Cepstral Analysis of EEG During Visual Perception and Mental Imagery Reveals the Influence of Artistic Expertise.

In this article, multichannel electroencephalogram (EEG) signals of artists and nonartists were analyzed during the performances of visual perception and mental imagery of paintings using cepstrum coefficients. Each of the calculated cepstrum coefficients and their parameters such as energy, average, standard deviation and entropy were separately used for distinguishing the two groups. It was found that a distinguishing coefficient might exist among the cepstrum coefficients, which could separate the two groups despite electrode placement. It was also observed that the two groups were distinguishable during the three states using the cepstrum coefficient parameters. However, separating the two groups was dependent on channel selection in this regard. The cepstrum coefficient parameters were found significantly lower for artists as compared to nonartists during the visual perception and the mental imagery, indicating a decreased average energy of EEG for artists. In addition, a similar significant decreasing trend in the cepstrum coefficient parameters was observed from occipital to frontal brain regions during the performances of the two cognitive tasks for the two groups, suggesting that visual perception and its mental imagery overlap in neuronal resources. The two groups were also classified using a neural gas classifier and a support vector machine classifier. The obtained average classification accuracies during the visual perception, the mental imagery, and at rest in the case of using the best selected distinguishable cepstrum coefficients were 76.87%, 77.5%, and 97.5%, respectively; however, a decrease in average recognition accuracy was found for classifying the two groups using the cepstrum coefficient parameters.

Introduction

Expertise and long-term training are attended by changes in certain aspects of brain activity while performing expertise-related tasks or processing expertise-related stimuli. Such differences are reflected in electroencephalogram (EEG) signals of experts.[1] Hence, EEG signals of professionals such as sportsmen and artists have been widely analyzed to date.[1234567891011121314151617] For instance, it has been found that professional marksmen exhibit significantly more alpha wave activity in left temporal, parietal, and occipital regions as compared to novices.[23] Crews et al.[4] have also observed that an increase in right-hemisphere alpha wave activity is related to decreased errors for expert golfers. An increased right-hemispheric alpha synchronization has been found for expert dancers as compared to novices during mental imagery of an improvisational dance.[5] It has been observed that musicians show an increased alpha wave activity as compared to nonmusicians when passively listening to music.[67] Petsche et al.[89] have also found that musicians and nonmusicians are different in the levels of EEG coherence and beta wave activity plays a major role in the music processing. Pang et al.[10] have demonstrated that artistic expertise is related to decrease in event-related potential (ERP) responses to visual stimuli. Bhattacharya et al.[11] have found that phase synchrony significantly increased for artists as compared to nonartists in the high-frequency bands during visual perception. Karkar et al.[12] classified the EEG signals of the two groups using scaling exponents and a neural network-based classifier with an average recognition of 81.6%. Shourie et al.[13] have also investigated differences between artists and nonartists in scaling exponents during the performances of visual perception, mental imagery, and at rest. They have observed that the two groups are distinguishable at rest using scaling exponents; however, a decrease in average recognition accuracy has been found for classifying the two groups when performing the same cognitive tasks. It has also been shown that the two groups are distinguishable using wavelet coefficients. The average classification accuracies are 72.9%, 75%, and 100% for classification of the two groups during the visual perception, the mental imagery, and at the resting condition, respectively.[14] It has been observed that the alpha wave activity significantly decreases for artists as compared to nonartists during the two cognitive tasks.[15] A significant increased approximate entropy (ApEn) has been observed for artists as compared to nonartists during visual perception and mental imagery.[16]

This review of research indicates that brain activity analysis reflects the intensive training or education that experts received and prompted the current study, which classifies EEG signals of artists and nonartists. Hence, we explored the various EEG features and found that previous research has reported good results for classification of biological signals at various states using cepstral coefficients.

