Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering

BBE 197 1–11 biocybernetics and biomedical engineering xxx (2017) xxx–xxx

Available online at www.sciencedirect.com

ScienceDirect journal homepage: www.elsevier.com/locate/bbe 1 2 3

Original Research Article

Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering

4 5 6

7 8

Q1

M.H. Quazi *, S.G. Kahalekar Department of Instrumentation Engineering, SGGSIE&T, Nanded (MS), India

article info

abstract

Article history:

Electroencephalogram (EEG) denotes a neurophysiologic measurement, which observes the

Received 16 November 2016

electrical activity of the brain through making a record of the EEG signal from the electrodes

Received in revised form

positioned on the scalp. The EEG signal gets mixed with other biological signals, called

11 March 2017

artifacts. Few artifacts include electromyogram (EMG), electrocardiogram (ECG) and elec-

Accepted 14 April 2017

trooculogram (EOG). Removal of artifacts from the EEG signal poses a great challenge in the

Available online xxx

medical field. Hence, the FLM (Firefly + Levenberg Marquardt) optimization-based learning algorithm for neural network-enhanced adaptive filtering model is introduced to eliminate

Keywords:

the artifacts from the EEG. Initially, the EEG signal was provided to the adaptive filter for

Electroencephalogram

yielding the optimal weights using the renowned optimization algorithms, called firefly

Levenberg Marquardt algorithm

algorithm and LM. These two algorithms are effectively hybridized and applied to the neural

Firefly algorithm

network to find the optimal weights for adaptive filtering. Then, the designed filtering

Artifacts signal

process renders an improved system for artifacts removal from the EEG signal. Finally, the

Signal to noise ratio

performance of the proposed model and the existing models regarding SNR, computation time, MSE and RMSE are analyzed. The results conclude that the proposed method achieves a high SNR of 42.042 dB. © 2017 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier B.V. All rights reserved.

11 9 10 12 13 14

1.

15 16 17 18 19 20

Recently, a number of EEG-based applications, such as wheelchair controllers and word speller programs [1–3] are used by the researchers. Using the non-invasive measurement of EEG, the activities of the brain can be observed by placing the electrodes on the scalp of the human brain in multiple areas [4]. The signal that results from recording the brain activities

Introduction

are not only the pure form of the brain signal because some defected signals, called artifacts (such as, power line noise, muscle contraction, heart activity and eye movement), also gets mixed with it [5,6]. Due to the existence of the artifact signal, the examination of EEG signal from the EEG recordings becomes more complex because it is perplexed with the neurological patterns. To generate a correct analysis and diagnosis [7], the unwanted signals must be eliminated from the recorded EEG signal.

* Corresponding author at: Department of Instrumentation Engineering, SGGSIE&T, Nanded (MS), India. E-mail address: [email protected] (M.H. Quazi). http://dx.doi.org/10.1016/j.bbe.2017.04.003 0208-5216/© 2017 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier B.V. All rights reserved. Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

21 22 23 24 25 26 27 28 29

BBE 197 1–11

2

biocybernetics and biomedical engineering xxx (2017) xxx–xxx

30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75

Once the artifacts from the EEG signal were removed, the original brain signal can be extracted from the EEG recordings [4]. Though the removal of artifacts from the EEG signal is a more complex process, it is utmost necessary for the progress of practical systems with low amplitude signal distortion [8]. Nowadays, more common methods are used to remove the artifacts from the EEG signals. Some of them includes linear filtering and regression model [9,10], ICA (independent component analysis) [11], WT (wavelet transform) [12–14] and (ANFIS) adaptive neuro-fuzzy inference system [7,15,16], adaptive filters and neural networks [3] and cascaded adaptive filters [17]. ICA is one of the important analyses for sorting out the EEG signals, which obtained from the electrodes that positioned on the scalp, into a self-governing mechanism. However, the ICA algorithm fails to yield better results during source separation. The artifacts signal was eliminated by the PCA technique, which is used to handle the high dimensional, boisterous and concurrent data. However, the orthogonal rotation gets reduced by the PCA analysis. Another technique is the eye-blink artifacts removal by the adaptive filter technique. Here, the adaptive filter subtracts the EEG source signal from the estimated inference signal to remove the artifacts [18]. Also, the adaptive filters remove the artifacts in real-time mode with less computation complexity. However, no complete assurance can be given that the reference signal is a perfect signal. Investigation on the fault classification performance by the fuzzy logic techniques has also been carried out [7]. The primary intention of this research is to design and develop a technique for removing the artifacts from the EEG signal. Here, we have planned to develop an FLM optimization-based learning algorithm for the neural network-enhanced adaptive filtering. This neural network-based adaptive filtering performs the removal of artifacts from the EEG signals. At first, the EEG signal is given to the proposed adaptive filtering to obtain the optimal weights using the well-known optimization algorithm, called firefly algorithm and LM (Levenberg Marquardt). These two algorithms were effectively hybridized and it is applied to the neural network to find the optimal weights for adaptive filtering. Then, the designed filtering was utilized for the removal of artifacts from the EEG signal. Finally, the comparison made against ICA, wavelet ICA, Fast ICA, wavelet-based method that is given in [19] and the LM based optimization, regarding SNR, percentage root mean square difference (PRD), and mean square error (MSE).

