Sixteen pianists (nine females, seven males; one professor and 15 students), all self-reported as right dominant, participated in the study. The participants had at least 5 years of professional piano playing and played, on average, (15.7 ± 7.63) hours per week. None of them reported low back pain symptoms. All musicians provided informed consent prior to the tests. All the procedures used in this study were performed in accordance with the Helsinki Declaration of 1975, as revised in 2000 and 2008, and approved by the Italian National Health Service. Table 1 shows the demographic data of the sampled population. Sex, years of experience, and weekly practice are reported for completeness; differences related to these factors have not been investigated in this study.

Two chairs were tested on two different days, at least one week apart: the standard piano stool (SS) and the alternative chair (A-Chair). The A-Chair was a modified version of the Varier Move model with an added adjustable lumbar support providing a trunk-thigh angle of 115° ± 10°. The height of the SS and the lumbar support of the A-Chair were adjusted by the researchers to guarantee optimal position and comfort, for each pianist at the start of each session (Figures 1A,B). The trunk-thigh angle was measured with a hand goniometer by our expert physiotherapist (FS) and our expert kinesiologist (SD’E).

The pianists played continuously during 1 h a set of standard exercises. Firstly, the E major scale, moto retto, four octaves, three velocities followed by the E major scale, moto contrario, two octaves, average velocity, two modalities (low intensity increasing in the ascending phase and high intensity decreasing in the ascending phase). These exercises were repeated in an alternating order for the 55 min.

Every 5 min the pianists switched to a standard music piece, well-known by pianists of all ages and experience (first tempo of Mozart sonata N16, K 545 in C major). This piece was played for 20 s before switching back to the standard exercises. As such, a total of twelve 20-s recordings of HDsEMG were acquired. At the end of each session, the pianists completed questionnaires exploring their comfort/discomfort.

The response of each subject to the two chairs was assessed in two ways:

Prior to positioning of the HDsEMG grids, the skin was cleaned with abrasive paste (NuPrep, Weaver and Company, United States) and cleaned with a wet cloth to avoid paste bridges between electrodes. The grids were then placed as indicated in Figure 1C, on each side of the spine at the lumbar level on the ESM using T11 and L4 spinous processes as anatomical landmarks to ensure consistency across participants and testing sessions. The expert physiotherapist in manual therapy (FS) and our expert kinesiologist (SD’E), manually located these body landmarks using multiple methods to improve accuracy (Robinson et al., 2009; Snider et al., 2011). The electrode grids were then placed with the medial column 1 cm laterally to the spinous process, on the thoraco-lumbar muscle region identified through palpation during a lumbar extension movement. Each grid had a total of 128 electrodes (16 rows × 8 columns per array) on each side of the back. The electrodes had a diameter of 3 mm and an inter-electrode distance of 10 mm (Figure 1D), as suggested in recommendations for best practices (Merletti and Muceli, 2019).

Monopolar HDsEMG signals were collected and differentiated along the column direction, approximately in the ESM fiber direction, with respect to a reference electrode placed on the knee. An amplifier of up to 400 channels (Quattrocento, OT Bioelettronica, Turin, Italy) was used, (second order analog band-pass filter with bandwidth of 10–500 Hz, CMRR = 95 dB, input impedance >90 MΩ over the entire bandwidth, 16-bit A/D conversion, sampling frequency = 2048 Hz, gain = 500, input referred noise level <1 μV_RMS and input resolution of 0.5 μV). The differential signals were further digitally filtered with a fourth order Butterworth bidirectional (non-causal) band-pass filter with high pass cut off at 20 Hz and low pass cutoff at 400 Hz.

Background noise was estimated by performing a separate test with the same protocol on five additional subjects lying prone and relaxed on a bed for 1 h. HDsEMG signals were recorded with the same setup and procedure used for the pianists. The 12 spatial means of the background noise maps ranged from 1.90 μV_RMS to 3.30 μV_RMS with a mean of 2.61 μV_RMS and a standard deviation of 0.46 μV_RMS. Hence, the background noise level was conservatively taken as 5.0 μV_RMS (about twice the average value and 1.5 times the maximum value).