For instance, seizure and nonseizure EEG signals were classified using cepstral coefficients and a standard Gaussian mixture model with an average recognition of 91.7%.[17] Normal and epileptic EEG signals were also classified using cepstral-derived features and a neural network-based classifier with an average accuracy of 100%.[1819] Other researchers have also shown the usefulness of cepstral-derived features for seizure detection.[2021] Cepstral coefficients from the autoregressive (AR) model coefficients related to the movement-related EEG signals were calculated for classifying extension, flexion, and resting types of movement using a dynamic Hidden Markov Model-based (HMM) classifier, and an average recognition of 74.6% was achieved.[22] The usefulness of cepstral analysis for measuring the depth of anesthesia was shown.[23] Four types of emotions, happiness, fear, sadness, and calmness, were distinguished using Mel Frequency Cepstral coefficients and a multilayer perceptron (MLP)-based classifier with an average accuracy of 90%.[24] Mel frequency cepstral coefficients and K nearest neighbor (KNN) classifier were used for distinguishing different pairwise vowel imagery with an overall accuracy of 75%.[25] The usefulness of cepstral analysis for distinguishing between severity levels of speech impairment has also been shown.[26] It has been reported that cepstrum coefficients can provide a robust classification performance under disturbances such as variation in muscle contraction effort, muscle fatigue, and electromyogram (EMG) electrode location shift.[27] The usefulness of cepstrum technique for identification of the various heart problems has also been shown.[282930] Five physiological variables (heart rate, blood temperature, oxygen saturation, systolic arterial blood pressure, and systolic pulmonary pressure) related to ill patients were investigated to classify them into two classes according to the time they need to reach a stable state after coronary bypass surgery using cepstral coefficients, and the usefulness of cepstral analysis in this approach has been shown. Cepstral analysis was also used as an indirect measurement of heart rate for tachyarrhythmia detection.[30]

This review of research confirms the usefulness of cepstral analysis for classifying biological signals. However, there was no extensive research that investigated differences between the two groups in terms of cepstral coefficient parameters such as energy, mean, standard deviation (SD), and entropy (ENT). Therefore, our intention was to understand whether cepstral coefficients’ differences could reflect artistic expertise. In this article, the two groups were compared during visual perception and mental imagery of some paintings. The effects of hemisphere (left vs. right) and region (frontal, centrotemporal, centroparietal, and occipital) were also considered. Results of this study may be used for measuring progress in novice artists.

We also used cepstrum coefficients’ parameters for classifying EEG signals of artists and nonartists during visual perception, mental imagery, and resting condition. While each of the power cepstrum coefficients has a different meaning, it is likely that the two groups are also separable using some of the cepstrum coefficients. In addition, differences between the two groups may only be found in some of the channels. But, the best channels for distinguishing the two groups using cepstrum coefficient were not determined.

In this research, the best distinguishable cepstrum coefficient for each channel was selected using Davies–Bouldin index (DBI). Cepstrum coefficient parameters such as energy, mean, SD, and ENT were also calculated and the best distinguishable channels for each of them were determined. Finally, the two groups were classified using the best selected cepstrum coefficients and the cepstrum coefficient parameters during the three mentioned conditions.

Materials and Methods

Data set

This study analyzed the EEG signals that were recorded in a study by Bhattacharya et al. Twenty female participants were equally divided into two groups—professional artists and nonartists participated in this study. The artists (mean age 44.3 years old) graduated from the Vienna Academy of Fine Arts with an MA degree, and the nonartists (mean age 37.5 years old) had no specific interest or education in visual arts. Participants had to perform four tasks of visual perception (looking at a painting) and four tasks of mental imagery (mentally imagining the painting shown previously) of some paintings. Each task was repeated with four different paintings (painting 1: Bean-Festival by Jordaens, painting 2: an etching by Rembrandt, painting 3: an abstract painting by Kandinsky, and painting 4: a portrait by Holbein) for 2 min. Each task was performed after a period of rest (1 min) and a distraction period of reading a newspaper article of neutral content [Figure 1].

Figure 1

A schematic diagram related to one task of visual perception (2 min) corresponding one task of mental imagery (2 min). Each task was performed after a period of rest (1 min) and a distraction period of reading a newspaper article of neutral content (2 min)[12]

The EEG signals were recorded by 19 electrodes (according to the international 10–20 system positions) with a sampling frequency of 128 Hz and a 12-bit A/D precision [Figure 2]. Electrode impedances were kept below 8 kΩ and the averaged signals of both earlobes were used as a reference.[1112] The EEG signals were digitally filtered between 0.3 Hz and 45 Hz with a 6th order Butterworth band-pass filter and were also carefully checked for artifacts, and artifactual segments caused by eye blinks, eye movements, or muscle tension were eliminated.[16]

Figure 2

19 electrode sites over the scalp according to the International 10–20 system. F refers to frontal, C for central, P for parietal, and O for occipital brain regions. Even electrodes are placed in the right hemisphere and odd electrodes are placed in the left hemisphere. The electrodes with a suffix z are placed in the midline zone[11]

Cepstral analysis

The cepstrum of a signal is defined as the inverse Fourier transform of logarithmic magnitude of the spectrum.[3031] The real cepstrum of signal x(t) was calculated as follows:

The returned sequence c(n) is a real-valued vector, the same size as the input signal x, and X(ω) is the Fourier transform of signal x. The power cepstrum was defined as follows:

where S(ω) is the power spectrum of x. Each of the power cepstrum coefficients has a different meaning. For example, c(0)2 shows the average energy:

Finer details of the spectrum shape are represented by higher cepstral coefficients.

