Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Classification of Lactate Level Using Resting-State EEG Measurements

Classification of Lactate Level Using Resting-State EEG Measurements Hindawi Applied Bionics and Biomechanics Volume 2021, Article ID 6662074, 8 pages https://doi.org/10.1155/2021/6662074 Research Article Classification of Lactate Level Using Resting-State EEG Measurements 1,2 3 4 Saad Abdulazeez Shaban , Osman Nuri Ucan , and Adil Deniz Duru Computer Science Department, College of Education for Pure Sciences, Diyala University, Diyala 32001, Iraq Electrical and Computer Engineering, School of Engineering and Natural Sciences, Altınbaş University, Istanbul 34217, Turkey Engineering Faculty, Electrical and Electronics Department, Istanbul University, 34850 Avcilar, Istanbul, Turkey Neuroscience and Psychology Research in Sports Lab, Faculty of Sport Science, Marmara University, 34668 Istanbul, Turkey Correspondence should be addressed to Saad Abdulazeez Shaban; saad.shaban@uodiyala.edu.iq Received 9 November 2020; Revised 1 January 2021; Accepted 15 January 2021; Published 8 February 2021 Academic Editor: Mohammed Yahya Alzahrani Copyright © 2021 Saad Abdulazeez Shaban et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The electroencephalography (EEG) signals have been used widely for studying the brain neural information dynamics and behaviors along with the developing impact of using the machine and deep learning techniques. This work proposes a system based on the fast Fourier transform (FFT) as a feature extraction method for the classification of human brain resting-state electroencephalography (EEG) recorded signals. In the proposed system, the FFT method is applied on the resting-state EEG recordings and the corresponding band powers were calculated. The extracted relative power features are supplied to the classification methods (classifiers) as an input for the classification purpose as a measure of human tiredness through predicting lactate enzyme level, high or low. To validate the suggested method, we used an EEG dataset which has been recorded from a group of elite-level athletes consisting of two classes: not tired, the EEG signals were recorded during the resting-state task before performing acute exercise and tired, the EEG signals were recorded in the resting-state after performing an acute exercise. The performance of three different classifiers was evaluated with two performance measures, accuracy and precision values. The accuracy was achieved above 98% by the K-nearest neighbor (KNN) classifier. The findings of this study indicated that the feature extraction scheme has the ability to classify the analyzed EEG signals accurately and predict the level of lactate enzyme high or low. Many studying fields, like the Internet of Things (IoT) and the brain computer interface (BCI), can utilize the findings of the proposed system in many crucial decision-making applications. techniques can play a vital role in analyzing and investigating 1. Introduction EEG signals and exploitation results in different applications Electroencephalography is the brain neural signals which like diagnosing human mental diseases [5], predicting emo- reflect the brain’s electrical potentials and are mainly used tional states, decision discovery for patient rehabilitation for studying brain neural information dynamics processing devices, or assistive technology for interactive input devices and employed to diagnose brain disturbances [1]. Normally, like gaming controllers and wheelchairs drivers [6]. Many those signals are time series signals [2] recorded by means of of the previous studies were investigating different brain a specialized skull helmet which contains multiple electrodes activity by means of magnetic resonance imaging (MRI) that distributed and attached to a specific position on the scalp, measures anatomical images noninvasively. Resting-state either in a wet or dry manner [3]. The data acquired from functional MRI activity is shown to differ between before and after performing aerobic exercise [7]. A recent study the recording of such signals are in a very large amount, and that is why they should be analyzed by specialized investigates the impact of a single acute exercise session on methods, rather than conventional visual ways [4]. Among the brain’s functional connectivity and showed an obvious those methods, data classification using machine learning increase in the functional connectivity of sensorimotor brain 2 Applied Bionics and Biomechanics networks that could be clearly assessed using functional MRI Input EEG recordings, before [8]. Even though the MR imaging shows high spatial resolu- and aer ex ft ercise tion, but because of the blood oxygenated level dependency attributes (BOLD), it shows a limited temporal resolution of the measured signals. Using the electroencephalography signals as a measure of electrophysiological brain activity Preprocessing EEG data produces enhancement in temporal resolution to a range of milliseconds [9]. The use of EEG as a technique of ana- lyzing in the field of psychology has been applied for a long time in many studies, but it was not that common in the Spectral analysis field of exercises and sports until recent times. A study investigated the impact of severe physical short-term exer- cise and long-term workout training on the EEG resting- state alpha frequency (iAPF) of the individual shows that Classification of EEG data frequency has increased after performing intense exercise [10], while another study shows an increment in the power of the frontal area observed in the EEG signals after perform- ing acute cycling exercises [11]. The high-intense running Predicting lactate enzyme high or low was found to have an effect on both EEG and the mood of the exerciser [12]. Brain activation during aerobic exercises was found to be increased, i.e., the EEG beta frequency band power is increased and the alpha frequency band is decreased Validating results with lactate during performing moderate-intensity short-time cycle measurements ergometer exercise, then returning to the power baseline after finishing [13]. Figure 1: The proposed classification model. One of the emerging aspects in this field is predicting the lactate dehydrogenase enzyme levels, whether being high or low, in the human blood by classifying the collected model performance. These features are analyzed along with the relation to blood lactate levels before and after perform- resting-state EEG data [14] from a subject along with mea- suring lactate levels in the blood. The idea behind that is to ing exercises. To the best of the author knowledge, until study whether lactate levels in the human blood could be pre- now, no one study in the literature is related to the assess- dicted of being increased or decreased affected by performing ment of classification performance using power spectral an acute exercise, as it was reported that the blood lactate density- (PSD-) based feature extraction machine learning classifiers when applied to the fatigue problem after acute level would reach its peak after maximal treadmill running exercise was made for a short period [15], meanwhile collect- exercise. Compared with several studies, experimental results ing EEG data before and after exercise and mark them as clarify that the suggested system could enhance the detection class 1 (before exercise or lactate-level-low) and class 2 (after rate. Figure 1 shows the schematics of the proposed system. exercise or lactate-level-high). This study is aimed at examin- ing this idea by suggesting a classification system that should 2. Materials and Methods discriminate two states of lactate level using EEG signals hav- ing different frequency bands recorded from a group of 2.1. Operational Tasks. In this study, the employed dataset healthy athlete subjects of the elite level, before and after per- that includes the resting-state EEG signals from [9], has forming a single bout of acute exercise. As EEG signals have been used. The proposed system in our study consists of different features that could be extracted using a variety of two main parts: one involves feeding lactate enzyme level methods [16], we should dominate the best discriminant fea- test measurements, and the other involves input the EEG ture that gives us the best classification score in terms of signal recordings and manipulating them. EEG data was accuracy. Among different features, the band power features recorded from a volunteers’ group of elite level athletes which represent the energy (power) of EEG signals were cho- (no:of subjects = 10), and all are representing members of sen