When the spatial average of the RMS values of the channels (pixels) of the entire map was higher than the noise level (5 μV_RMS), a Region of Activity (ROA) was defined using the Active Contour Segmentation algorithm (Caselles et al., 1997) available on Matlab v10 (The MathWorks Inc., Natick, MA, United States). This RMS spatial average over the ROA was defined as RMS_ROA. Otherwise the RMS_ROA was defined over the entire map.

The RMS_ROA values were small (range from 3.39 μV_RMS to 20.39 μV_RMS, with peak-to-peak sEMG values of the channels in a range of 50–200 μV) and power line interference was evident. Power line interference was removed using the spectral interpolation technique (Mewett et al., 2004). This technique was applied by (1) computing the Fourier transform of each 20-s signal, (2) removing harmonics in 10 frequency windows (centered on the first 10 harmonics of the power line) between 48 and 52 Hz, 96 and 104 Hz, 144 and 156 Hz and so on up to the 10th harmonic, (3) replacing the removed harmonics with new harmonics obtained by interpolation between previous and subsequent harmonics, (4) applying the inverse Fourier transform to re-obtain the “cleaned” signal in the time domain. ECG interference, visible in the monopolar signals, was substantially absent in the differential signals. Examples of differential signals (after spectral interpolation), obtained from column 8 of the left grid and column 1 of the right grid are provided in Figure 2.

Topographical images of the longitudinal differential HDsEMG RMS values (estimated on a 20 s epoch) provided by each grid were obtained and the ROA was computed using Caselles et al. (1997) method for segmentation. Examples of these images are provided in Figure 3 where the spatial mean of the RMS values within the contoured ROA is the RMS_ROA.

EMG measurements were performed on the RMS maps over the twelve trials (20 s every 5 min during 55 min) and used to investigate and compare chair-specific spatial and time patterns of the images. The quantitative outcome measures used were:

where x and y are the discrete coordinates of pixel (i,j) in mm defined with respect to the origin located in the top left corner of the map. I (i,j) is the intensity of pixel (i,j); and M and N are the total number of rows and columns, respectively. This computation only includes the pixels within the ROA; the pixels outside of the ROA were not computed in the summation.

The difference in displacement of the centroid of the A-Chair with respect to the SS, for the left (L) and right (R) sides was calculated as:

and

The distribution of the X_CM, Y_CM was Gaussian according to the Kolmogorov–Smirnov test. An additional ANOVA multivariate analysis was applied to explore the effects of chair, side, and time on the centroid to (a) identify significant changes in the location of the ROA centroid in time and by side of the ESM, and (b) test if the coordinates of the centroid were significantly affected by the chairs on each side of the back. An image interpolation by a factor of 15 was applied (1 pixel = 0.6 mm).

The qualitative assessment was performed using different musculoskeletal questionnaire-based evaluations that were compared between the two chairs. The statistical assessment was performed treating the evaluation outcomes as continuous variables (Beam and Wieand, 1991; Nevill et al., 2002; Harpe, 2015). These outcomes were:

Body part discomfort comparisons were performed between chairs (across pianists) using General Linear Mixed Models (GLMM). Dunn’s post hoc was applied to allow multiple comparison adjustments.

The relationship between quantitative and qualitative measurements was explored using Spearman’s correlation. This was done by correlating the RMS_ROA changes in time between chairs (across subjects), with each qualitative measure obtained (GCR, HZc, HZd, and BPD). This representation is only indicative since the 10 grades of each scale cannot be assumed to represent equally spaced values of comfort/discomfort.

All the statistical analyses were carried out using Minitab v19 (Minitab LLC, PA, United States).

