*3.7. Statistical Analysis*

In order to appreciate differences in the performances between groups based on statistical support, some verification must be done. Firstly, a Shapiro-Wilk test has been performed with the goal of checking normal distribution. In this test, the null hypothesis (H0) considers that the data set came

from a normally distributed population. The Shapiro-Wilk Test is very appropriate for small sample sizes (<50 samples).

Next, the parametric Welch's T-test has been used to estimate if pianists' performances differ significantly from non-pianists' achievements. This is a two-sample test which is used to test the hypothesis that two populations have equal means. So, the null hypothesis (H0) considers equal means between the two groups under discussion.

In both cases, a significance level α = 0.05 is assumed to be appropriate.

### **4. Results**

To carry out the experiment explained above, samples were taken from 8 subjects, 4 of whom were pianists and the other four were not (non-musicians), functioning as a control group.

In a controlled motor imagery BCI experiment, accuracy of between 80% and 90% is expected after 6–9 training sessions of 20 min each [47]. Different investigations have presented different thresholds for the "efficiency" of the BCI, but a reference value for acceptable results is 70% [29]. However, according to the state of the art, certain subjects may have difficulties using BCI systems.

Control of BCI systems requires learning both the system and the user; there must be mutual adaptation. Due to this, it is expected that the yields in the classification will increase with the user's training. Thus, the first two were introductory and learning, achieving the best performance on the third and last day.

These results are obtained from the offline analysis with the MATLAB program, using the 80 sequences together from trials 2 and 3 combined, from each session. Due to this, we will focus on the presentation of the results concerning the last day (session 3). The first trial of each day was used to train the CSP+LDA used in the online analysis in order to provide the closed loop feedback to the user.

Offline processing allows various results to be obtained, since the cross-validation technique can be applied to different subsets of data. This analysis will be done on the second and third trials of the third session, in which the user had feedback. Various authors such as [48] consider the first trial to be a training phase to give participants the opportunity to learn how to carry out the motor imagery, concentrating their analysis on the following phases. Moreover, the performance on the set of the 40 sequences of the second trial will be studied and we will also proceed with the 40 sequences of the third trial. In addition, independently, the 80 sequences of both tests will be taken together. The results corresponding to the group of pianists are summarized in Figure 4, and those of the control group, below (Figure 5).

**Figure 4.** Offline results of the group of pianists in the third session. Measurements in Run 2 and 3 were taken from feedback trials.

**Figure 5.** Offline results of the non-pianist group in the third session. Measurements in Run 2 and 3 were taken from feedback trials.

As stated above, a Shapiro–Wilk test can determine whether the data present a normal distribution. The results indicated that the data was normally distributed (*p*-values > 0.05). To compare the means of both groups, we performed the parametric Welch's T-test. The results indicated in an overall consideration that the performance of the pianists differed significantly from that of the non-pianists (*p*-value = 0.0344445, α = 0.05). Figure 6 indicates the comparison between Runs (2, 3 and 2-and-3), showing that, in all circumstances, the difference between the averages is large enough to be statistically relevant and below the significance level of 0.05.

**Figure 6.** Comparison of offline results in both groups. Measurements in Run 2 and 3 were taken from feedback trials.

The scalp topography illustrates how the physiological sources project to the scalp. Figure 7 shows two examples of projected EEG signal after a CSP filter [49]. Subplots (a) and (b) present topographies of Pianist 3 in right and left MI along Run 2 and 3 at 22 Hz. Subplots (c) and (d) present topographies of Non pianist 2. All of them were computed along Run 2 and 3 at 22 Hz. Dense red or blue areas show where the greatest differences in the projected signals were found. As can be seen, for Pianist 3, channel F4 and Cz were the most actives, being C3 and C4 in the medium scale of colors. However, in Non-Pianist 2, channels C3, C4 and Pz are marked with strong red color. In general, we observe a greater area of activation in the Non-Pianist subject as compared with that in the Pianist participant.

**Figure 7.** Examples of Projected electroencephalogram (EEG) signal after a Common Spatial Pattern (CSP) filter for Left and Right MI of Pianist 3 and Non-Pianist 2. Subfigures (**a**) and (**b**) belong to Pianist 3 (left/right, respectively), and subfigures (**c**) and (**d**) belong to Non-pianist 2 (left/right, respectively).

#### **5. Discussion**

Despite the disparity of results depending on the set of sequences analyzed, some conclusions can be outlined. Paying attention to the group of pianists, we can see that almost all of them, whatever the test taken, achieve yields in the third session greater than 70%. Pianist 1 would be a possible BCI-Inefficient user, although, as we have seen, we could only confirm this after 6 or 9 training sessions. This kind of users would require a different approach in order to improve their performance. In this sense, some studies have been developed with the aim of deploy a specific design of the experiment [50]. This would explain why the subject has not evolved or shown better performance. Among all of them, Pianist 3 stands out, reaching an almost perfect efficiency of 97.5% in the third trial. This result is very striking, considering that these rates are usually reached in more advanced sessions.

A common aspect in both groups is the drop in performance in the third trial, which occurs in 5 of the 8 cases. Mood, motivation, frustration, etc., are factors that determine the performance of the BCI. In total, the trial lasts an hour, and it is quite difficult to maintain the optimal concentration state during that time, carrying out the movement imagery tasks. Tejedor [51] calls it the "maturation effect", so that the optimum point of the results of a group of participants who are undergoing an experiment would be between the beginning (where the procedure is not yet mastered) and the end (where tiredness, fatigue, lack of interest, etc. come into play). In addition to this concept, Tejedor indicates that we have also to bear in mind the "experimental mortality", this is, the dropout of the volunteers along the experiment due to a lack of interest or change in the familiar/labor circumstances. Frustration can also trim the learning curve when we are considering inexperienced subjects. For such reason, motivation is an important issue to take into account in order to get acceptable results [52].