76

2.

77 78 79 80 81 82 83 84 85 86

While recording the EEG signal through the biomedical equipment, a variety of large signal contaminations or noise may affect the original signal data. The artifacts would dramatically alter the signal that was recorded at all scalp sites, especially at those sites that are closest to the source of the noise. Hence, the artifact or noise cancelation serves as a necessary stage of EEG processing [1,4,7,8,19–23]. One of the biggest challenges of using EEG is the very small signal-to-noise ratio of the brain signals that we are trying to observe, due to the coupling of the wide variety of noise

Problem formulation

sources [24]. Additionally, the EEG signals are of very small amplitudes, and because of that, they can be easily contaminated by noise. The main difficulty of the EEG artifact removal problem is the selection of the threshold level. The reason is that it should not remove the original EEG signal coefficients and at the same time, it should never keep the artifact signals as the original ones. In the case of the time varying signals that arise from the human body, adaptive filtering is found to serve as an appropriate method for the EEG artifact removal. Although the adaptive linear filters are the most widely used filters from among the various available adaptive filters, their performance is not satisfactory for dealing with the nonlinear problems. In [20], an adaptive FLN–RBFN-based filter was proposed to remove ocular, muscular and cardiac artifacts from the EEG signals. But, this adaptive neural network had not considered the weights optimally, when the learning process was performed. Hence, the finding of the optimal weights for adaptive filtering is a heuristic search problem, and it should be performed optimally by handling all the constraints to obtain a good artifacts removal performance.

87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109

3. Adaptive noise cancelation for artifacts removal

110 111

Basically, adaptive noise cancelation is used to eliminate the artifacts from the EEG signal. Fig. 1 shows the basic block diagram representation for achieving adaptive noise cancelation [20]. As shown in the figure, the adaptive noise cancelation process requires two inputs. The first input is generated from the EEG signal source and it is represented as S(t). Then, the second input is collected from the source of the artifact signal and it is denoted as A(t). The noise source considered here represents the origin, where the various artifacts such as, EOG, EMG and ECG generate. The noisy source signal of the artifacts signal gets passed through unidentified non-linear dynamics, resulting in the generation of the interference signal I(t). Then, the combination of both the interference signal and the clean signal generates the primary input signal and it is represented as follows:

112 113 114 115 116 117 118 119 120 121 122 123 124 125 126

(1)

127

where S(t) is defined as the input source signal and I(t) indicates the interference signal that is generated from the noise source. Meanwhile, the signal generated from the noise source is applied to an adaptive filtering process to get the filtered output. The filtered output is as close to the interference signal that is generated from the result of nonlinear dynamics. The aim of this adaptive noise cancelation is to retrieve the clean EEG signal. In order to retrieve the clean EEG signal, the filtered output is subtracted from the primary input and it is shown below:

128 129 130 131 132 133 134 135 136 137 138

So ðtÞ ¼ SðtÞ þ IðtÞFðtÞ

(2)

139

where F(t) denotes the filtered output and the output signal of the entire adaptive noise cancelation is represented as So(t).

140 141 142

PðtÞ ¼ SðtÞ þ IðtÞ

Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

BBE 197 1–11 biocybernetics and biomedical engineering xxx (2017) xxx–xxx

3

Fig. 1 – Basic block diagram for adaptive noise cancelation.

143

4.

NARX system model

144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161

The NARX neural network model, which is defined as the combination of the multilayer feed forward network, recurrent loop and time delay, is employed in this paper. This section shows the detailed description of the NARX neural network, which is used for modeling the nonlinear systems. Furthermore, the NARX network is more suitable for modeling the time series prediction analysis. Basically, the time series response is based on the successively correlated signal, wherein the signal value depends on the past values and the input signal. If the system input is found to be an assessable quantity, then that method can be considered as NARX [25]. The NARX system model contains three vector layers, namely, the input layer, the hidden layers and the output layer. The input layer contains three kinds of collected information vectors such as, exogenous input vector, delayed regressed output vector and delayed exogenous input vector. After performing the neural network operation, the output vector is produced as L(n + 1).