Subcutaneous adipose tissue thickness is known to reduce the sEMG RMS values (Kuiken et al., 2003). SAT thicknesses were measured by three operators, and checked for differences between three measurement sites (the ESM region was divided into three sub-regions at T11, L1, L3 levels) and between left and right side, using an ultrasound scanner (Echo Blaster 128, Telemed, Lithuania). Since the Kolmogorov–Smirnov test confirmed the Gaussian distribution of the SAT values, the two-way ANOVA was used to assess significant differences between sites (across participants). The correlation between SAT and RMS amplitude was then investigated using Pearson’s correlation coefficient.

The research question addressed in this work aims at assessing changes of sEMG RMS values and of subjective perception of comfort and discomfort induced by using the A-Chair compared to using the SS in 16 subjects. Twelve 20 s measurements were performed during 55 min of play, 5 min apart, on each subject for both chairs. Statistical tests concerning sEMG RMS values were the same as adopted in previous work (Russo et al., 2019) and based on the Wilcoxon signed-rank test. Dunn’s post hoc tests were used to allow for multiple comparisons adjustments.

The qualitative assessment was carried out through questionnaire-based Likert-type evaluation which contained more than five categories. As such, an ordinal approximation of a continuous variable was used, and the evaluation of the outcomes of these scales was treated as a continuous variable (Beam and Wieand, 1991; Nevill et al., 2002; Harpe, 2015).

The spatial mean of RMS computed over the ROA (if RMS value >5 μV), or over the entire map (if RMS value <5 μV) ranged from 3.39 to 20.39 μV_RMS. An example of the signal quality is shown on a 4 s recording window (out of the 20 s) in Figure 2. Examples of RMS maps are shown in Figure 3.

In seven subjects (44%) out of 16, the signals from most electrode pairs, presented visually evident burst-like activity patterns as observed previously (Cattarello et al., 2018; Russo et al., 2019) on the same muscles. These bursts were observed to last 100–390 ms and repeating about 1.5–2.5 times per second. They are barely visible in Figure 2B. These bursts were not related to either side or chair and were not investigated further as this was not the aim of this work.

Typical topographical maps of the RMS values computed over the right and left grids at the start (recording 1, at 0 min) and end (recording 12, at 55 min) obtained from both chairs are shown in Figure 3. The mean RMS_ROA was different between chairs. The effect of the chair on the RMS_ROA of each map was quantified by the mean (over 12 measurements) percentage change of the RMS_ROA for each subject sitting on the A-Chair with respect to the SS, defined as –100 [RMS_ROA(A-Chair)–RMS_ROA (SS)]/RMS_ROA (SS). Wilcoxon ranked sign tests with Dunn’s post hoc were used to compare the median differences for each pianist (N = 16), computed across trials (N = 12) between chairs (N = 2), for each side. These comparisons (Table 2) showed that 14 (87.5%) of the 16 subjects had statistically greater values of the RMS_ROA on the SS than on the A-Chair on both left and right sides. As mentioned in section “Discussion,” these results are in agreement with Wilcoxon signed-rank tests carried out on violinists (Russo et al., 2019).

The A-Chair was generally associated to lower RMS_ROA amplitude throughout the trials when compared to the SS. This lower amplitude reflects less involvement of the ESM whilst using the A-Chair. This observation was further confirmed when changes of the RMS_ROA were assessed for each pianist by chair and side of the ESM (Table 2). The mean percent difference of the RMS_ROA across all pianists was (35.47 ± 21.46) % on the right side and (29.69 ± 23.01) % on the left side. This difference between sides was significantly different from zero (Wilcoxon signed-rank test with post hoc Dunn’s test).

The mixed effects model was applied to assess the RMS_ROA change in time individually and across the sample population to compute the differences taking fixed and random factors into consideration (fixed factors: chairs and sides; random factors: pianist and time). The statistical tests revealed that both chairs (mixed effects model, F = 423.18, p < 0.001) and sides contributed to the RMS_ROA variation with the right side presenting greater RMS amplitude than the left side (F = 21.09, p < 0.001). When these factors were assessed for each subject, the individual RMS_ROA differences were confirmed to be influenced by the chairs and side (Z = 2.69, p = 0.004). A global decreasing trend in time of RMS_ROA was observed on both chairs, with a mean difference over time between chairs of 2.80 μV with (RMS_ROA of SS) > (RMS_ROA of A-Chair). The model goodness-of-fit explained 96.97% (R²) of the RMS_ROA with a st. dev. of 0.70. This highlights that the changes on the RMS_ROA were directly influenced by the chairs and sides of the ESM.