Comparing both figures, it is quite evident to conclude that the performance of pianists is, in general, higher than that of the control group, which seems to suggest that the proposed research hypothesis is fulfilled. Indeed, when calculating the mean of the results in each of the trials of the third session (the second, the third, and the set of second and third), we observe that the average is always higher in the pianists, as shown by the Figure 6, where the standard deviations of each mean value are also indicated.

Not only is it significant that the means are superior, regardless of the way in which the data is processed, but also that the performance is always close to 70%. Furthermore, recall that these results are impoverished by the presence of a supposed subject with BCI illiteracy, which would also explain why the mean of the set of trials 2 and 3 remains a few hundredths away from the optimal value of BCI performance. In the case of the group of non-pianists, except for some partial results, this figure is not reached. It is striking that the fourth subject of the group of non-pianists has an extremely low result when joining the two sessions.

The present study was intended to examine the analysis of the performance of a BCI device by means of the motor imagery carried out by pianists. In this sense, this work has synthesized some relevant results regarding the neurobiological differences that musical practice entails, showing that said functional specialization entails significant anatomical differences. Moreover, the foundations of the technology of the brain-computer interfaces have been exposed. These artifacts are able to detect the electrical activity of the brain with electrodes that perform an EEG and, through processing with machine learning algorithms, this biological signal is transferred into an order that can have different applications, such as the control of smart homes or robots.

The musician's brain is a paradigmatic case of neuroscience. However, in the existing literature, no previous study has been found that analyzes the performance of musicians with BCI systems, using motor imagery. The literature does seem to show that having musical prowess suggests better performance, but in the absence of a specific study in this regard, it was questionable whether, in fact, musical training and structural differences in the brain of musicians entailed better performance. To answer this hypothesis, an experimental procedure based on BCI has been designed for this paper.

Music is a global activity that involves the development of different capacities and intelligences: it improves memory, concentration, reading, self-esteem, emotional intelligence, psychomotricity, etc. In this sense, given the need to control BCI devices in the future, we can affirm that musical practice could improve motor coordination as well as neural plasticity, as revealed in the literature review, hence favoring their optimal use. To conclude that musical training is one of the factors that favors performance with motor imagery does nothing more than claim the importance of musical education in our society.

### **6. Conclusions**

In this work, the performance of a set of pianists as they interacted with an MI-based BCI system was assessed and compared with a control group. The Common Spatial Patterns (CSP) and Linear Discriminant Analysis (LDA) machine learning algorithms were applied to the EEG signals for feature extraction and classification, respectively. The results revealed that the pianists achieved a higher mean level of BCI control—by means of MI—during the final trial (74.69%) in comparison to the control group (63.13%).

Regarding the above, it can be concluded that there seems to be indications that musical training is indeed a factor that improves the performance of a BCI device through movement imagery. As mentioned before, the performances achieved by the pianists in the last trial are on the order of 10 points higher than the non-pianists, regardless of the data set analyzed, which suggests that the previous hypothesis is true.

In future research, these results could be completed, on the one hand, by expanding the sample and, on the other, by supporting training for several sessions in order to reach more definitive conclusions. With this, we could explore the improvement of the performance of the two cases we found that were BCI inefficient. As we cannot venture a hypothesis based on only three sessions, with a greater number of measurements, we would be able to explore the evolution of these participants and deepen our understanding of so-called BCI illiteracy.

**Author Contributions:** Conceptualization, J.-V.R.-R., I.R.-R., G.R.-B. and J.-V.R.; methodology, J.-V.R.-R., I.R.-R. and G.R.-B.; software, G.R.-B.; validation, J.-V.R.-R., I.R.-R. and G.R.-B.; formal analysis, J.-V.R.-R., I.R.-R. and G.R.-B.; investigation, J.-V.R.-R., I.R.-R. and G.R.-B.; resources, J.-V.R.-R., I.R.-R. and G.R.-B.; data curation, J.-V.R.-R., I.R.-R. and G.R.-B.; writing—original draft preparation, J.-V.R.-R. and I.R.-R.; writing—review and editing, J.-V.R.-R., I.R.-R., G.R.-B., J.-V.R. and J.-M.M.-G.-P.; visualization, J.-V.R.-R., I.R.-R., G.R.-B., J.-V.R. and J.-M.M.-G.-P.; supervision, I.R.-R., G.R.-B., J.-V.R. and J.-M.M.-G.-P.; project administration, G.R.-B., J.-V.R. and J.-M.M.-G.-P.; funding acquisition, J.-V.R. and J.-M.M.-G.-P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work has been funded by the Ministerio de Ciencia e Innovación, Spain (TEC2016-78028-C3-2-P and PID2019-107885GB-C33), and by European Fonds Européen de Développement Économique et Régional (FEDER) funds. This paper has been partially supported by Ministerio de Ciencia, Innovacion y Universidades grant number PGC2018-0971-B-100 and Fundación Séneca de la Región de Murcia grant number 20783/PI/18.

**Acknowledgments:** Ignacio Rodríguez-Rodríguez would like to thank the support of Programa Operativo FEDER Andalucía 2014–2020 under Project No. UMA18-FEDERJA-023 and Universidad de Málaga, Campus de Excelencia Internacional Andalucía Tech.

**Conflicts of Interest:** The authors declare no conflict of interest.