to represent a discriminant criterion and are computed official karate team. These subjects had performed the blood by means of power spectral density of each EEG signal fre- lactate level test before doing the exercise, and the results quency band for a given channel. Frequency band power is were allocated to represent low-level lactate (not tired) class. regarded as a gold standard feature to be applied in applica- Initial lactate measurements were found to be at the baseline tions like brain computer interface (BCI) by many studies value of around 2 millimoles/litre. In the first step, subjects [6, 17]. Band power features are calculated to evaluate the were sitting in a calm fashion with eyes closing (EC) condi- brain’s activity changes over a given time window (typically tion and asked to stay as-calm-as possible and thinking about of a few seconds) encountered by performing an acute exer- nothing for 3 minutes. Meanwhile, the EEG signals are being cise session. Then, the extracted feature data is arranged in collected from subjects with a sampling rate of 1000 Hz using a vector, manipulated and modified using preprocessing BrainAmp ExG amplifier from 16 channels (Fz, Fp1, Fp2, F3, techniques to clean data from artifacts and enhance the F4, Cz, C3, C4, T7, T8, Pz, P3, P4, Oz, O1, and O2) with dry Applied Bionics and Biomechanics 3 Fp1 Fp2 F3 F4 Fz T7 C3 Cz C4 T8 Pz P3 P4 O1 O2 Oz Figure 2: Cap 16-electrodes layout. electrode caps. Figure 2 shows the distribution of 16 removing epochs that have an absolute amplitude greater than electrodes over the brain scalp. 100 μV by using band-pass filtering technique. Figure 3 The next step in this phase requires each subject to sepa- shows epochs of 1 second EEG signal of one subject recorded rately perform an acute exercise of a short-time shuttle run from 16 channels before and after performing the exercise. with 20 meters for each shuttle. This running protocol is an 2.2. Feature Extraction. The EEG signals are nonstationary incrementally progressive test that is used to predict personal time-series signals, and once the raw version of EEG data physical sensations experienced during exercises like maxi- was recorded then passed the preprocessing step, the next mum oxygen consumption, increased heart rate, muscle step is to get related attributes through the feature extraction fatigue, and increased sweating. It consists of 20 m running process. To get the better distinguishing feature from EEG that requires increasing running pace while time decreases signals, we have applied the fast Fourier transform (FFT) as levels proceed with a beep stimulus between levels. method to provide frequency representation of the signals, While performing the exercise, the performance is mon- which helps to measure the power spectrum of data for each itored using rated perceived exertion (RPE) scale. RPE is a frequency band, delta (0-4 Hz), theta (4-8), alpha (8-13), beta scale for measuring physical activity intensity by asking the (13-30), and gamma (30-45), within a time window or epoch. activist about how he feels his body is working without inter- For the frequency spectral analysis, the nonstationarity can rupting the exercise. The exercise ended when each subject be tolerated and the EEG signal assumed stationary for the reports a 16 RPE level according to the Borg rating of epoch period. Fast Fourier transform (FFT) is a signal pro- perceived exertion [18]. cessing method which is used to transform the signal from The next experimental phase starts, after a short resting its time domain to the equivalent frequency domain repre- period of one minute, by measuring blood lactate levels for sentation by dividing the signal function into a continuous each subject, and it was found at a high level of around 16 frequency band known as frequency spectrum [20]. If FðkÞ millimoles/litre. Then, the lactate test is repeated 4 times with is the fast Fourier transform of a function f ðzÞ, then it is 2 minutes between them, and each test result found to be at defined by using equation (1) as follows: the same high levels with no drop to baseline within the EEG data measurement phase. Then, the EEG data measur- kz kz Fk = 〠 ing was repeated with EC condition for 3 minutes and ðÞ fz ðÞw + 〠 fz ðÞw , ð1Þ Z Z z even z odd assigned to be a high-level lactate (tired) class. Both datasets of measured EEG signal, pre, and postexercise contain arti- facts generated by some muscular movement, eye blinking, for k =0,1, : ⋯ :, Z − 1, where FðkÞ is the Fourier coefficient and heartbeats that can contaminate the quality of EEG data of f ðzÞ, which is assumed to have a complex value, z even and [19], and that is why the data has been cleaned from noises by z odd correspond to the EEG samples of f ðzÞ, which were 4 Applied Bionics and Biomechanics Pre exercise first second EEG –50 –100 0 200 400 600 800 1000 Time (ms) Post exercise first second EEG –20 –40 0 200 400 600 800 1000 Time (ms) Figure 3: Pre and post exercise 16-channel EEG data. even and odd numbered, respectively. w = exp ð−2zj/ZÞ, information gain are not. Gini index measures the impurity where z is equal to 3.14, and j is the imaginary part. of data from a set of training tuples, as in equation (2). 2.3. Classification. The process of EEG data formatting and performing frequency band power calculation is done by a Gini data =1 − 〠 Pi , ð2Þ ðÞ self-developed MATLAB routine, and the resulting data, i=1 pre, and postexercise are then combined and formed a single bulk dataset of size 7759 rows and 80 columns and fed to a set where Pi represents the probability that a tuple in data of classification algorithms (classifiers) with 80 features and belongs to a specific class, say Ci. Information gain is an attri- two class labels, “1” is representing lactate-level-low before bute selection measure that tries to find the attribute which exercise, and “2” is representing lactate-level-high after exer- has the highest information gain that minimizes the required cise. Among those applied classifiers, the KNN, decision tree information to classify tuple and is defined by equation (3). (DT), and logistic regression (LR) have reported scoring better than others, like the linear discriminant analysis (LDA) classi- fier and support vector machine (SVM) classifier; thus, only InfðÞ data = −〠 Pi log ðÞ Pi , ð3Þ the highest-scoring classifiers have been listed in Results. i=1 2.3.1. Decision Tree. A decision tree is a machine learning where Pi is the probability that a tuple in data belongs to a model in which each nonleaf node denotes a test on a feature, specific class Ci. each branch node represents an outcome of the test, and each terminal node holds a class label. The root node is the top- 2.3.2. K-Nearest Neighbor. The K-nearest neighbor (KNN) most node. Assume an X tuple with an unknown class label, classifiers learn by comparing a given test vector with similar the feature values of X are tested against the tree with path training vectors (SRs). The training vectors are described by traced along from the root node to a leaf node, which repre- number of n attributes. When given an unknown vector, a sents the class prediction for the given tuple [21]. Each deci- K-nearest neighbor classifier seeks the pattern field for the k sion tree employs an attribute selection method that specifies training vectors (nearest neighbors) that are closest to the a procedure for choosing the best attribute that discriminates unknown vector. The “closeness” is defined in terms of a the tuple depending on the class. This procedure uses attri- distance, such as Manhattan distance, which defines distance bute selection measure as a metric function to evaluate split d between x1 and x2 vectors, as in equation (4): for feature selection such as the information gain and the Gini index. Some attribute selection measures impose the djðÞ , l =jj xj1 − xl1 +jj xj2 − xl2 +jj xj3 − xl3 : ð4Þ tree to become binary like the Gini index; others like Micro (V) Micro (V) Applied Bionics and Biomechanics 5 Table 1: Classification score of applied models. Table 2: Sensitivity and specificity of the classifiers. Model KNN Decision tree Logistic regression Model KNN Decision tree Logistic regression Accuracy 98.4% 98.0% 70.0% Sensitivity 98.8% 97.2% 76.5% Specificity 97.0% 94.2% 64.0% Then, the probability is used as a measure to assign the input x to the most probable class (nearest one), as in equation (5): recognition (i.e., the proportion of the second-class belong- ing tuples that are correctly identified) [24]. These two ðÞ i Py = j ∣ X = x = −〠Ly = j , ð5Þ ðÞ measures are defined as follows in equations (6) and (7), i∊B respectively: where B represents the set of K neighbors of the training TP vector which are nearest to input x, and LðxÞ represents ð7Þ Sensitivity = , indicator that acts as a function which sets to 1 if the input TP + FP x is true and set to 0 if not. TN ð8Þ 2.3.3. Logistic Regression. Logistic regression is a popular Specificity = , TN + FN model for solving classification problems, and the term “Logistic” comes from the underlying Logit function used where TP represents the positive tuples of data that were in this model for classification, the natural logarithm of odds correctly classified by the model, whereas FP represents the ratio [22]. Logistic regression estimates as probability the positive tuples that were falsely classified. On the other hand, impacts of independent variables on the outcome variables. TN represents the negative tuples of data that were correctly Simple logistic model is shown in equation (5). classified by the model. In contrast, FN represents the nega- tive tuples that were falsely classified by the model. Table 2 shows those measures for each of the applied classification Logit y = naturallog odds =ln , ð6Þ ðÞ ðÞ 1 − y models. Thus, we note that the KNN and decision tree classi- fiers have a high accuracy along with high sensitivity and where the logitðyÞ is representing the probabilities from 0 specificity which indicates their ability to correctly classify to 1. both the positive and negative tuples, which are in contrast The technology used for classification was the Classifica- to the logistic regression classifier that showed a moderate tion Learner applications available in the MATLAB R2018a sensitivity and specificity scores meaning that it can recog- software. nize positive and negative tuples at a lower rate. To validate entire input data, we used the technique of K- Furthermore, classifiers show the following precision fold cross-validation, which splits data into K folds (parts). values, which represent percentage of instances labelled as Among these K folds, K-1 folds are used to train proposed positive and are actually such, for both classes, lactate-level- model and the remaining fold is used for testing purpose. low and lactate-level-high denoting pre-exercise (not tired) The procedure is replicated for K times until all subsets are and post-exercise (tired) tiredness recognition, respectively, validated; then, all the results are averaged for final accuracy for different classifiers as in Table 3a, b, and c as follows: prediction [23]. 4. Discussion 3. Results In the present study, we had investigated the ability to predict The extraction of EEG band power feature yields a significant whether the lactate level is low or high in the human body enhancement in the classifier’s accuracy scores, especially for using EEG signals of subjects after performing an acute exer- KNN, decision tree and logistic regression classifiers. The cise. The subjects were athletes of the elite level from the main finding of our study is proving the ability to clearly pre- national team of Turkey. The achieved results indicate that dict human blood lactate levels using resting-state EEG sig- predicting blood lactate levels, high or low, using electroen- nals when applying suitable techniques, power spectral cephalogram brain data can be done accurately in terms of density in our case. The classification score versus applied classification scores when implemented for healthy athlete method results are listed in the Table 1. To the best of the who endures a single bout of acute exercises. The discrimina- authors’ knowledge, there is no study in the literature related tion ability is driven by the changes encountered in the band to the classification performance measure using FFT and power values of EEG signal bands after doing an exercise machine learning classifiers, investigating the fatigue prob- [25]. This hypothesis was proven by variations that occurred lem after acute exercise. with alpha and beta frequency band power that investigated Another measure is to calculate the specificity and sensi- after implementing a maximal effort exercise and shows an tivity of the classifier. Sensitivity is also referred to as the rate increment in beta absolute power in a group of electrodes of true positive recognition (i.e., the proportion of the first- [26]. In our study, the best scoring classification model was class belonging tuples that are correctly identified); on the KNN with 98.4% accuracy with a ratio of training data and other hand, specificity represents the rate of true negative testing data 80 : 20, which was found to be a high scoring 6 Applied Bionics and Biomechanics Table 3: Precision values for different classifiers. (a) Precision values for decision tree Total 3867 3893 7760 Lactate-level-low 3790 76 3866 Actual class Lactate-level-high 77 3817 3894 Lactate-level-low Lactate-level-high Total Predicted class (b) Precision values for KNN Total 3870 3890 7760 Lactate-level-low 3804 62 3866 Actual class Lactate-level-high 66 3828 3894 Lactate-level-low Lactate-level-high Total Predicted class (c) Precision values for logistic regression Total 4359 3401 7760 Lactate-level-low 2959 907 3866 Actual class Lactate-level-high 1400 2494 3894 Lactate-level-low Lactate-level-high Total Predicted class Table 4: Score results of various objective studies. Study Year Electrode no. Study objective Applied algorithms Accuracy % Classify facial movement and expressions [29] 2016 14 KNN 98.0 by a noninvasive EEG signals Analyze EEG signals from different cognitive [30] 2018 1 LDA, SVM, KNN 95.0, 100.0, 100.0 states to control BCI devices EEG signal classification recorded while ANN, SVM, MLP, 88.89, 98.75 98.57, [33] 2014 128 doing a complex cognitive task KNN, Naïve Bayes 98.21 83.57 Monitoring EEG signal while driving to [34] 2016 12 SVM 91.28 detect the driver fatigue state EEG signal features extraction and [35] 2019 19 SVM 90.0 classification during fatigue exercise Classify resting-state EEG to predict Present study 2020 16 KNN, DT, LR 98.4, 98.0, 70.0 lactate, high or low model in other studies, especially when applied against appli- work which investigated the effect of a single bout of acute cations with a low-dimensional [27] feature vector of data exercise, the effect of increasing running exercise intensities [28]. The KNN was found to perform effectively to extract on spontaneous EEG was investigated by a study, which and classify feature vector for different facial movements found that the overall spectrum power in EEG significantly and expressions measured by noninvasive EEG devices. The increased in all frequency bands with increasing intensities accuracy was around 98% driven by implementing segmen- of exercise, lactate level has increased, and even after a period of 15- to 30-minute recovery, lactate enzyme level has tation to the complete signal waveform [29]. Even though classification could be applied using other EEG features like decreased but still significantly higher than baseline and dis- the average spectral centroid, average standard deviation, or cernible [31]. The subsequent decrease in spectrum power average energy entropy, but still the power spectral density was seen in a subset of frequency bands in some cortical offers the highest accuracy with all classifiers and was found regions suggesting a decrease in cortical activation after exercise intensities, as a hypothesis of brainstem inhibitory to score 100% with KNN when analyzing EEG signals from different human cognitive states employed to control brain mechanism, may occur [32]. Table 4 shows the results of computer interface (BCI) devices [30]. In contrast to our various objective studies comparable to our work. Applied Bionics and Biomechanics 7 [6] F. Lotte, L. Bougrain, A. Cichocki et al., “A review of classifica- 5. Conclusions tion algorithms for EEG-based brain–computer interfaces: a The proposed work represents the use of band power spectral 10 year update,” Journal of Neural Engineering, vol. 15, no. 3, density along with machine learning techniques for classifi- article 031005, 2018. cation and analysis of EEG signals recorded during resting- [7] T. B. Weng, G. L. Pierce, W. G. Darling, D. Falk, V. A. Mag- state tasks. The band’s power feature of EEG signals was notta, and M. W. Voss, “The acute effects of aerobic exercise extracted using FFT for all of the 16 channels of each sub- on the functional connectivity of human brain networks,” Brain Plasticity, vol. 2, no. 2, pp. 171–190, 2017. ject’s EEG recording. Three different classification models (KNN, decision tree, and logistic regression) were applied, [8] A. S. Rajab, D. E. Crane, L. E. Middleton, A. D. Robertson, M. Hampson, and