The real cepstrum is related to the power cepstrum via the relationship 2× real cepstrum=power cepstrum.

In this research, real cepstrum coefficient parameters such as energy (E), mean (M), SD, and ENT were also computed for investigating the effect of art training in the EEG signals. The cepstrum coefficient energy (E) was calculated by the following equation:

where N is the number of cepstrum coefficients. The cepstrum coefficient mean (M) was calculated using the following equation:

The cepstrum coefficient SD was calculated as follows:

The cepstrum coefficient ENT was calculated using the following equation:

Statistical analysis

A series of 4×2×4×2 (PAINTING×HEMISPHERE×REGION×GROUP) analysis of variances (ANOVAs) with repeated measures were computed to determine whether the differences in the cepstrum coefficient parameters among the variables were significant. The PAINTING variable consisted of four paintings that participants had to look at and then imagine. The HEMISPHERE variable referred to two levels: left and right (the midline electrodes (Pz, Cz, Fz) were not included). The REGION variable consisted of four levels as follows: frontal (Fp1, F3, F7, Fp2, F4, F8), centrotemporal (C3, T3, C4, T4), parietotemporal (P3, T5, P4, T6), and occipital (O1, O2). The GROUP variable referred to two levels: artist and nonartist. Huynh–Feldt procedure was used to correct sphericity assumptions degrees of freedom, and Bonferroni method was used for multiple comparisons. The repeated measure ANOVAs were computed separately for the visual perception and the mental imagery tasks.[13]

Davies–Bouldin index

The DBI is a metric for cluster validity and measures discriminability between clusters, which is designed based on scattering matrices. A lower DBI indicates more discriminability between the clusters. Conversely, a higher DBI shows more similarity between the clusters. The DBI calculation procedure can be expressed as follows:[32]

where C is the number of clusters and Di is equal to R, for the most similar cluster to cluster i as follows:

where R, is the similarity between cluster i and j and is calculated using the following equation:

where S is a measure of within cluster scatter for cluster i, and M, is a measure of separation between the ith and the jth clusters.

where Z is the centroid of cluster i, and J is the size of cluster i. In this research, p and q are set to 2.

The range of classification accuracy may be predicted using DBI. In this approach, if the calculated DBI was lower than 2.5, one can expect that classification accuracy was more than 60%. If the DBI was lower than 1, then the recognition accuracy might be more than 90%.[14]

Classification

Neural gas classifier

Neural gas is a competitive network in that it has no certain topology. The number of its neurons is constant during a learning procedure. The neurons of the network are adapted according to their distance to training data. The neural gas algorithm can be summarized as shown below:[3334]

1 – Initialize the neurons’ set A with N neurons:

A = {c1, c2, … c}      (13)

Each neuron has a reference vector w that is chosen randomly.

2 – Calculate the distance between a training input ζ and each neuron.

3 – Order all neurons of A according to their distance to a training input ζ. It means that the sequence of indices (i1, i2,…, iN−1) should be found such that w1 is the closest reference vector to ζ, w2 is the second closest reference vector to ζ and w, and k=0, 1, …, N−1 is the kth reference vector close to ζ.

4 – Update the reference vectors as shown below:

5 – Increase the time parameter t.

6 – If t < tmax, continue with step 2.

After training, the neurons cover the space of the training data. For classification of an unknown input, its distances to all of the neurons should be calculated. The label of the closest neuron to the unknown input determines its label.[14]

SVM Classifier

The linear support vector machine (SVM) is a classification algorithm that distinguishes two groups using a separating hyperplane in the training data space and by penalizing mistakenly classified data points. The two groups are divided by a clear gap that is as wide as possible. An unknown input is then mapped into that same space and predicted to belong to a category based on which side of the gap they fall on.[35]

Results

The three cepstrum coefficient parameters were calculated for the EEG signals of the two groups during the performances of the visual perception tasks and the mental imagery tasks. A series of 4×2×4×2 (PAINTING×HEMISPHERE×REGION×GROUP) ANOVAs with repeated measures were computed to determine the significant differences in the cepstrum coefficient parameters between the variables. The obtained results are shown in Figures 3–8.

