, when the IP and IF features are used,
⎢⎣ g ( w1 ⋅ xN + b1) ⋮ g ( wN ⋅ xN + bN )⎥⎦ the average accuracy for the non-trained group is not worse
N ×N
than that for the trained group. Therefore, the phase feature
6
J. Neural Eng. 12 (2015) 036004 X Yin et al
Figure 5. Decoding accuracies of different EEG feature types and their combinations. (a) Results for every subject; (b) Results for different
groups.
may provide significantly different information compared to The accuracy of the concentration-combined feature is higher
the amplitude feature. than that of the optical-combined feature by 0.03, which
implies that the concentration feature should be chosen
preferentially.
3.2. fNIRS
7
J. Neural Eng. 12 (2015) 036004
Table 2. Decoding accuracies of different fNIRS signal types and their combinations.
subj6 0.77 ± 0.05 0.79 ± 0.09 0.68 ± 0.06 0.79 ± 0.06 0.67 ± 0.02 0.67 ± 0.09 0.71 ± 0.05 0.77 ± 0.04 0.79 ± 0.03
trained 0.76 ± 0.01 0.78 ± 0.01 0.75 ± 0.05 0.78 ± 0.01 0.73 ± 0.04 0.73 ± 0.04 0.77 ± 0.04 0.79 ± 0.02 0.79 ± 0.00
non-trained 0.69 ± 0.05 0.68 ± 0.01 0.66 ± 0.02 0.69 ± 0.02 0.67 ± 0.04 0.66 ± 0.05 0.69 ± 0.04 0.71 ± 0.04 0.72 ± 0.03
all 0.73 ± 0.05 0.73 ± 0.05 0.70 ± 0.06 0.74 ± 0.05 0.70 ± 0.05 0.70 ± 0.06 0.73 ± 0.05 0.75 ± 0.05 0.76 ± 0.05
X Yin et al
J. Neural Eng. 12 (2015) 036004 X Yin et al
Figure 6. Decoding accuracies of different fNIRS signal types and their combinations. (a) Results for every subject; (b) Results for different
groups.
subject 6. Hence, the EEG and fNIRS features can supple- amplitude aspect, such as band power, ERD/ERS, and CSP.
ment each other and improve the classification of different In this study, IP feature and its derivative feature (IF) were
motor imagery tasks. compared with the amplitude feature and its square feature
(power feature). The results show that when used alone, the
phase feature outperforms the amplitude feature by approxi-
mately 10% for the average of all the subjects. When the
4. Discussion amplitude and phase features are used simultaneously, the
classification accuracy between hand clenching force and
4.1. Classification of sole EEG speed motor imageries can be improved to 88% for both
Amplitude and phase are two aspects of a signal. Most fea- trained and non-trained groups. The classification accuracy of
tures used in EEG-based BCI applications are focused on the time-phase-frequency also outperforms the results using CSP
9
J. Neural Eng. 12 (2015) 036004 X Yin et al
Table 3. Decoding accuracies of fNIRS, EEG, and fNIRS-EEG by 15%. The time-phase-frequency feature proposed in the
features. current study can also be extended to time-spatial-phase-fre-
Hb-HbO-HbT- quency feature by adopting the CSP method.
Hb-HbO- Power-IA- HbD-&-power-
Subject HbT-HbD IP-IF IA-IP-IF
4.2. Classification of sole fNIRS
subj1 0.80 ± 0.06 0.89 ± 0.07 0.89 ± 0.02
subj2 0.79 ± 0.04 0.86 ± 0.01 0.87 ± 0.02 fNIRS is a new measurement method that can be used for BCI
subj3 0.69 ± 0.07 0.87 ± 0.06 0.87 ± 0.03 applications. Both concentration features and optical features
subj4 0.70 ± 0.05 0.88 ± 0.01 0.88 ± 0.01 were used in current research. After comparing the two fea-
subj5 0.76 ± 0.04 0.90 ± 0.07 0.93 ± 0.03 ture types, the results demonstrate that the concentration
subj6 0.79 ± 0.03 0.88 ± 0.01 0.92 ± 0.03
signal outperforms the optical signal, implying that the con-
trained 0.79 ± 0.00 0.88 ± 0.01 0.89 ± 0.02
centration signal is more effective in reflecting brain activa-
non-trained 0.72 ± 0.03 0.88 ± 0.01 0.89 ± 0.03
all 0.76 ± 0.05 0.88 ± 0.01 0.89 ± 0.02 tions than optical signals. Moreover, compared with the Hb,
HbO, and HbT features, the HbD feature can obtain the best
Figure 7. Decoding accuracies of fNIRS, EEG, and fNIRS-EEG merged feature. (a) Results for every subject; (b) Results for different
groups.