All the results from the questionnaires were Gaussian-distributed as confirmed by the Kolmogorov–Smirnov test unless otherwise explicitly mentioned.

Spearman’s correlation was used to assess the relationship between the time change of the RMS_ROA and the GCR, HZc, and HZd. A correlation was found on the left side of the ESM between the change of the RMS_ROA on the A-Chair with respect to the SS and the GCR scores [r = 0.630; CI = (0.145, 0.871); p = 0.009], which highlights that at lower sEMG amplitude, the A-Chair is perceived to be more comfortable.

The mean SAT on the right (R) and left (L) side was 8 mm and 9 mm, respectively. No significant differences were found between 12 measurements of the SAT on the R and L side of each subject at any of the three anatomical levels. The lack of significant differences between sides of the SAT thicknesses indicates that the RMS_ROA differences between R and L sEMG amplitudes and ROAs as well as between their percent changes are due to the neural drive, postural factors and chairs.

Finally, on the L side, the A-Chair was associated to a non-significant negative correlation, implying that the RMS_ROA decreased as the SAT thickness increased [r = −0.415, CI = (−0.765, 0.123), p = 0.124]; this non-significant correlation was also observed on the SS, with RMS_ROA decreasing as SAT thickness increased [r = −0.246, CI = (−0.674, 0.304), p = 0.376]. On the R side, the A-Chair was associated to a significant positive correlation between RMS_ROA and SAT thickness, with RMS_ROA increasing as SAT thickness increased [r = 0.579, CI = (0.095, 0.842), p = 0.024]; on the SS, a non-significant positive correlation was found with RMS values increasing as SAT thickness increased [r = 0.453, CI = (−0.078, 0.783), p = 0.090] (see section “General Considerations”).

A novelty of this work is the use of a large electrode grid on each side of the spine and the definition of a ROA for each electrode grid to test the ESM of pianists. This eliminates the confounding factor due to the location of a single electrode pair, whose signal may be quite different depending on electrode location as evident from Figure 3. The need for HDsEMG technology stems also from previous work on violinists (Russo et al., 2019), and for other muscles, as indicated in the recent work of Vieira and Botter (2021). In previous reports, individual electrode pairs were placed on the back extensor muscles, including the Longissimus and Iliocostalis (Honarmand et al., 2018), with sEMG recordings for 10 min before and 10 min after 5 h of playing while the pianists were sitting on a standard piano stool or on a chair with back support. These authors also measured comfort using scales and concluded that “There were significantly lesser muscular activity, more ability to perform isometric back extension and better personal comfort while sitting on a chair with backrest.” Similarly, our exploratory work using a HDsEMG suggests that the use of a chair with lumbar support (A-Chair) ensuring a trunk-thigh angle of 115° ± 10° significantly reduces the RMS_ROA of sEMG detected over the ESM, in performing pianists, by (35.47 ± 21.46) % on the right side and (29.69 ± 23.01) % on the left side (Wilcoxon signed-rank test with post hoc Dunn’s p < 0.01).

Qualitative comfort changes indicate that the A-Chair provided greater comfort (when assessed using the GCR) than the SS. Additionally, the A-Chair was associated to more positive feelings with “likeliness” being significantly greater when compared to the SS. These qualitative results are found to be in line with the quantitative measurements of RMS and findings reported (Honarmand et al., 2018).

Two discomfort assessments were found to be of significance with the A-Chair causing greater feeling of “cramping” than the SS, and for the SS to cause a greater perception of “numbness” when compared to the A-Chair. No statistical difference was found by body part.