162

4.1.

163 164 165

Fig. 2 shows the internal architecture of the NARX neural network. The mathematical form of the NARX neural work is represented as follows:

166

Lðn þ 1Þ ¼ f ðLðnÞ; . . .; LðnDL Þ; VðnÞ; . . .; VðnDV ÞÞ

167 168 169 170 171 172 173 174 175

where L(n) is defined as the exogenous input vector. The delayed values of the regressed output vector are represented as L(n  1), L(n  2), . . ., L(n  DL). Then, the series of delayed exogenous input vector is denoted as V(n), V(n  1),   , V (n  DV). At the initial stage of the NARX neural network operation, the weights assign to the exogenous input vector and the hidden units as well as between the regressed output vector and the hidden units.

176 177 178 179

Internal architecture of the NARX neural network

(3)

4.2. LM-Firefly: proposed learning algorithm for the NARX model A hybrid algorithm, developed by the combination of both the Firefly [26] and Levenberg–Marquardt (LM) algorithm [13], is

used to improve the optimization performance in this paper. Initially, the firefly algorithm considers as the learning algorithm of the NARX neural network. The firefly algorithm depends on the blinking characteristics of the fireflies. The possible solution vector in the firefly algorithm derives from the position of every firefly. Later, the position of every firefly is changed based on the brightness. A firefly that does not find any brightness makes a random move. Here, the brightness value based on the performance that results from the optimization problem. The random movement of the fireflies increases the computation complexity, in addition to taking more time for converging at a solution. As a consequence, nonfeasible solutions and slower convergence result from the firefly algorithm. To overcome these drawbacks of the firefly algorithm, a faster convergence nonlinear optimization algorithm, called the LM algorithm, is used along with the firefly algorithm. The LM algorithm is used to find the location of the objects, which can be defined as the difference between the investigational values and the model equations. The LM algorithm is performed based on the combination of both the steepest descent and the Gauss-Newton method. However, the LM algorithm is used to perform a better-supervised learning with high speed. So, the NARX neural network imparts a faster training and more accuracy with the hybrid learning algorithm. Fig. 3 shows the proposed hybrid learning algorithm for the NARX neural network, in which three kinds of weights are combined using the learning algorithm. Here, the weight input vector is generated from the weight of the exogenous input, regressed output and the functional results of both the exogenous as well as the regressed output. At first, the randomly generated input vector is given to the input of the both the LM and the FF learning algorithm. After performing the optimization process, the weight vector is generated from both the Levenberg–Marquardt learning algorithm and the Firefly learning algorithm are WLM = Wl1 + Wl2 + Wl3 +    + Wlk and Wff = Wf1 + Wf2 + Wf3 +    + Wfk respectively. Then, both of these weight vectors are combined using the proposed hybrid learning algorithm, and a new weight vector Whb = Wz1 + Wz2 + Wz3 +    + Wzk is generated. The steps that are involved in the hybrid learning algorithm is discussed as follows:

Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223

BBE 197 1–11

4

biocybernetics and biomedical engineering xxx (2017) xxx–xxx

Fig. 2 – Internal architecture of the NARX neural network.

Fig. 3 – Proposed hybrid learning algorithm for the NARX neural network. Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

BBE 197 1–11 biocybernetics and biomedical engineering xxx (2017) xxx–xxx

224 225 226 227 228 229 230

Step 1: Initially, assign the size of the population with dimension d. Additionally, make a random initialization of the values of both the attractiveness and the intensity factor. Moreover, initialize the fireflies randomly based on the weight vector. The intensity of the Firefly is represented as follows: gd ¼

231 232 233 234 235

2

1 þ ld

where gd is defined as the intensity of the firefly at a distance d. Then, the attractiveness of the firefly can be represented as follows: ad ¼

236 237 238 239 240 241 242 243 244

g0

a0 1 þ ld2

Step 2: Evaluate the fitness function values. Step 3: Calculate the brightness of every Firefly and make a comparison between the fireflies based on the brightness value. Update the position of each firefly based on the brightness of the neighboring firefly. The weighted function used for the firefly algorithm is represented as follows:

245

where S(W) is defined as the value of the performance index, ET denotes the target outputs and E refers to the simulated output.

278 279 280

Step 10: Update the input weights based on the following equations of the LM algorithm.