10
J. Neural Eng. 12 (2015) 036004 X Yin et al
results when used alone. Finally, although the average accu- sub-classes, we only separated hand clenching force from
racy of the Hb-HbO-HbT-HbD feature combination outper- speed motor imagery. In future work, we will identify dif-
forms that of the Hb-HbO-HbT feature combination by only ferent levels of hand clenching force and speed motor
1%, the HbD feature provides new information for cerebral imageries.
activations. In conclusion, the paradigm of hand clenching force and
speed motor imageries proposed in this study is feasible in
increasing control command numbers for BCI applications.
4.3. Classification of EEG–fNIRS
Furthermore, the EEG–fNIRS merged feature can better
The results in this paper show that when the EEG–fNIRS improve classification accuracy compared with the single-
merged feature is used, the decoding accuracy can be signal type approach.
improved by 1% to 5%. The best accuracy using the merged
feature is 93% ± 3% for subject 5, which is high enough,
considering that only one hand is used in the experiment. The Acknowledgments
merged feature outperforms the sole EEG feature and the sole
fNIRS feature. Although the classifier-level merged method This work is supported by the National Natural Science
has been widely used in previous studies [24], the feature- Foundation (NNSF) of China under Grant Nos. 61203368,
level merged method is more fundamental in improving 81470084, and 61463024, and by the National High Tech-
classification performance. Furthermore, the optimized fea- nology Research and Development Program of China (863
tures offer cognitive neuroscience experts a convenient way Program) under Grant No. 2012AA02A605.
of explaining the neuromechanism behind the optimized
features.
EEG presents the neural electrical activity during brain References
activation, whereas fNIRS measures hemoglobin oxygenation
information. There are three reasons why the EEG–fNIRS
merged features can achieve better performance than sole [1] Wolpaw J R, Birbaumer N, Mcfarland D J, Pfurtscheller G and
Vaughan T M 2002 Brain-computer interfaces for
EEG and sole fNIRS. First, when EEG and fNIRS are used communication and control Clinical Neurophysiology 113
simultaneously, more task-relevant information content can 767–91
be enhanced about the motor imagery state of the subject. [2] Vidal J 1973 Toward direct brain-computer communication
Second, these two techniques can complement each other in Ann. Rev. Biophys. Bioeng. 2 157–80
presenting cortex activation during motor imagery. Finally, [3] Scherer R and Pfurtscheller G 2013 Thought-based interaction
with the physical world Trends Cogn. Sci. 17 490–2
feature optimization can also extract the most relevant fea- [4] Ortiz-Rosario A and Adeli H 2013 Brain-computer interface
tures, which can separate hand clenching force and speed technologies: from signal to action Rev. Neuroscience 24
motor imageries. 537–52
[5] Kubler A, Holz E and Botrel L 2013 Brain-computer
interfacing: science fiction has come true Brain 136 2001–4
4.4. Limitations of the study [6] Donoghue J P 2008 Bridging the brain to the world: a
perspective on neural interface systems Neuron 60 511–21
In our study, we used only a single-distance fNIRS approach, [7] De Massari D, Ruf C A, Furdea A, Matuz T,
and the hemodynamic responses caused by skin blood flow Van Der Heiden L, Halder S, Silvoni S and Birbaumer N
may be present [44–48]. Although a Chebyshev II infinite 2013 Brain communication in the locked-in state Brain 136
impulse response filter at a cutoff frequency of 0.1 Hz was 1989–2000
[8] Sun H Y, Xiang Y and Yang M D 2011 Neurological
used to reduce physiological noise, the systemic interference rehabilitation of stroke patients via motor imaginary-based
may inevitably be present in fNIRS measurements. It is brain-computer interface technology Neural Regen. Res. 6
possible that changes in respiration that cause changes of CO2 2198–202
values in the blood also affect the measured signals [49]. [9] Marshall D, Coyle D, Wilson S and Callaghan M 2013 Games,
However, our objective was to distinguish two motor ima- gameplay, and BCI: the state of the art IEEE T. Comput. Int.