Figure 3

Changes in EEG cepstrum coefficient energy during visual perception

Figure 8

Changes in EEG cepstrum coefficient mean during mental imagery

In the cepstrum coefficient energy related to the visual perception tasks, a significant main effect REGION (F=20.18, P < 0.001) was observed, indicating a decrease in cepstrum coefficient energy from occipital to frontal brain regions. This effect was more pronounced for nonartists; however, this interaction between REGION and GROUP failed to reach statistical significance (F=1.63, P=0.18). We also found a significant main effect GROUP with lower cepstrum coefficient energy for artists (F=8.41, P < 0.01). The remaining ANOVA effects in the cepstrum coefficient energy were not significant [Figure 3].

In the cepstrum coefficient mean related to the visual perception tasks, a significant main effect REGION was found, indicating a decrease in cepstrum coefficient mean from parietotemporal to frontal brain regions (F=7.42, P < 0.001). This effect was more pronounced for artists. We also observed a significant main effect GROUP, suggesting that artists displayed a lower cepstrum coefficient mean as compared to nonartists (F=9.30, P < 0.01). Artists and nonartists exhibited different variation patterns of cepstrum coefficient mean during the visual perception tasks. This effect is evidenced by a significant interaction between PAINTING, REGION, and GROUP (F=2.05, P < 0.05). The remaining ANOVA effects in the cepstrum coefficient mean were not significant [Figure 4].

Figure 4

Changes in EEG cepstrum coefficient mean during visual perception

In the cepstrum coefficient Std related to the visual perception tasks, a significant main effect REGION was found, suggesting a decrease in cepstrum coefficient Std from occipital to frontal brain regions (F=17.13, P < 0.001). In addition, a significant main effect GROUP was observed with lower cepstrum coefficient Std for artists as compared to nonartists (F=10.26, P < 0.01). The remaining ANOVA effects in the cepstrum coefficient Std failed to reach statistical significance [Figure 5]. The cepstrum coefficient parameter was averaged across the four trials of the two groups during the visual perception and is shown in Figure 6.

Figure 5

Changes in EEG cepstrum coefficient Std during visual perception

Figure 6

The cepstrum coefficients: (a) energy, (b) mean, and (c) Std averages of the two groups across the four trials during the visual perception

In the cepstrum coefficient energy related to the mental imagery tasks, a significant main effect PAINTING was observed (F=4.48, P < 0.01). Artists and nonartists exhibited different patterns of cepstrum coefficient energy during visualization of the four paintings. This effect was evidenced by a significant interaction between PAINTING and GROUP (F=9.84, P < 0.001). In addition, a significant main effect REGION was found (F=17.22, P < 0.001), indicating a decrease in cepstrum coefficient energy from parietotemporal to frontal brain regions. A significant main effect GROUP with lower cepstrum coefficient energy was also observed for artists (F=5.75, P < 0.05). The remaining ANOVA effects in the cepstrum coefficient energy were not significant [Figure 7].

Figure 7

Changes in EEG cepstrum coefficient energy during mental imagery

In the cepstrum coefficient mean during the mental imagery tasks, a significant main effect REGION was found (F=3.10, P < 0.05), indicating an increased cepstrum coefficient mean in parietotemporal brain region. In addition, a significant interaction between PAINTING and REGION was observed (F=793.5, P < 0.01). This effect was further moderated by GROUP (PAINTING×REGION×GROUP interaction: F=974.0, P < 0.05). Artists tend to show a different pattern of cepstrum coefficient mean as compared to nonartists; however, the main effect GROUP failed to reach statistical significance (F=2.56, P=0.11). The remaining ANOVA effects were not significant [Figure 8].