B. J. MacIntosh, “A single session of exercise and their performance was reported. The classification accu- increases connectivity in sensorimotor-related brain networks: racy of KNN and decision tree found to be above 98%. This a resting-state fMRI study in young healthy adults,” Frontiers makes the study the unique and pioneer one to discuss and in Human Neuroscience, vol. 8, p. 625, 2014. prove the ability to use resting-state EEG signals as an accu- [9] A. D. Duru, T. H. Balcıoğlu, C. E. Ö. Çakır, and D. G. Duru, rate measure for the human tiredness level through predict- “Acute changes in electrophysiological brain dynamics in elite ing lactate enzyme level high or low. The band power was karate players,” Iranian Journal of Science and Technology, found to be a very useful EEG feature to classify these signals Transactions of Electrical Engineering, vol. 44, pp. 565–579, after performing acute exercise sessions. Hence, the proposed feature extraction and classification system have the signifi- [10] B. Gutmann, A. Mierau, T. Hülsdünker et al., “Effects of Phys- cance to be applied on real-time EEG applications like BCI, ical Exercise on Individual Resting State EEG Alpha Peak Fre- IoT, military, or medical applications to predict the individ- quency,” Neural Plasticity, vol. 2015, Article ID 717312, 6 ual physical tiredness state that can assist in many crucial sit- pages, 2015. uations. As a suggested study expansion, the classifiers could [11] H. Enders, F. Cortese, C. Maurer, J. Baltich, A. B. Protzner, and be applied to EEG data collected for each subject individually B. M. Nigg, “Changes in cortical activity measured with EEG with applying the same previous procedures, and the results during a high-intensity cycling exercise,” Journal of Neuro- could be compared in both cases. This may be implemented physiology, vol. 115, no. 1, pp. 379–388, 2016. in future work possibly with applying more algorithms and [12] S. Schneider, C. D. Askew, J. Diehl et al., “EEG activity and preprocessing techniques for the purpose of achieving higher mood in health orientated runners after different exercise classification accuracy scores. intensities,” Physiology and Behavior, vol. 96, no. 4-5, pp. 709–716, 2009. Data Availability [13] K. A. Kubitz and A. A. Mott, “EEG power spectral densities during and after cycle ergometer exercise,” Research Quarterly The EEG dataset which was used in the present study is avail- for Exercise and Sport, vol. 67, no. 1, pp. 91–96, 1996. able from the author, Adil Deniz Duru, through a reasonable [14] A. M. A. Mohamed, O. N. Uçan, O. Bayat, and A. D. Duru, request. “Classification of resting-state status based on sample entropy and power spectrum of electroencephalography (EEG),” Applied Bionics and Biomechanics, vol. 2020, Article ID Conflicts of Interest 8853238, 10 pages, 2020. No conflict of interest is declared by the authors regarding [15] N. Fujitsuka, T. Yamamoto, T. Ohkuwa, M. Saito, and M. Miyamura, “Peak blood lactate after short periods of max- this paper publication. imal treadmill running,” European Journal of Applied Physiol- ogy, vol. 48, no. 3, pp. 289–296, 1982. References [16] A. S. Al-Fahoum and A. A. Al-Fraihat, “Methods of EEG signal features extraction using linear analysis in frequency and time- [1] S. Sanei and J. A. Chambers, EEG Signal Processing, 312, Wiley, frequency domains,” ISRN Neuroscience, vol. 2014, Article ID England, 2007. 730218, 7 pages, 2014. [2] T. H. H. Aldhyani and M. R. Joshi, “Integration of time series [17] D. Trad, T. Al-Ani, and M. Jemni, “Motor imagery signal clas- models with soft clustering to enhance network traffic fore- sification for BCI system using empirical mode décomposition casting,” in 2016 Second International Conference on Research and bandpower feature extraction,” Brain, vol. 7, no. 2, pp. 5– in Computational Intelligence and Communication Networks 16, 2016. (ICRCICN), pp. 212–214, Kolkata, India, September 2016. [18] G. A. V. Borg, “Psychophysical bases of perceived exertion,” [3] L.-W. Ko, C. Yang, P.-L. Wu et al., “Development of a smart Medicine and Science in Sports and Exercise, vol. 14, no. 5, helmet for strategical BCI applications,” Sensors, vol. 19, pp. 377–381, 1982. no. 8, article 1867, 2019. [19] S. Romeroa, M. A. Mañanasa, and M. J. Barbanoj, “A compar- [4] R. Vigario, J. Sarela, V. Jousmiki, M. Hamalainen, and E. Oja, ative study of automatic techniques for ocular artifact reduc- “Independent component approach to the analysis of EEG and tion in spontaneous EEG signals based on clinical target MEG recordings,” IEEE Transactions on Biomedical Engineer- variables: a simulation case,” Computers in Biology and Medi- ing, vol. 47, no. 5, pp. 589–593, 2000. cine, vol. 38, no. 3, pp. 348–360, 2008. [5] T. H. Aldhyani, A. S. Alshebami, and M. Y. Alzahrani, “Soft computing model to predict chronic diseases,” Journal of [20] M. R. Canal, “Comparison of wavelet and short time Fourier Information Science and Engineering, vol. 36, no. 2, pp. 365– transform methods in the analysis of EMG signals,” Journal 376, 2020. of Medical Systems, vol. 34, no. 1, pp. 91–94, 2010. 8 Applied Bionics and Biomechanics [21] J. R. Quinlan, “Induction of decision trees,” Machine Learning, vol. 1, no. 1, pp. 81–106, 1986. [22] C.-Y. J. Peng, K. L. Lee, and G. M. Ingersoll, “An introduction to logistic regression analysis and reporting,” The Journal of Educational Research, vol. 96, no. 1, pp. 3–14, 2010. [23] T. Fushiki, “Estimation of prediction error by using K-fold cross-validation,” Statistics and Computing, vol. 21, no. 2, pp. 137–146, 2011. [24] J. Han, M. Kamber, and J. Pei, Data Mining Concepts and Techniques 3'rd ed., 703, Morgan Kaufmann, 2011. [25] A. A. Abdul-latif, I. Cosic, D. K. Kumar, B. Polus, and C. Da Costa, “Power changes of EEG signals associated with muscle fatigue: the root mean square analysis of EEG bands,” in Pro- ceedings of the 2004 Intelligent Sensors, Sensor Networks and Information Processing Conference, pp. 531–534, Melbourne, VC, Australia, December 2004. [26] H. Moraes, C. Ferreira, A. Deslandes et al., “Beta and alpha electroencephalographic activity changes after acute exercise,” Arquivos de Neuro-Psiquiatria, vol. 65, no. 3-A, pp. 637–641, [27] H. I. Alsaadi, R. M. Almuttairi, O. Bayat, and O. N. Ucan, “Computational intelligence algorithms to handle dimension- ality reduction for enhancing intrusion detection system,” Journal of Information Science and Engineering, vol. 36, pp. 293–308, 2020. [28] J. F. Borisoff, S. G. Mason, A. Bashashati, and G. E. Birch, “Brain–computer interface design for asynchronous control applications: improvements to the LF-ASD asynchronous brain switch,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 6, pp. 985–992, 2004. [29] U. I. Awan, U. H. Rajput, G. Syed, R. Iqbal, I. Sabat, and M. Mansoor, “Effective classification of EEG signals using K- nearest neighbor algorithm,” in 2016 International Conference on Frontiers of Information Technology (FIT), pp. 120–124, Islamabad, Pakistan, December 2016. [30] M. Rashid, N. Sulaiman, M. Mustafa, S. Khatun, and B. Bifta, fication “The Classi of EEG Signal Using Different Machine Learning Techniques for BCI Application,” in Robot Intelli- gence Technology and Applications. RiTA 2018. Communica- tions in Computer and Information Science, vol 1015,J.H. Kim, H. Myung, and S. M. Lee, Eds., Springer, Singapore, 2018. [31] D. Mechau, S. Mücke, H. Liesen, and M. Weiß, “Effect of increasing running velocity on electroencephalogram in a field test,” European Journal of Applied Physiology, vol. 78, no. 4, pp. 340–345, 1998. [32] B. D. Hatfield, “Exercise and mental health: the mechanisms of exercise-induced psychological states,” in Psychology of sports, exercise and fitness, L. Diamant, Ed., pp. 17–50, Hemisphere Publishing, Washington DC, USA, 1991. [33] H. U. Amin, A. S. Malik, R. F. Ahmad et al., “Feature extrac- tion and classification for EEG signals using wavelet transform and machine learning techniques,” Australasian Physical & Engineering Sciences in Medicine, vol. 38, no. 1, pp. 139–149, [34] Y. Xiong, J. Gao, Y. Yang, X. Yu, and W. Huang, “Classifying driving fatigue based on combined entropy measure using EEG signals,” International Journal of Control and Automa- tion, vol. 9, no. 3, pp. 329–338, 2016. [35] Z. Yang and H. Ren, “Feature extraction and simulation of EEG signals during exercise-induced fatigue,” IEEE Access, vol. 7, pp. 46389–46398, 2019. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Applied Bionics and Biomechanics Hindawi Publishing Corporation

Classification of Lactate Level Using Resting-State EEG Measurements

Loading next page...