Ai Games 5 82–99
geries and extract considerable useful patterns to improve [10] Bonnet L, Lotte F and Lecuyer A 2013 Two brains, one game:
accuracy, and not to understand the neural mechanisms of design and evaluation of a multiuser BCI video game based
functional brain activity. Our future work will thus examine a on motor imagery IEEE T. Comput. Int. Ai Games 5 185–98
two-detector approach to eliminating the systemic inter- [11] Lachaux J-P, Rodriguez E, Martinerie J and Varela F J 1999
ference and checking whether this approach can significantly Measuring phase synchrony in brain signals Human Brain
Mapping 8 194–208
improve classification accuracy. [12] Gysels E and Celka P 2004 Phase synchronization for the
In addition, the hand clenching force and speed motor recognition of mental tasks in a brain-computer interface
imageries in this study both come from the same right hand. IEEE T. Neur. Sys. Reh. 12 406–15
The results show that when the EEG–fNIRS feature is used, [13] Li J I E and Zhang L 2009 Phase interval value analysis for the
the classification of these two motor imageries reaches 89%. motor imagery task in BCI J.Circuit Sys. Comp. 18 1441–52
[14] Coyle S, Ward T, Markham C and Mcdarby G 2004 On the
The usage of motor parameter imageries in BCI applications suitability of near-infrared (NIR) systems for next-
is, therefore, feasible. Although the hand clenching force generation brain–computer interfaces Phys. Measure. 25
motor imagery and speed motor imagery both contain three 815–22
11
J. Neural Eng. 12 (2015) 036004 X Yin et al
[15] Jobsis F F 1977 Noninvasive, infrared monitoring of cerebral [33] Cui X, Bray S and Reiss A L 2010 Functional near infrared
and myocardial oxygen sufficiency and circulatory spectroscopy (NIRS) signal improvement based on negative
parameters Science 198 1264–7 correlation between oxygenated and deoxygenated
[16] Hoshi Y 2005 Functional near-infrared spectroscopy: potential hemoglobin dynamics NeuroImage 49 3039–46
and limitations in neuroimaging studies Int. Rev. Neurobiol. [34] Jain A K, Duin R P W and Mao J C 2000 Statistical pattern
66 237–66 recognition: a review IEEE T. Pattern Analysis Machine Int.
[17] Hoshi Y 2007 Functional near-infrared spectroscopy: current 22 4–37
status and future prospects J. Biomed. Opt. 12 062106 [35] Yu X, Chum P and Sim K B 2014 Analysis the effect of PCA
[18] Scholkmann F, Kleiser S, Metz A J, Zimmermann R, for feature reduction in non-stationary EEG based motor
Mata Pavia J, Wolf U and Wolf M 2014 A review on imagery of BCI system Optik 125 1498–502
continuous wave functional near-infrared spectroscopy and [36] Brown G, Pocock A, Zhao M-J and Luján M 2012 Conditional
imaging instrumentation and methodology Neuroimage 85 likelihood maximisation: a unifying framework for
6–27 information theoretic feature selection J. Machine Learn.