281 282 283

1

DW ¼ ½GT G þ hi GT E where G is the Jacobian matrix, GT is the Jacobian transform matrix and h is the learning rate. The learning rate parameter is updated using the decay function v. As the value of S(W) increases, the learning rate gets multiplied by the decay function rate (v). Then, the value of S(W) is recomputed using the weighted function W + DW as the trail weight. Similarly, the learning rate is divided by the decay rate, whenever the function S(W) increases. Then, the incremented values of the weights is found by the formula,

284 285 286 287 288 289 290 291 292 293 294

W ¼ W þ DW Step 11: Update the weight as per the following equation, when the trail-weighed function is less than the performance index.

295 296 297 298 299 300

h ¼ hv 2

Wtþ1 ¼ Wt þ a0 eld ðWcb Wt Þ þ ge ff

246 247 248 249 250 251 252 253 254 255 256 257 258

5

where the absorption coefficient of the light is denoted as l and e is a random number. Further, the current best solution is represented by Wcb. Step 4: Once again, find the fitness value and the values of attractiveness as well as the intensity. Step 5: Rank the fireflies to find the current best solution or the global solution. Step 6: Continue this process, until the maximum number of iteration occurs. Once the firefly algorithm is applied, both the input vector and the updated weight are combined as follows:

259

Here, the learning rate is updated with the decay value by using the current learning rate. Then, return to Step 8 and continue the process.

301 302 303 304

Step 12: Measure the learning rate as per the following equation, when the performance index is less than the trail function.

305 306 307

h¼

308

h v

Step 13: Update the randomly generated weight vector using the LM formula, which is shown below:

309 310 311 312 313

LM Wtþ1 ¼ Wt ½H þ m  T1  Q

Pff ¼ Vðw; Wnew Þ 260 261 262

where w is defined as the input vector and Wnew is defined as the weighted vector is generated from the firefly algorithm.

where H is the Hessian matrix of the system and defined as the product of both the Jacobian matrix and the Jacobian transform matrix.

314 315 316 317

263 264

Step 7: Compute the error between the expected value and the current value, as shown below.

Step 14: Generate the gradient matrix from the Jacobian matrix and represent it as follows:

318 319

265

Q ¼ GT  E

n 1X ff Eff ¼ ðP Pgi Þ n i¼1 i ff

266 267 268

where Pgi denotes the original ground truth value and Pi defines the output of the firefly algorithm.

269 270 271

Step 8: Using the LM algorithm, initialize the weight vector based on the size of the hidden weights that are used in the NARX neural network, as shown below.

272 274 273 275 276

W ¼ Wa1 ; Wa2 ; . . .; Wak

320

where Q is the gradient matrix and is defined as the product of both the Jacobian transform matrix and the error value.

321 322 323

Step 15: Represent the output of the LM algorithm based on the original input vector weight and newly updated weight, which is expressed as follows:

324 325 326 327

PLM ¼ Vðw; Wnew Þ i

Step 9: Compute the sum of squared error based on the following equation.

Step 16: At the end of the LM algorithm, calculate the error value using the updated weight and the ground truth value as follows:

SðWÞ ¼ ET E

ELM ¼

277

n 1X ðPLM Pgi Þ n i¼1 i

Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

328 329 330 331 332 333

BBE 197 1–11

6

biocybernetics and biomedical engineering xxx (2017) xxx–xxx

334 335 336 337

where n is the total number of iterations, Pgi is the ground truth is the vector that is generated value of the input vector and PLM i from the output of the LM algorithm.

generating any interference signal such as EOG, ECG and EMG signal.

338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356

Step 17: Finally, generate the resultant weight vector based on the results of both the firefly algorithm and the LM algorithm. These two algorithms have enabled optimization in this paper. To perform the optimization process, the error value of both the LM as well as the FF algorithm is considered for the analysis. When the error value of the LM algorithm is lesser than the error value of the FF algorithm (ELM < EFF), the weight vector is considered from LM ). Similarly, when the result of the LM algorithm (Wtþ1 ¼ Wtþ1 the error value of the FF algorithm is lesser than the error value of the LM algorithm (ELM > EFF), the weight vector is FF ). As selected from the result of the FF algorithm (Wtþ1 ¼ Wtþ1 a consequence, the current error value becomes lesser than the previous error value (Et+1 < Et), and the value of the damping factor gets reduced. The value of damping factor increases when the current error is greater than the previous error (Et+1 > Et). Finally, the weight vector is generated through comparing the error value of the FLM learning algorithm.

5.