 
/lp/hindawi-publishing-corporation/classification-of-lactate-level-using-resting-state-eeg-measurements-N9idPb0wRb
Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2021 Saad Abdulazeez Shaban et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ISSN
1176-2322
eISSN
1754-2103
DOI
10.1155/2021/6662074
Publisher site
See Article on Publisher Site

Abstract

Hindawi Applied Bionics and Biomechanics Volume 2021, Article ID 6662074, 8 pages https://doi.org/10.1155/2021/6662074 Research Article Classification of Lactate Level Using Resting-State EEG Measurements 1,2 3 4 Saad Abdulazeez Shaban , Osman Nuri Ucan , and Adil Deniz Duru Computer Science Department, College of Education for Pure Sciences, Diyala University, Diyala 32001, Iraq Electrical and Computer Engineering, School of Engineering and Natural Sciences, Altınbaş University, Istanbul 34217, Turkey Engineering Faculty, Electrical and Electronics Department, Istanbul University, 34850 Avcilar, Istanbul, Turkey Neuroscience and Psychology Research in Sports Lab, Faculty of Sport Science, Marmara University, 34668 Istanbul, Turkey Correspondence should be addressed to Saad Abdulazeez Shaban; saad.shaban@uodiyala.edu.iq Received 9 November 2020; Revised 1 January 2021; Accepted 15 January 2021; Published 8 February 2021 Academic Editor: Mohammed Yahya Alzahrani Copyright © 2021 Saad Abdulazeez Shaban et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The electroencephalography (EEG) signals have been used widely for studying the brain neural information dynamics and behaviors along with the developing impact of using the machine and deep learning techniques. This work proposes a system based on the fast Fourier transform (FFT) as a feature extraction method for the classification of human brain resting-state electroencephalography (EEG) recorded signals. In the proposed system, the FFT method is applied on the resting-state EEG recordings and the corresponding band powers were calculated. The extracted relative power features are supplied to the classification methods (classifiers) as an input for the classification purpose as a measure of human tiredness through predicting lactate enzyme level, high or low. To validate the suggested method, we used an EEG dataset which has been recorded from a group of elite-level athletes consisting of two classes: not tired, the EEG signals were recorded during the resting-state task before performing acute exercise and tired, the EEG signals were recorded in the resting-state after performing an acute exercise. The performance of three different classifiers was evaluated with two performance measures, accuracy and precision values. The accuracy was achieved above 98% by the K-nearest neighbor (KNN) classifier. The findings of this study indicated that the feature extraction scheme has the ability to classify the analyzed EEG signals accurately and predict the level of lactate enzyme high or low. Many studying fields, like the Internet of Things (IoT) and the brain computer interface (BCI), can utilize the findings of the proposed system in many crucial decision-making applications. techniques can play a vital role in analyzing and investigating 1. Introduction EEG signals and exploitation results in different applications Electroencephalography is the brain neural signals which like diagnosing human mental diseases [5], predicting emo- reflect the brain’s electrical potentials and are mainly used tional states, decision discovery for patient rehabilitation for studying brain neural information dynamics processing devices, or assistive technology for interactive input devices and employed to diagnose brain disturbances [1]. Normally, like gaming controllers and wheelchairs drivers [6]. Many those signals are time series signals [2] recorded by means of of the previous studies were investigating different brain a specialized skull helmet which contains multiple electrodes activity by means of magnetic resonance imaging (MRI) that distributed and attached to a specific position on the scalp, measures anatomical images noninvasively. Resting-state either in a wet or dry manner [3]. The data acquired from functional MRI activity is shown to differ between before and after performing aerobic exercise [7]. A recent study the recording of such signals are in a very large amount, and that is why they should be analyzed by specialized investigates the impact of a single acute exercise session on methods, rather than conventional visual ways [4]. Among the brain’s functional connectivity and showed an obvious those methods, data classification using machine learning increase in the functional connectivity of sensorimotor brain 2 Applied Bionics and Biomechanics networks that could be clearly assessed using functional MRI Input EEG recordings, before [8]. Even though the MR imaging shows high spatial resolu- and aer ex ft ercise tion, but because of the blood oxygenated level dependency attributes (BOLD), it shows a limited temporal resolution of the measured signals. Using the electroencephalography signals as a measure of electrophysiological brain activity Preprocessing EEG data produces enhancement in temporal resolution to a range of milliseconds [9]. The use of EEG as a technique of ana- lyzing in the field of psychology has been applied for a long time in many studies, but it was not that common in the Spectral analysis field of exercises and sports until recent times. A study investigated the impact of severe physical short-term exer- cise and long-term workout training on the EEG resting- state alpha frequency (iAPF) of the individual shows that Classification of EEG data frequency has increased after performing intense exercise [10], while another study shows an increment in the power of the frontal area observed in the EEG signals after perform- ing acute cycling exercises [11]. The high-intense running Predicting lactate enzyme high or low was found to have an effect on both EEG and the mood of the exerciser [12]. Brain activation during aerobic exercises was found to be increased, i.e., the EEG beta frequency band power is increased and the alpha frequency band is decreased Validating results with lactate during performing moderate-intensity short-time cycle measurements ergometer exercise, then returning to the power baseline after finishing [13]. Figure 1: The proposed classification model. One of the emerging aspects in this field is predicting the lactate dehydrogenase enzyme levels, whether being high or low, in the human blood by classifying the collected model performance. These features are analyzed along with the relation to blood lactate levels before and after perform- resting-state EEG data [14] from a subject along with mea- suring lactate levels in the blood. The idea behind that is to ing exercises. To the best of the author knowledge, until study whether lactate levels in the human blood could be pre- now, no one study in the literature is related to the assess- dicted of being increased or decreased affected by performing ment of classification performance using power spectral an acute exercise, as it was reported that the blood lactate density- (PSD-) based feature extraction machine learning classifiers when applied to the fatigue problem after acute level would reach its peak after maximal treadmill running exercise was made for a short period [15], meanwhile collect- exercise. Compared with several studies, experimental results ing EEG data before and after exercise and mark them as clarify that the suggested system could enhance the detection class 1 (before exercise or lactate-level-low) and class 2 (after rate. Figure 1 shows the schematics of the proposed system. exercise or lactate-level-high). This study is aimed at examin- ing this idea by suggesting a classification system that should 2. Materials and Methods discriminate two states of lactate level using EEG signals hav- ing different frequency bands recorded from a group of 2.1. Operational Tasks. In this study, the employed dataset healthy athlete subjects of the elite level, before and after per- that includes the resting-state EEG signals from [9], has forming a single bout of acute exercise. As EEG signals have been used. The proposed system in our study consists of different features that could be extracted using a variety of two main parts: one involves feeding lactate enzyme level methods [16], we should dominate the best discriminant fea- test measurements, and the other involves input the EEG ture that gives us the best classification score in terms of signal recordings and manipulating them. EEG data was accuracy. Among different features, the band power features recorded from a volunteers’ group of elite level athletes which represent the energy (power) of EEG signals were cho- (no:of subjects = 10), and all are representing members of sen to represent a discriminant criterion and are computed official karate team. These subjects had performed the blood by means of power spectral density of each EEG signal fre- lactate level test before doing the exercise, and the results quency band for a given channel. Frequency band power is were allocated to represent low-level lactate (not tired) class. regarded as a gold standard feature to be applied in applica- Initial lactate measurements were found to be at the baseline tions like brain computer interface (BCI) by many studies value of around 2 millimoles/litre. In the first step, subjects [6, 17]. Band power features are calculated to evaluate the were sitting in a calm fashion with eyes closing (EC) condi- brain’s activity changes over a given time window (typically tion and asked to stay as-calm-as possible and thinking about of a few seconds) encountered by performing an acute exer- nothing for 3 minutes. Meanwhile, the EEG signals are being cise session. Then, the extracted feature data is arranged in collected from subjects with a sampling rate of 1000 Hz using a vector, manipulated and modified using preprocessing BrainAmp ExG amplifier from 16 channels (Fz, Fp1, Fp2, F3, techniques to clean data from artifacts and enhance the F4, Cz, C3, C4, T7, T8, Pz, P3, P4, Oz, O1, and O2) with dry Applied Bionics and Biomechanics 3 Fp1 Fp2 F3 F4 Fz T7 C3 Cz C4 T8 Pz P3 P4 O1 O2 Oz Figure 2: Cap 16-electrodes layout. electrode caps. Figure 2 shows the distribution of 16 removing epochs that have an absolute amplitude greater than electrodes over the brain scalp. 100 μV by using band-pass filtering technique. Figure 3 The next step in this phase requires each subject to sepa- shows epochs of 1 second EEG signal of one subject recorded rately perform an acute exercise of a short-time shuttle run from 16 channels before and after performing the exercise. with 20 meters for each shuttle. This running protocol is an 2.2. Feature Extraction. The EEG signals are nonstationary incrementally progressive test that is used to predict personal time-series signals, and once the raw version of EEG data physical sensations experienced during exercises like maxi- was recorded then passed the preprocessing step, the next mum oxygen consumption, increased heart rate, muscle step is to get related attributes through the feature extraction fatigue, and increased sweating. It consists of 20 m running process. To get the better distinguishing feature from EEG that requires increasing running pace while time decreases signals, we have applied the fast Fourier transform (FFT) as levels proceed with a beep stimulus between levels. method to provide frequency representation of the signals, While performing the exercise, the performance is mon- which helps to measure the power spectrum of data for each itored using rated perceived exertion (RPE) scale. RPE is a frequency band, delta (0-4 Hz), theta (4-8), alpha (8-13), beta scale for measuring physical activity intensity by asking the (13-30), and gamma (30-45), within a time window or epoch. activist about how he feels his body is working without inter- For the frequency spectral analysis, the nonstationarity can rupting the exercise. The exercise ended when each subject be tolerated and the EEG signal assumed stationary for the reports a 16 RPE level according to the Borg rating of epoch period. Fast Fourier transform (FFT) is a signal pro- perceived exertion [18]. cessing method which is used to transform the signal from The next experimental phase starts, after a short resting its time domain to the equivalent frequency domain repre- period of one minute, by measuring blood lactate levels for sentation by dividing the signal function into a continuous each subject, and it was found at a high level of around 16 frequency band known as frequency spectrum [20]. If FðkÞ millimoles/litre. Then, the lactate test is repeated 4 times with is the fast Fourier transform of a function f ðzÞ, then it is 2 minutes between them, and each test result found to be at defined by using equation (1) as follows: the same high levels with no drop to baseline within the EEG data measurement phase. Then, the EEG data measur- kz kz Fk = 〠 ing was repeated with EC condition for 3 minutes and ðÞ fz ðÞw + 〠 fz ðÞw , ð1Þ Z Z z even z odd assigned to be a high-level lactate (tired) class. Both datasets of measured EEG signal, pre, and postexercise contain arti- facts generated by some muscular movement, eye blinking, for k =0,1, : ⋯ :, Z − 1, where FðkÞ is the Fourier coefficient and heartbeats that can contaminate the quality of EEG data of f ðzÞ, which is assumed to have a complex value, z even and [19], and that is why the data has been cleaned from noises by z odd correspond to the EEG samples of f ðzÞ, which were 4 Applied Bionics and Biomechanics Pre exercise first second EEG –50 –100 0 200 400 600 800 1000 Time (ms) Post exercise first second EEG –20 –40 0 200 400 600 800 1000 Time (ms) Figure 3: Pre and post exercise 16-channel EEG data. even and odd numbered, respectively. w = exp ð−2zj/ZÞ, information gain are not. Gini index measures the impurity where z is equal to 3.14, and j is the imaginary part. of data from a set of training tuples, as in equation (2). 2.3. Classification. The process of EEG data formatting and performing frequency band power calculation is done by a Gini data =1 − 〠 Pi , ð2Þ ðÞ self-developed MATLAB routine, and the resulting data, i=1 pre, and postexercise are then combined and formed a single bulk dataset of size 7759 rows and 80 columns and fed to a set where Pi represents the probability that a tuple in data of classification algorithms (classifiers) with 80 features and belongs to a specific class, say Ci. Information gain is an attri- two class labels, “1” is representing lactate-level-low before bute selection measure that tries to find the attribute which exercise, and “2” is representing lactate-level-high after exer- has the highest information gain that minimizes the required cise. Among those applied classifiers, the KNN, decision tree information to classify tuple and is defined by equation (3). (DT), and logistic regression (LR) have reported scoring better than others, like the linear discriminant analysis (LDA) classi- fier and support vector machine (SVM) classifier; thus, only InfðÞ data = −〠 Pi log ðÞ Pi , ð3Þ the highest-scoring classifiers have been listed in Results. i=1 2.3.1. Decision Tree. A decision tree is a machine learning where Pi is the probability that a tuple in data belongs to a model in which each nonleaf node denotes a test on a feature, specific class Ci. each branch node represents an outcome of the test, and each terminal node holds a class label. The root node is the top- 2.3.2. K-Nearest Neighbor. The K-nearest neighbor (KNN) most node. Assume an X tuple with an unknown class label, classifiers learn by comparing a given test vector with similar the feature values of X are tested against the tree with path training vectors (SRs). The training vectors are described by traced along from the root node to a leaf node, which repre- number of n attributes. When given an unknown vector, a sents the class prediction for the given tuple [21]. Each deci- K-nearest neighbor classifier seeks the pattern field for the k sion tree employs an attribute selection method that specifies training vectors (nearest neighbors) that are closest to the a procedure for choosing the best attribute that discriminates unknown vector. The “closeness” is defined in terms of a the tuple depending on the class. This procedure uses attri- distance, such as Manhattan distance, which defines distance bute selection measure as a metric function to evaluate split d between x1 and x2 vectors, as in equation (4): for feature selection such as the information gain and the Gini index. Some attribute selection measures impose the djðÞ , l =jj xj1 − xl1 +jj xj2 − xl2 +jj xj3 − xl3 : ð4Þ tree to become binary like the Gini index; others like Micro (V) Micro (V) Applied Bionics and Biomechanics 5 Table 1: Classification score of applied models. Table 2: Sensitivity and specificity of the classifiers. Model KNN Decision tree Logistic regression Model KNN Decision tree Logistic regression Accuracy 98.4% 98.0% 70.0% Sensitivity 98.8% 97.2% 76.5% Specificity 97.0% 94.2% 64.0% Then, the probability is used as a measure to assign the input x to the most probable class (nearest one), as in equation (5): recognition (i.e., the proportion of the second-class belong- ing tuples that are correctly