[19] Sano A and Bakardjian H 2009 Movement-related cortical Res. 13 27–66
evoked potentials using four-limb imagery Int. J. Neurosci. [37] Meyer P E, Schretter C and Bontempi G 2008 Information-
119 639–63 theoretic feature selection in microarray data using variable
[20] Hashimoto Y and Ushiba J 2013 EEG-based classification of complementarity IEEE J. Sel. Topics Signal Proces. 2
imaginary left and right foot movements using beta rebound 261–74
Clinical Neurophysiology 124 2153–60 [38] Huang G B, Zhu Q Y and Siew C K and Ieee 2004 Extreme
[21] Friedrich E V C, Scherer R and Neuper C 2013 Long-term learning machine: A new learning scheme of feedforward
evaluation of a 4-class imagery-based brain-computer neural networks IEEE Int. Joint Conf. Neural Networks
interface Clin. Neurophys. 124 916–27 (New York) (IEEE)
[22] Jochumsen M, Niazi I K, Mrachacz-Kersting N, Farina D and [39] Huang G-B, Zhou H, Ding X and Zhang R 2012 Extreme
Dremstrup K 2013 Detection and classification of learning machine for regression and multiclass classification
movement-related cortical potentials associated with task IEEE Transactions on Systems, Man, and Cybernetics—Part
force and speed J. Neural Eng. 10 056015 B: Cybernetics 42 513–29
[23] Yuan H, Perdoni C and He B 2010 Relationship between speed [40] Huang G B, Zhu Q Y and Siew C K 2006 Extreme learning
and EEG activity during imagined and executed hand machine: Theory and applications Neurocomputing 70
movements J. Neural Eng. 7 026001 489–501
[24] Fazli S, Mehnert J, Steinbrink J, Curio G, Villringer A, [41] Rao C R and Mitra S K 1972 Generalized inverse of a matrix
Muller K R and Blankertz B 2012 Enhanced performance by and its applications Proc. of the Sixth Berkeley Symp. on
a hybrid NIRS-EEG brain computer interface Neuroimage Mathematical Statistics and Probability Volume 1: Theory
59 519–29 of Statistics (Berkeley, CA) (University of California Press)
[25] Matthews F and Pearlmutter B A 2008 Hemodynamics for [42] Liang N Y, Saratchandran P, Huang G B and Sundararajan N
brain-computer interfaces IEEE Signal Proces. Mag. 25 2006 Classification of mental tasks from EEG signals using
87–94 extreme learning machine Int. J. Neural Sys. 16 29–38
[26] Mcfarland D J, Mccane L M, David S V and Wolpaw J R 1997 [43] Ramoser H, Muller-Gerking J and Pfurtscheller G 2000
Spatial filter selection for EEG-based communication Optimal spatial filtering of single trial EEG during imagined
Electroen. Clin. Neuro. 103 386–94 hand movement Reh. Eng., IEEE T. 8 441–6
[27] Mcfarland D J, Miner L A, Vaughan T M and Wolpaw J R [44] Erdoğan S B, Yücel M A and Akın A 2014 Analysis of task-
2000 Mu and beta rhythm topographies during motor evoked systemic interference in fNIRS measurements:
imagery and actual movements Brain Top. 12 177–86 insights from fMRI NeuroImage 87 490–504
[28] Delpy D T, Cope M, Van Der Zee P, Arridge S, Wray S and [45] Gagnon L, Yücel M A, Boas D A and Cooper R J 2013 Further
Wyatt J 1988 Estimation of optical pathlength through tissue improvement in reducing superficial contamination in NIRS
from direct time of flight measurement Phys. Med. Biol. 33 using double short separation measurements NeuroImage 85
1433–42 127–35
[29] Cope M 1991 The development of a near infrared spectroscopy [46] Kirilina E, Jelzow A, Heine A, Niessing M, Wabnitz H,
system and its application for non invasive monitoring of Brühl R, Ittermann B, Jacobs A M and Tachtsidis I 2012 The
cerebral blood and tissue oxygenation in the newborn infants physiological origin of task-evoked systemic artefacts in
PhD dissertation Department of Medical Physics and functional near infrared spectroscopy Neuroimage 61 70–81
Biology, University College London, UK [47] Tachtsidis I, Leung T S, Chopra A, Koh P H, Reid C B and
[30] Duncan A, Meek J H, Clemence M, Elwell C E, Tyszczuk L, Elwell C E 2009 False Positives in Functional Nearinfrared
Cope M and Delpy D T 1995 Optical pathlength Topography, in Oxygen Transport to Tissue XXX (New
measurements on adult head, calf and forearm and the head York: Springer) p 307–14
of the newborn infant using phase resolved optical [48] Takahashi T, Takikawa Y, Kawagoe R, Shibuya S,
spectroscopy Phys. Medicine Bio. 40 295–304 Iwano T and Kitazawa S 2011 Influence of skin blood flow
[31] Le Van Quyen M, Foucher J, Lachaux J, Rodriguez E, Lutz A, on near-infrared spectroscopy signals measured on the
Martinerie J and Varela F J 2001 Comparison of Hilbert forehead during a verbal fluency task Neuroimage 57
transform and wavelet methods for the analysis of neuronal 991–1002
synchrony J. Neuroscience Meth. 111 83–98 [49] Scholkmann F, Gerber U, Wolf M and Wolf U 2013 End-tidal
[32] Yoshino K and Kato T 2012 Vector-based phase classification CO2: an important parameter for a correct interpretation in
of initial dips during word listening using near-infrared functional brain studies using speech tasks Neuroimage 66
spectroscopy Neuroreport 23 947–51 71–9
12