357 358

4.3. Multi channel artifacts removal using LMF based NARX model

359 360 361 362 363 364 365 366 367 368 369 370

Fig. 4 shows the block diagram representation of the NARX model that uses the MF-based learning algorithm to achieve multichannel artifacts removal. The electrical activity of the brain can be measured using the electroencephalogram method. In this method, 15 electrodes are placed on the scalp of several areas of the brain. The signal, generated from the multiple electrodes, is defined as the multi-channel signal. Then, the signal is generated from the multiple electrodes is passed through the NARX model using the proposed LMF algorithm. The proposed NARX model generates the artifacts removed signal or clean EEG signal for the corresponding channel or electrodes, without

Results and discussion

371 372 373

In this paper, the real signal is obtained from the physionet database, and the performance analysis is carried out. For demonstrating the proposed NARX model with the hybrid learning method, a real signal is contaminated with various artifact signal sources such as EOG, EMG, and ECG.

374 375 376 377 378

5.1.

379

Experimental setup

The proposed technique is implemented using MATLAB, and the experimentation conduct with the real-time signals that are available in physionet.

380 381 382

5.1.1.

383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403

Dataset description

The input signals are downloaded from the physionet by accessing the URL (https://physionet.org/cgi-bin/atm/ATM? database=ptbdb&tool=plot_waveforms). Here, the input database selected for the EEG is CHB-MIT scalp EEG database (chbmit). From this EEG database, the initial five signals are downloaded by giving the length as 1 min. For adding the ECG artifact, the ECG signal is downloaded from apnea-ECG database (apnea-ecg) for the length of 1 min. Similarly, the EMG signal is downloaded from the URL (https://physionet. org/physiobank/database/emgdb/emg_healthy.txt). The EOG signals are downloaded from the physionet by accessing the URL (https://physionet.org/cgi-bin/atm/ATM?database= ptbdb&tool=plot_waveforms). Here, the input database selected for the EOG is UCD Sleep Apnea database (ucddb) and the time length is 1 min. For adding the artifact signal such as, ECG, EOG and EMG with the EEG signal, the first channel of the signal is considered as artifact signal. Then, the artifact signal is added with all of the 15 channels of the five EEG signals taken for the experimentation. The physiologic signals are sampled at 256 samples per second with 16-bit resolution.

Fig. 4 – Block diagram representation to achieve multichannel artifacts removal using LMF-based NARX model. Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

BBE 197 1–11 biocybernetics and biomedical engineering xxx (2017) xxx–xxx

404

5.2.

405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429

Various methods such as independent component analysis, wavelet independent component analysis, fast independent component analysis, a neural network with LM, cascaded adaptive filter and the proposed NARX neural network are used for the extraction of the artifact signal from the input EEG signal. The independent component analysis is used to reveal the hidden signal or the artifact signal from the input EEG signal. Also, the maximally independent source signals can be statistically estimated using the independent component analysis approach. The estimation of an independent component analysis is performed one by one based on their negentropy, which is used to perform a faster computation than the independent component analysis. The wavelet decomposition algorithm is used to combine the wavelet transform with the independent component analysis, so as to overcome the disadvantages that the two techniques impart separately. Fig. 5 shows the waveform representation of the ECG artifact. The physiological artifacts such as ECG, EOC, and EMG arise from a variety of body activities. The artifacts signal of electrocardiography (ECG) caused by the heart beats, which generate the rhythmic activity in the EEG signal. Also, the responses of skin such as sweating may vary the electrodes' impedance and generate the ECG signal, in place of the expected EEG signal.

430

5.3.

Performance analysis

431 432 433 434 435 436 437 438 439 440 441

5.3.1.

Performance analysis based on signal to noise ratio

In this section, the performance of the proposed model is analyzed based on the signal to noise ratio, as shown in Fig. 6. Signal to noise ratio is one of the performance measurements, which is used to detect the noise level in the original signal. Fig. 6(a) shows the SNR-based analysis that depends on the ECG artifacts signal of the EEG signal. In the first signal source, the SNR value of the proposed model is 42.007 dB. This SNR value is very high, when compared to the SNR of other existing models like, ICA, WICA, FICA and cascaded adaptive filter that takes values as 5.4679 dB, 5.8425 dB, 10.433 dB and 39.215 dB,