In the cepstrum coefficient Std during the mental imagery tasks, a significant main effect PAINTING was observed (F=4.35, P < 0.01), indicating that artists and nonartists showed different patterns of cepstrum coefficient Std when visualizing the different paintings. This was suggested by a significant interaction between PAINTING and GROUP (F=7.33, P < 0.001). Artists displayed lower cepstrum coefficient Std as compared to nonartists (main effect GROUP: F=7.80, P < 0.01). In addition, a significant main effect REGION was observed, indicating a decrease in cepstrum coefficient Std from occipital to frontal brain regions (F=20.11, P < 0.001). The remaining ANOVA effects were not significant [Figure 9]. The cepstrum coefficient parameter was averaged across the four trials of the two groups during the mental imagery and shown in Figure 10. Summary of ANOVA effects is also shown in Table 1.

Figure 9

Changes in EEG cepstrum coefficient Std during mental imagery

Figure 10

The cepstrum coefficients: (a) energy, (b) mean, and (c) Std averages of the two groups across the four trials during the mental imagery

Table 1

Summary of ANOVA effects

Davies–Bouldin's index calculation

The cepstrum coefficients were calculated for the EEG signals of the two groups during the visual perception, the mental imagery, and at the resting condition. Then, the cepstrum coefficient parameters (energy, mean, SD, and ENT) were calculated for the obtained cepstrum coefficient. DBI was calculated for cepstrum coefficient parameters of each channel separately. The obtained results are shown in Table 2, and the best distinguishable channels (with the lowest DBI) are highlighted.

Table 2

The Davies–Bouldin index values for the average, the standard deviation (SD), the energy, and the entropy (ENT) of the cepstrum coefficients related to the two groups during the visual perception, the mental imagery, and at the resting condition

As shown in Table 2, it was found that the two groups are distinguishable during the three states using the cepstrum coefficient parameters. But, the separation of the two groups was dependent on channel selection. This was because DBI was obtained higher than 3 for some channels, and therefore, these channels are not appropriate for distinguishing the two groups. In addition, DBI decreases at the resting condition as compared to the visual perception and the mental imagery. Therefore, it was expected that the average classification accuracy of the two groups at the resting condition must be higher.

Next, DBI was calculated for each of the obtained cepstrum coefficient. Then, the cepstrum coefficient with the lowest value in the DBI was determined for each of the tasks and channels separately. The obtained results are shown in Table 3, and the indices of the best distinguishing cepstrum coefficients were noted. As shown in Table 3, no noticeable differences were observed between the different channels and DBI was obtained lower than 3 for all channels and the three conditions. Therefore, a distinguishing cepstrum coefficient may exist among the cepstrum coefficients that can discriminate the two groups despite the placement of the electrode.

Table 3

The cepstrum coefficients with the lowest Davies–Bouldin index values for each of the channels

Classification of the EEG signals

In this research, the two groups were classified using cepstrum coefficient parameters in their corresponding selected channels and a SVM and a neural gas classifier separately. For instance, in the case of the visual perception, the greatest discriminability was observed in Fz channel for the cepstrum coefficient mean and O2 for the remaining cepstrum coefficient parameters [Table 1]. Therefore, O2 and Fz channels were selected for classifying the two groups during the visual perception. For each classification, the calculated features were permuted and divided into two groups: training data (80%) and test data (20%). Finally, for each feature and state, the averaged sensitivity (true positive rate), specificity (true negative rate), and accuracy (true decision rate) across ten separate classifications were calculated using the following equations:

where TPR, SPC, and ACC represent sensitivity, specificity, and accuracy, respectively. In addition, positive referred to identified (artist) and negative referred to rejected (nonartist) (True positive (TP)=correctly identified, False positive (FP)=incorrectly identified, True negative (TN)=correctly rejected, and False negative (FN)=incorrectly rejected). The obtained results are shown in Tables 4 and 5.

Table 4

The results of classifying the two groups using SVM classifier

Table 5

The results of classifying the two groups using neural gas classifier

As shown in Tables 4 and 5, the greatest accuracy for classifying the two groups during the visual perception was 71.87% using the cepstrum coefficient energy, mean, SD, and neural gas classifier. In the case of the mental imagery, the greatest accuracy was 71.25% using the cepstrum coefficient SD and neural gas classifier. The greatest accuracy for classifying the two groups at the resting condition was 85% using the cepstrum coefficient energy and SVM classifier. The average classification accuracies of the two groups using the four cepstrum coefficient parameters were found higher at the resting condition and lower during the mental imagery.