identified) [24]. These two ðÞ i Py = j ∣ X = x = −〠Ly = j , ð5Þ ðÞ measures are defined as follows in equations (6) and (7), i∊B respectively: where B represents the set of K neighbors of the training TP vector which are nearest to input x, and LðxÞ represents ð7Þ Sensitivity = , indicator that acts as a function which sets to 1 if the input TP + FP x is true and set to 0 if not. TN ð8Þ 2.3.3. Logistic Regression. Logistic regression is a popular Specificity = , TN + FN model for solving classification problems, and the term “Logistic” comes from the underlying Logit function used where TP represents the positive tuples of data that were in this model for classification, the natural logarithm of odds correctly classified by the model, whereas FP represents the ratio [22]. Logistic regression estimates as probability the positive tuples that were falsely classified. On the other hand, impacts of independent variables on the outcome variables. TN represents the negative tuples of data that were correctly Simple logistic model is shown in equation (5). classified by the model. In contrast, FN represents the nega- tive tuples that were falsely classified by the model. Table 2 shows those measures for each of the applied classification Logit y = naturallog odds =ln , ð6Þ ðÞ ðÞ 1 − y models. Thus, we note that the KNN and decision tree classi- fiers have a high accuracy along with high sensitivity and where the logitðyÞ is representing the probabilities from 0 specificity which indicates their ability to correctly classify to 1. both the positive and negative tuples, which are in contrast The technology used for classification was the Classifica- to the logistic regression classifier that showed a moderate tion Learner applications available in the MATLAB R2018a sensitivity and specificity scores meaning that it can recog- software. nize positive and negative tuples at a lower rate. To validate entire input data, we used the technique of K- Furthermore, classifiers show the following precision fold cross-validation, which splits data into K folds (parts). values, which represent percentage of instances labelled as Among these K folds, K-1 folds are used to train proposed positive and are actually such, for both classes, lactate-level- model and the remaining fold is used for testing purpose. low and lactate-level-high denoting pre-exercise (not tired) The procedure is replicated for K times until all subsets are and post-exercise (tired) tiredness recognition, respectively, validated; then, all the results are averaged for final accuracy for different classifiers as in Table 3a, b, and c as follows: prediction [23]. 4. Discussion 3. Results In the present study, we had investigated the ability to predict The extraction of EEG band power feature yields a significant whether the lactate level is low or high in the human body enhancement in the classifier’s accuracy scores, especially for using EEG signals of subjects after performing an acute exer- KNN, decision tree and logistic regression classifiers. The cise. The subjects were athletes of the elite level from the main finding of our study is proving the ability to clearly pre- national team of Turkey. The achieved results indicate that dict human blood lactate levels using resting-state EEG sig- predicting blood lactate levels, high or low, using electroen- nals when applying suitable techniques, power spectral cephalogram brain data can be done accurately in terms of density in our case. The classification score versus applied classification scores when implemented for healthy athlete method results are listed in the Table 1. To the best of the who endures a single bout of acute exercises. The discrimina- authors’ knowledge, there is no study in the literature related tion ability is driven by the changes encountered in the band to the classification performance measure using FFT and power values of EEG signal bands after doing an exercise machine learning classifiers, investigating the fatigue prob- [25]. This hypothesis was proven by variations that occurred lem after acute exercise. with alpha and beta frequency band power that investigated Another measure is to calculate the specificity and sensi- after implementing a maximal effort exercise and shows an tivity of the classifier. Sensitivity is also referred to as the rate increment in beta absolute power in a group of electrodes of true positive recognition (i.e., the proportion of the first- [26]. In our study, the best scoring classification model was class belonging tuples that are correctly identified); on the KNN with 98.4% accuracy with a ratio of training data and other hand, specificity represents the rate of true negative testing data 80 : 20, which was found to be a high scoring 6 Applied Bionics and Biomechanics Table 3: Precision values for different classifiers. (a) Precision values for decision tree Total 3867 3893 7760 Lactate-level-low 3790 76 3866 Actual class Lactate-level-high 77 3817 3894 Lactate-level-low Lactate-level-high Total Predicted class (b) Precision values for KNN Total 3870 3890 7760 Lactate-level-low 3804 62 3866 Actual class Lactate-level-high 66 3828 3894 Lactate-level-low Lactate-level-high Total Predicted class (c) Precision values for logistic regression Total 4359 3401 7760 Lactate-level-low 2959 907 3866 Actual class Lactate-level-high 1400 2494 3894 Lactate-level-low Lactate-level-high Total Predicted class Table 4: Score results of various objective studies. Study Year Electrode no. Study objective Applied algorithms Accuracy % Classify facial movement and expressions [29] 2016 14 KNN 98.0 by a noninvasive EEG signals Analyze EEG signals from different cognitive [30] 2018 1 LDA, SVM, KNN 95.0, 100.0, 100.0 states to control BCI devices EEG signal classification recorded while ANN, SVM, MLP, 88.89, 98.75 98.57, [33] 2014 128 doing a complex cognitive task KNN, Naïve Bayes 98.21 83.57 Monitoring EEG signal while driving to [34] 2016 12 SVM 91.28 detect the driver fatigue state EEG signal features extraction and [35] 2019 19 SVM 90.0 classification during fatigue exercise Classify resting-state EEG to predict Present study 2020 16 KNN, DT, LR 98.4, 98.0, 70.0 lactate, high or low model in other studies, especially when applied against appli- work which investigated the effect of a single bout of acute cations with a low-dimensional [27] feature vector of data exercise, the effect of increasing running exercise intensities [28]. The KNN was found to perform effectively to extract on spontaneous EEG was investigated by a study, which and classify feature vector for different facial movements found that the overall spectrum power in EEG significantly and expressions measured by noninvasive EEG devices. The increased in all frequency bands with increasing intensities accuracy was around 98% driven by implementing segmen- of exercise, lactate level has increased, and even after a period of 15- to 30-minute recovery, lactate enzyme level has tation to the complete signal waveform [29]. Even though classification could be applied using other EEG features like decreased but still significantly higher than baseline and dis- the average spectral centroid, average standard deviation, or cernible [31]. The subsequent decrease in spectrum power average energy entropy, but still the power spectral density was seen in a subset of frequency bands in some cortical offers the highest accuracy with all classifiers and was found regions suggesting a decrease in cortical activation after exercise intensities, as a hypothesis of brainstem inhibitory to score 100% with KNN when analyzing EEG signals from different human cognitive states employed to control brain mechanism, may occur [32]. Table 4 shows the results of computer interface (BCI) devices [30]. In contrast to our various objective studies comparable to our work. Applied Bionics and Biomechanics 7 [6] F. Lotte, L. Bougrain, A. Cichocki et al., “A review of classifica- 5. Conclusions tion algorithms for EEG-based brain–computer interfaces: a The proposed work represents the use of band power spectral 10 year update,” Journal of Neural Engineering, vol. 15, no. 3, density along with machine learning techniques for classifi- article 031005, 2018. cation and analysis of EEG signals recorded during resting- [7] T. B. Weng, G. L. Pierce, W. G. Darling, D. Falk, V. A. Mag- state tasks. The band’s power feature of EEG signals was notta, and M. W. Voss, “The acute effects of aerobic exercise extracted using FFT for all of the 16 channels of each sub- on the functional connectivity of human brain networks,” Brain Plasticity, vol. 2, no. 2, pp. 171–190, 2017. ject’s EEG recording. Three different classification models (KNN, decision tree, and logistic regression) were applied, [8] A. S. Rajab, D. E. Crane, L. E. Middleton, A. D. Robertson, M. Hampson, and B. J. MacIntosh, “A single session of exercise and their performance was reported. The classification accu- increases connectivity in sensorimotor-related brain networks: racy of KNN and decision tree found to be above 98%. This a resting-state fMRI study in young healthy adults,” Frontiers makes the study the unique and pioneer one to discuss and in Human Neuroscience, vol. 8, p. 625, 2014. prove the ability to use resting-state EEG signals as an accu- [9] A. D. Duru, T. H. Balcıoğlu, C. E. Ö. Çakır, and D. G. Duru, rate measure for the human tiredness level through predict- “Acute changes in electrophysiological brain dynamics in elite ing lactate enzyme level high or low. The band power was karate players,” Iranian Journal of Science and Technology, found to be a very useful EEG feature to classify these signals Transactions of Electrical Engineering, vol. 44, pp. 565–579, after performing acute exercise sessions. Hence, the proposed feature extraction and classification system have the signifi- [10] B. Gutmann, A. Mierau, T. Hülsdünker et al., “Effects of Phys- cance to be applied on real-time EEG applications like BCI, ical Exercise on Individual Resting State EEG Alpha Peak Fre- IoT, military, or medical applications to predict the individ- quency,” Neural Plasticity, vol. 2015, Article ID 717312, 6 ual physical tiredness state that can assist in many crucial sit- pages, 2015. uations. As a suggested study expansion, the classifiers could [11] H. Enders, F. Cortese, C. Maurer, J. Baltich, A. B. Protzner, and be applied to EEG data collected for each subject individually B. M. Nigg, “Changes in cortical activity measured with EEG with applying the same previous procedures, and the results during a high-intensity cycling exercise,” Journal of Neuro- could be compared in both cases. This may be implemented physiology, vol. 115, no. 1, pp. 379–388, 2016. in future work possibly with applying more algorithms and [12] S. Schneider, C. D. Askew, J. Diehl et al., “EEG activity and preprocessing techniques for the purpose of achieving higher mood in health orientated runners after different exercise classification accuracy scores. intensities,” Physiology and Behavior, vol. 96, no. 4-5, pp. 709–716, 2009. Data Availability [13] K. A. Kubitz and A. A. Mott, “EEG power spectral densities during and after cycle ergometer exercise,” Research Quarterly The EEG dataset which was used in the present study is avail- for Exercise and Sport, vol. 67, no. 1, pp. 91–96, 1996. able from the author, Adil Deniz Duru, through a reasonable [14] A. M. A. Mohamed, O. N. Uçan, O. Bayat, and A. D. Duru, request. “Classification of resting-state status based on sample entropy and power spectrum of electroencephalography (EEG),” Applied Bionics and Biomechanics, vol. 2020, Article ID Conflicts of Interest 8853238, 10 pages, 2020. No conflict of interest is declared by the authors regarding [15] N. Fujitsuka, T. Yamamoto, T. Ohkuwa, M. Saito, and M. Miyamura, “Peak blood lactate after short periods of max- this paper publication. imal treadmill running,” European Journal of Applied Physiol- ogy, vol. 48, no. 3, pp. 289–296, 1982. References [16] A. S. Al-Fahoum and A. A. Al-Fraihat, “Methods of EEG signal features extraction using linear analysis in frequency and time- [1] S. Sanei and J. A. Chambers, EEG Signal Processing, 312, Wiley, frequency domains,” ISRN Neuroscience, vol. 2014, Article ID England, 2007. 730218, 7 pages, 2014. [2] T. H. H. Aldhyani and M. R. Joshi, “Integration of time series [17] D. Trad, T. Al-Ani, and M. Jemni, “Motor imagery signal clas- models with soft clustering to enhance network traffic fore- sification for BCI system using empirical mode décomposition casting,” in 2016 Second International Conference on Research and bandpower feature extraction,” Brain, vol. 7, no. 2, pp. 5– in Computational Intelligence and Communication Networks 16, 2016. (ICRCICN), pp. 212–214, Kolkata, India, September 2016. [18] G. A. V. Borg, “Psychophysical bases of perceived exertion,” [3] L.-W. Ko, C. Yang, P.-L. Wu et al., “Development of a smart Medicine and Science in Sports and Exercise, vol. 14, no. 5, helmet for strategical BCI applications,” Sensors, vol. 19, pp. 377–381, 1982. no. 8, article 1867, 2019. [19] S. Romeroa, M. A. Mañanasa, and M. J. Barbanoj, “A compar- [4] R. Vigario, J. Sarela, V. Jousmiki, M. Hamalainen, and E. Oja, ative study of automatic techniques for ocular artifact reduc- “Independent component approach to the analysis of EEG and tion in spontaneous EEG signals based on clinical target MEG recordings,” IEEE Transactions on Biomedical Engineer- variables: a simulation case,” Computers in Biology and Medi- ing, vol. 47, no. 5, pp. 589–593, 2000. cine, vol. 38, no. 3, pp. 348–360, 2008. [5] T. H. Aldhyani, A. S. Alshebami, and M. Y. Alzahrani, “Soft computing model to predict chronic diseases,” Journal of [20] M. R. Canal, “Comparison of wavelet and short time Fourier Information Science and Engineering, vol. 36, no. 2, pp. 365– transform methods in the analysis of EMG signals,” Journal 376, 2020. of Medical Systems, vol. 34, no. 1, pp. 91–94, 2010. 8 Applied Bionics and Biomechanics [21] J. R. Quinlan, “Induction of decision trees,” Machine Learning, vol. 1, no. 1, pp. 81–106, 1986. [22] C.-Y. J. Peng, K. L. Lee, and G. M. Ingersoll, “An introduction to logistic regression analysis and reporting,” The Journal of Educational Research, vol. 96, no. 1, pp. 3–14, 2010. [23] T. Fushiki, “Estimation of prediction error by using K-fold cross-validation,” Statistics and Computing, vol. 21, no. 2, pp. 137–146, 2011. [24] J. Han, M. Kamber, and J. Pei, Data Mining Concepts and Techniques 3'rd ed., 703, Morgan Kaufmann, 2011. [25] A. A. Abdul-latif, I. Cosic, D. K. Kumar, B. Polus, and C. Da Costa, “Power changes of EEG signals associated with muscle fatigue: the root mean square analysis of EEG bands,” in Pro- ceedings of the 2004 Intelligent Sensors, Sensor Networks and Information Processing Conference, pp. 531–534, Melbourne, VC, Australia, December 2004. [26] H. Moraes, C. Ferreira, A. Deslandes et al., “Beta and alpha electroencephalographic activity changes after acute exercise,” Arquivos de Neuro-Psiquiatria, vol. 65, no. 3-A, pp. 637–641, [27] H. I. Alsaadi, R. M. Almuttairi, O. Bayat, and O. N. Ucan, “Computational intelligence algorithms to handle dimension- ality reduction for enhancing intrusion detection system,” Journal of Information Science and Engineering, vol. 36, pp. 293–308, 2020. [28] J. F. Borisoff, S. G. Mason, A. Bashashati, and G. E. Birch, “Brain–computer interface design for asynchronous control applications: improvements to the LF-ASD asynchronous brain switch,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 6, pp. 985–992, 2004. [29] U. I. Awan, U. H. Rajput, G. Syed, R. Iqbal, I. Sabat, and M. Mansoor, “Effective classification of EEG signals using K- nearest neighbor algorithm,” in 2016 International Conference on Frontiers of Information Technology (FIT), pp. 120–124, Islamabad, Pakistan, December 2016. [30] M. Rashid, N. Sulaiman, M. Mustafa, S. Khatun, and B. Bifta, fication “The Classi of EEG Signal Using Different Machine Learning Techniques for BCI Application,” in Robot Intelli- gence Technology and Applications. RiTA 2018. Communica- tions in Computer and Information Science, vol 1015,J.H. Kim, H. Myung, and S. M. Lee, Eds., Springer, Singapore, 2018. [31] D. Mechau, S. Mücke, H. Liesen, and M. Weiß, “Effect of increasing running velocity on electroencephalogram in a field test,” European Journal of Applied Physiology, vol. 78, no. 4, pp. 340–345, 1998. [32] B. D. Hatfield, “Exercise and mental health: the mechanisms of exercise-induced psychological states,” in Psychology of sports, exercise and fitness, L. Diamant, Ed., pp. 17–50, Hemisphere Publishing, Washington DC, USA, 1991. [33] H. U. Amin, A. S. Malik, R. F. Ahmad et al., “Feature extrac- tion and classification for EEG signals using wavelet transform and machine learning techniques,” Australasian Physical & Engineering Sciences in Medicine, vol. 38, no. 1, pp. 139–149, [34] Y. Xiong, J. Gao, Y. Yang, X. Yu, and W. Huang, “Classifying driving fatigue based on combined entropy measure using EEG signals,” International Journal of Control and Automa- tion, vol. 9, no. 3, pp. 329–338, 2016. [35] Z. Yang and H. Ren, “Feature extraction and simulation of EEG signals during exercise-induced fatigue,” IEEE Access, vol. 7, pp. 46389–46398, 2019.

Journal

Applied Bionics and BiomechanicsHindawi Publishing Corporation

Published: Feb 8, 2021

References