Experimental results

7

respectively. For the second signal, the SNR value of the ICA and the WICA model are 3.7654 dB and 4.0605 dB respectively, and the SNR value of FICA is 8.2046 dB. The SNR value of cascaded adaptive filter is 34.299 dB. Here, the proposed hybrid optimization algorithm achieves the maximum SNR ratio of 38.359 dB. Once again, the SNR performance of the various models analyzes with the third signal source. Here, the SNR value attained by ICA and WICA are 5.4679 dB and 5.8425 dB. The SNR value of FICA is 9.1664 dB, which is greater than the values of both the ICA and the WICA models and the SNR value of cascaded adaptive filter is 33.712 dB. Fig. 6(b) shows the signal to noise ratio analysis of the EMG artifacts signal that is presented in the EEG. By analyzing Fig. 6 (b), the SNR values of the various artifacts removal models obtained. For the first signal, the SNR value of the ICA and the WICA is 5.4348 dB and 5.5671 dB, respectively. The SNR value obtained by the cascaded adaptive filter is 39.34 dB while the SNR value of the proposed model is 42.042 dB. For the second EEG signal applied to the EMG artifacts removal process, both the NN_LM and the proposed model have achieved the same SNR value as 38.397 dB, and the SNR value of the cascaded adaptive filter is 35.124 dB. Finally, we observe that the proposed model could effectively remove the EMG signal artifact of EEG using the FLM learning algorithm. Fig. 6(c) shows the Signal to Noise Ratio analysis of the EEG signal with EOG artifacts. For the first signal, the SNR values are 5.7113 dB, 5.7118 dB, 6.002 dB, and 39.293 dB for the artifacts removal models, such as ICA, WICA, FICA, and cascaded adaptive filter respectively. On the other hand, SNR value obtained by the NN_LM and NN_FLM model is 42.042 dB. From Fig. 6(c), we understand that the proposed method reaches the higher SNR value, which achieves better performance of artifact removal.

442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474

5.3.2. value

475 476 477 478 479 480 481

Performance analysis based on mean square error

Fig. 7(a) shows the performance analysis of the proposed method, involving the ECG artifact signal, using mean square error. The mean square error (MSE) of the WICA and NN_LM model is very high when compared to the proposed model. The proposed model achieves as minimum mean square error

Fig. 5 – ECG artifact cancelation process. Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

BBE 197 1–11

8

biocybernetics and biomedical engineering xxx (2017) xxx–xxx

Fig. 6 – Signal to noise ratio analysis of EEG signal with various artifacts.

482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508

value of 4882.6. Meanwhile, the maximum mean square error that is obtained by the ICA model is 6915.9. Minimum value of the mean square error results in a better performance of the artifacts removal process. The proposed optimizationbased learning algorithm also achieves the minimum mean square error for removing the artifacts from the input EEG signal. Fig. 7(b) shows the EMG artifacts signal removal process based on the mean square error performance. Initially, the MSE value of the various models such as ICA, WICA, FICA, NN_LM, cascaded adaptive filter and the proposed NN_FLM model are analyzed from the various signal source. When analyzing the first signal source, the proposed NN_FLM-based method achieves the minimum MSE value of 4310. Other existing models such as ICA, WICA, FICA, NN_LM, and cascaded adaptive filter have obtained the maximum MSE value as 5532.4, 5507.9, 5506.3, 4356.8, and 5156.1, respectively. Fig. 7(c) shows the performance analysis based on the MSE for the EOG signal removal from EEG. Here, the signals from a total of five patients considered. By analyzing the first patient's brain signal, the MSE of models like ICA, WICA, and FICA are orderly obtained as 5508.2, 5502.4, and 5497.2. A lesser MSE produces better performance. Furthermore, the NN_LM and cascaded adaptive filter models have achieved the MSE value as 4667 and 5218.8. However, the proposed NN_FLM method has obtained the minimum MSE of 4369. From the results above, we observed that the proposed model achieves

minimum MSE value to improve the performance of the artifacts removal process.

509 510

5.3.3. Performance analysis based on root mean square error (RMSE) value

511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534

Fig. 8 shows the RMSE analysis of the EEG signal with various artifacts such as ECG, EMG, and EOG. Fig. 8(a) shows the RMSE analysis to assess the performance of the proposed model, involving the ECG artifact signal. Usually, the minimum value of the RMSE generates a better performance. The five signal sources are taken from the physionet database. The RMSE attained by the models like, FICA, NN_LM, and the cascaded adaptive filter are 63.982, 0.26736 and 19.613 respectively whereas, the value of RMSE obtained by the proposed model is 0.23922. From the figure, the least RMSE value is obtained by the proposed model, when compared to the other existing models such as FICA and NN_LM and cascaded adaptive filter. Fig. 8(b) shows the performance analysis of the EMG signal artifact removal process based on the RMSE value. For the first signal source, the proposed model obtains the RMSE value of 0.22156, which is very less than the other existing models. The RMSE value of NN_LM model is 0.2232 and it nearly equals the RMSE of the proposed method. However, the proposed model obtains the least RMSE value. From the results, we understand that the proposed model only renders a better artifacts removal performance, when compared to the other existing models.

Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

BBE 197 1–11 biocybernetics and biomedical engineering xxx (2017) xxx–xxx

Fig. 7 – Mean square error analysis.

Fig. 8 – RMSE analysis. Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

9

BBE 197 1–11

10

biocybernetics and biomedical engineering xxx (2017) xxx–xxx

Table 1 – Comparative discussion of the proposed model with the existing models such as ICA, WICA, FICA, NN_LM, and cascaded adaptive filter.

ICA [11] WICA FICA NN_LM Cascaded adaptive filter [17] NN_FLM (proposed)

MSE

RMSE

SNR

5508.2 5502.4 5492.6 4356.8 5156.1 4310

65.194 64.765 63.982 0.2232 19.587 0.22156

5.7113 5.8425 10.433 42.042 39.34 42.042

535 536 537 538 539 540 541 542 543 544 545 546

Fig. 8(c) shows the performance analysis based on RMSE for the EOG artifact signal removal process. For the fifth signal source, the RMSE value of the various artifacts removal models such as ICA, WICA, FICA, NN_LM, cascaded adaptive filter and the proposed NN_FLM model are 85.709, 85.048, 85.038, 0.5889, 21.849, and 0.58697 respectively. From Fig. 8(c), the artifact removal models such as FICA, NN_LM and the proposed NN_FLM model obtained the RMSE value of 82.501, 0.3737, and 0.37008 respectively for the third signal source. Finally, we conclude that the proposed model of NN_FLM has obtained the lowest RMSE value that improves the performance of the artifact removal process.

547

5.4.

548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568

Table 1 shows comparative discussion the of the proposed NN_FLM model with the existing models, such as ICA, WICA, FICA, NN_LM, and cascaded adaptive filter based on MSE, RMSE, and SNR. From the table, the value of MSE attained by the existing models, such as ICA, WICA, FICA, NN_LM, and cascaded adaptive filter are 5508.2, 5502.4, 5492.6, 4356.8, and 5156.1recpectively while, the proposed model achieves the MSE value of 4310 which is smaller than the existing models. The value of RMSE obtained by the existing models such as ICA, WICA, FICA, NN_LM, and cascaded adaptive filter are 65.194, 64.765, 63.982, 0.2232, and 19.587 respectively. The proposed model achieves the RMSE value of 0.22156 which is also smaller than the existing models. The minimum value of MSE and RMSE generates the better performance for removing the artifacts. The value of SNR obtained by the existing models such as ICA, WICA, FICA, NN_LM, and cascaded adaptive filter are 5.7113 dB, 5.8425 dB, 10.433 dB, 42.042 dB, and 39.34 dB respectively whereas, the proposed model attains the higher SNR value of 42.042 dB. From the table, it is clear that the proposed model achieves the better performance while compared with the existing models.

569

6.

570 571 572 573 574 575 576 577

In this paper, a hybrid optimization algorithm for the neural network-enhanced adaptive filtering has been proposed to remove the artifacts signal from the multi-channel EEG data. The proposed hybrid optimization algorithm has been used to find the weights of the NARX neural network, and it has been defined as the combination of both the LM as well as the FF algorithm, so as to overcome the individual shortcomings of the two algorithms. The proposed model of NN_FLM has been

Comparative discussion

Conclusion

implemented using MATLAB, and the experimentation has been conducted using the real-time signals that are available in physionet. Here, 15 channel data are considered and the artifacts like EOG, EMG, and ECG signals include for the analysis. A comparison has been made subsequently using the independent component analysis, wavelet-independent component analysis, fast independent component analysis and neural network-based LM algorithm, regarding signal to noise ratio, root mean square error, mean square error and computational time. From the results, we conclude that the proposed FLM optimization-based learning algorithm attains a better signal to noise ratio of 42.042 dB.