Lastly, the two groups were also classified using the best distinguishing coefficients and the mentioned classifiers. For each feature and state, the channel with the lowest DBI was selected for classifying the two groups. For instance, in the case of visual perception, 4th cepstrum coefficient in channel Fp1 was selected for classifying the two groups. Finally, the averaged accuracy (true decision rate) across ten separate classifications was calculated for each feature and state. The obtained results are shown in Table 6.

Table 6

The results of classifying the two groups using the selected cepstrum coefficient and the two classifiers

As shown in Table 6, the greatest accuracies were obtained using SVM classifier. The average classification accuracies were found higher as compared to the obtained average classification accuracies using the four cepstrum coefficient parameters. In addition, the average accuracy of the two groups was higher at the resting condition as compared to the visual perception and the mental imagery.

Discussion

In this research, the EEG signals of artists and nonartists were analyzed and classified using cepstrum coefficients. It was observed that the cepstrum coefficient parameters were significantly lower for artists as compared to nonartists during the performances of the two cognitive tasks. Real cepstrum is related to power spectrum and power cepstrum is related to average energy of signal.[36] Therefore, the decreased cepstrum coefficient energy implies decrease in average energy of EEG for artists. Bhattacharya and Petsche have demonstrated that visual perception of art consists of at least the following three steps: (1) analysis of basic features such as colors, forms, and shapes, (2) classification of this raw information into coherent and fundamental forms, and (3) giving an appropriate meaning to these fundamental forms using previous knowledge stored in long-term memory. Features such as artistic educational background, personalities, pronounced interest in culture, and good visual memory are involved in step 3. Therefore, such personal qualities acquired by art training influence the performances of visual perception and mental imagery of individuals.[11] The mentioned differences between artists and nonartists influence their EEG signals and the EEG features such as the three cepstrum coefficient parameters can reflect them.

The obtained results may be used for measuring progress in novice artists. In this approach, an individual had to carry out the four visual perception tasks and the four mental imagery tasks, and then the cepstrum coefficient parameters are calculated. This experiment is repeated after a period of art training or extensive experiments in art, and then the calculated cepstrum coefficient parameters related to the two experiments are compared. A reduction in the cepstrum coefficient parameters after art training or extensive experiments may imply improved art-related abilities of novice artists. This measuring progress procedure can be improved using the other distinguishable features with the cepstrum coefficient parameters. For instance, Shourie et al.[1516] have reported a significant decreased alpha power and a significant increased approximate ENT for artists as compared to nonartists during the performances of the two cognitive tasks. Therefore, a reduction in both alpha power and the cepstrum coefficient parameters and an increase in approximate ENT may be related to improved art-related abilities of novice artists. Result of future researches in various distinguishable features may be used to improve the progress measuring procedure.

A significant decreasing trend in the three cepstrum coefficient parameters has also been found from occipital to frontal brain region for both groups. Therefore, this decreasing trend is related to the task type and is not influenced by artistic expertise. In addition, the trends of changes in the three cepstrum coefficient parameters related to the different regions were similar for the two cognitive tasks. This result is in accordance with the results reported by Kosslyn et al.[37] and Karkar et al.,[12] which indicate that visual perception and its mental visualization strongly overlap in terms of their neural resources.

It was also found that the two groups are distinguishable using the four cepstrum coefficient parameters (energy, mean, SD, and ENT); however, the placement of the electrode is significant in this regard. No considerable difference was found between the four cepstrum coefficient parameters for separating the two groups. A decrease in DBI was also observed at the resting condition, suggesting that the average classification accuracy for the two groups must be higher at rest. The obtained classification accuracies confirm this issue. This result is in accordance with the results reported by Shourie et al.,[1314] which indicate that discriminabilities in scaling exponent and wavelet coefficients between the two groups are higher at rest as compared to the performance of similar cognitive tasks.

It was also found that a distinguishing cepstrum coefficient might exist among the cepstrum coefficients, which can separate the two groups despite electrode placement. A decreased DBI was found for the best distinguishing cepstrum coefficient of each channel as compared to the four cepstrum coefficient parameters. Therefore, it is expected that classification accuracy of the two groups must be higher when using the best distinguishable coefficients. The obtained classification accuracies confirm this issue. The greatest accuracies for classifying the two groups during the visual perception and the mental imagery were obtained as 76.87% and 77.5%, suggesting an improved average classification accuracy as compared to the results reported in[14] (Shourie et al. classified the two groups using wavelet coefficient during the visual perception and the mental imagery with the average accuracy of 75%).

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.