The lack of significant differences in the SAT values between sides of the ESM indicates that any RMS_ROA differences between sides cannot not be attributed to the SAT. Theoretically, if two measurements of sEMG RMS are performed using the SS and A-Chair one week apart, then SAT thickness, muscle and SAT-skin conductivities remained the same, the % variation of sEMG RMS should be independent of these three parameters. In this study, a significant correlation was observed between RMS_ROA values and SAT thickness on the right side of the ESM for both chairs, with RMS_ROA values increasing as SAT thickness increased. In contrast, on the left side, a correlation was found with RMS_ROA values decreasing as SAT thickness increased, but this was not statistically significant. The relation between RMS_ROA and SAT values requires further investigations that exceed the purpose of this work.

Individual comfort/discomfort, as well as sEMG RMS_ROA amplitude and PRMD, are not only related (or caused by) muscular activity: they can also be due to the mechanical conditions of the intervertebral disks and the curvature of the spine, clinically associated to back pain (Laird et al., 2014; Chun et al., 2017). This issue is not addressed in this work because no subject had back pain and clinical considerations were not the focus of this work. The literature concerning ergonomics and occupational medicine usually considers sEMG as an index of exposure, quantified through the Exposure Variation Analysis (EVA), whose values should be reduced to reduce exposure and risk (Hägg et al., 2000; Reynolds et al., 2014). However, this technique still finds limited use among clinicians and deserves further exploration.

The burst-like patterns observed on the HDsEMG of the ESM were not related to either side or chair type (visual observations). Similar patterns have been studied in sitting violinists (Khorrami Chokami et al., 2021) and should be further investigated.

Our results are in agreement with those obtained in previous work on sitting violinists (Russo et al., 2019) and show that the use of a chair with lumbar support (such as A-Chair) has a significant impact on the RMS_ROA value measured from the ESM on both violinists and pianists. When using the A-chair rather than the standard orchestra chair, nine out of nine violinists showed a significantly lower RMS_ROA on both the R and L sides (Wilcoxon signed rank test). When using the A-chair rather than the piano stool, 14 out of 16 pianists showed a lower RMS_ROA on the L side and R side (all statistically significant, Wilcoxon signed rank test). The probability of this happening by chance, according to the binomial distribution, is less than 0.004 in each case. Other qualitative findings are similar in the two studies.

Burst-like ESM activation is more evident in the violin players (eight out of nine subjects) (Russo et al., 2019) than in the piano players (seven out 16 subjects). This issue deserves further investigation.

This is the first work addressing the ESM of pianists with large electrode grids providing overall topographical maps. As shown for other muscles (Vieira and Botter, 2021), the HDsEMG approach is expected to identify the effect of two chairs on the ESM of pianists better than a single electrode pair that provides only a local sample. A total of eleven previous publications deal with HDsEMG analysis of back muscles, out of which three are from our own group, with others mostly addressing lower back pain, myofascial trigger points, lifting exercises, or other physiological investigations. Other authors have further investigated the ESM of pianists using only one or a few electrode pairs.

The relevance of HDsEMG is confirmed by Figure 3 which shows that the map of RMS amplitude over the ESM is not uniform and the signal intensity is stronger near the spine and not homogeneously distributed, confirming previous findings (De Nooij et al., 2009) and supporting the observations of Vieira and Botter (2021) about the fact that using single electrode pairs may lead to different conclusions depending on the location of the electrode pair on the muscle. The spatial average of the sEMG RMS over a ROA provides an indication of the ESM activity more reliable than that provided by a single electrode pair.

There was a significant difference between the RMS_ROA of the right and left side of the ESM. This difference may be due to hand dominance and deserves further investigation.

It is also concluded, from the relationship between quantitative and qualitative measures, that using a chair with lumbar support and a trunk-thigh angle of 115° ± 10° provides greater comfort, less exertion of the ESM, and less discomfort than the standard piano stool. This is in agreement with the findings of a previous work on sitting violinists (Russo et al., 2019).

Finally, a decreasing trend of mean RMS was observed over time, but no significant change of the ROA’s centroid was observed possibly because the inter-subject variability was considerable.