578 579 580 581 582 583 584 585 586 587 588 589

references

590

[1] Jirayucharoensak S, Israsena P. Automatic removal of EEG artifacts using ICA and lifting wavelet transform. Proceedings of International Conference on Computer Science and Engineering; 2013. p. 136–9. [2] Niedermeyer E, da Silva FL. Electroencephalography: basic principles, clinical applications, and related fields. Lippincott Williams & Wilkins; 2005. [3] Nguyen H-AT, Musson J, Li F, Wang W, Zhang G, Xu R, et al. EOG artifact removal using a wavelet neural network. Neurocomputing 2012;97:374–89. [4] Turnip A, Junaidi E. Removal artifacts from EEG signal using independent component analysis and principal component analysis. Proceedings of International Conference on Technology, Informatics, Management, Engineering & Environment; 2014. p. 19–21. [5] Vazquez RR, Perez HV, Ranta R, Louis VD, Maquin D, Maillard L. Blind source separation, wavelet de-noising and discriminant analysis for EEG artifacts and noise cancelling. Biomed Signal Process Control 2012;7:389–400. [6] Geetha G, Geethalakshmi SN. Artifact removal from EEG using spatially constrained independent component analysis and wavelet de-noising with Otsu's thresholding technique. Procedia Eng 2012;30:1064–71. [7] FemilinSheniha S, SujaPriyadharsini S, Edward Rajan S. Removal of artifact from EEG signal using differential evolution algorithm. Proceedings of International Conference on Communication and Signal Processing; 2013. p. 3–5. [8] Gorecka J, Walerjan P. Artifacts extraction from EEG data using the infomax approach. Biocybern Biomed Eng 2011;31 (4):59–74. [9] Correa G. Artifact removal from EEG signals using adaptive filters in cascade, vol. 90. IOP Publishing Ltd.; 2007. p. 1–10. [10] Croft RJ, Barry RJ. Removal of ocular artifact from the EEG: a review. Neurophysiol Clin/Clin Neurophysiol 2000;30 (1):5–19. [11] Li Y. Automatic removal of the eye blink artifact from EEG using an ICA-based template matching approach. Physiol Meas 2006;27:425–36. [12] Senthilkumar P. Removal of ocular artifacts in the EEG through wavelet transform without using an EOG reference channel. Int J Open Probl Comput Sci 2008;1 (3):188–200. [13] Gaidhane VH, Singh V, Hote YV, Kumar M. New approaches for image compression using neural network. Intell Learn Syst Appl 2011;3:220–9. [14] Krishnaveni V. Automatic identification and Removal of ocular artifacts from EEG using wavelet transform. Meas Sci Rev 2006;6(4):45–57.

591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640

Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003

BBE 197 1–11 biocybernetics and biomedical engineering xxx (2017) xxx–xxx

641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660

[15] Suja Priyadharsini S, Edward Rajan S. An efficient softcomputing technique for extraction of EEG signal from tainted EEG signal. Appl Soft Comput 2012;12:1131–7. [16] Liu T, Yao D. Removal of the ocular artifacts from EEG data using a cascaded spatio-temporal processing. Comput Methods Prog Biomed 2006;83:95–103. [17] Correa AG, Laciar E, Patiño HD, Valentinuzzi ME. Artifact removal from EEG signals using adaptive filters in cascade. J Phys Conf Ser 2007;90(1):1–10. [18] Hsiao-Lung C, Yu-Tai T, Ling-Fu M, Tony W. The removal of ocular artifacts from EEG signals using adaptive filters based on ocular source components. Ann Biomed Eng 2010;38(11). [19] Akhtar MT, Mitsuhasini W, James CJ. Employing spatially constrained ICA and wavelet de-noising, for automatic removal of artifacts from multichannel EEG data. Signal Process 2012;92:401–16. [20] Jafarifarmand A, Badamchizadeh MA. Artifacts removal in EEG signal using a new neural network enhanced adaptive filter. Neurocomputing 2013;103:222–31.

11

[21] Zou Y, Nathan V, Jafari R. Automatic identification of artifact-related independent components for artifact removal in EEG recordings. Biomed Health Inform 2014;20 (1):73–81. [22] Kim M-K, Kim S-P. Artifact removal from EEG signals using the total variation method. Proceedings of International Conference on ASCC; 2015. p. 1–4. [23] Turnip A. Automatic artifacts removal of EEG signals using robust principal component analysis. Proceedings of International Conference on Technology, Informatics, Management, Engineering & Environment; 2014. p. 19–21. [24] Guruva Reddy A, Narava S. Artifact removal from EEG signals. Comput Appl 2013;77(13). [25] Menezes Jr JMP, Barreto GA. Long-term time series prediction with the NARX network: an empirical evaluation. Neuro Comput 2008;71:3335–43. [26] Subhani Shareef SK, Rasul Mohideen E, Ali L. Directed Firefly algorithm for multimodal problems. Proceedings of International Conference on Computational Intelligence and Computing Research; 2015.

661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681

Please cite this article in press as: Quazi MH, Kahalekar SG. Artifacts removal from EEG signal: FLM optimization-based learning algorithm for neural network-enhanced adaptive filtering. Biocybern Biomed Eng (2017), http://dx.doi.org/10.1016/j.bbe.2017.04.003