Active biofeedback changes the spatial distribution of upper trapezius muscle activity during computer work

Abstract

The aim of this study was to investigate the spatio-temporal effects of advanced biofeedback by inducing active and passive pauses on the trapezius activity pattern using high-density surface electromyography (HD-EMG). Thirteen healthy male subjects performed computer work with superimposed feedback either eliciting passive (rest) or active (approximately 30% MVC) pauses based on fuzzy logic design and a control session with no feedback. HD-EMG signals of upper trapezius were recorded using a 5 × 13 multichannel electrode grid. From the HD-EMG recordings, two-dimensional maps of root mean square (RMS), relative rest time (RRT) and permuted sample entropy (PeSaEn) were obtained. The centre of gravity (CoG) and entropy of maps were used to quantify changes in the spatial distribution of muscle activity. PeSaEn as a measure of temporal heterogeneity for each channel, decreased over the whole map in response to active pause (P < 0.05) underlining a more homogenous activation pattern. Concomitantly, the CoG of RRT maps moved in caudal direction and the entropy of RMS maps as a measure of spatial heterogeneity over the whole recording grid, increased in response to active pause session compared with control session (no feedback) (P < 0.05). Active pause compared with control resulted in more heterogeneous coordination of trapezius compared with no feedback implying a more uneven spatial distribution of the biomechanical load. The study introduced new aspects in relation to the potential benefit of superimposed muscle contraction in relation to the spatial organization of muscle activity during computer work.

Appendix: Biofeedback design

The feedback instructions were generated in an online application designed in LabView (BioFeedSmart, Aalborg University, Aalborg, Denmark). Five-second epochs of trapezius EMG with 2-s overlap of ipsi-lateral clavicular and descending parts of the muscle were fed to the decision machine. RMS and PeSaEn were calculated over 500-ms epochs. RMS values were averaged and normalized to RVE. A Mamdani-style fuzzy inference was developed as the core of the decision machine. The fuzzy inference took four inputs consisting of RMS and PeSaEn of ipsi-lateral clavicular and descending parts called entries. For each entry, there were three fuzzy linguistic attributes (“LOW”, “MEDIUM” and “HIGH”). The “HIGH” attributes were set based on instructed rest at the beginning of the experimental protocol for PeSaEn and a calibration contraction composed of 90° bilateral forearm flexion for NRMS. The membership functions were defined symmetric triangular (“MEDIUM”) and trapezoidal (“LOW”/“HIGH”). Membership functions’ parameters (Fig. 6, τ _i ) were forced to keep the symmetric structure and satisfy $\tau_{4} = \tau_{3} + \frac{1}{4}(\tau_{5} - \tau_{3} ).$ Three fuzzy attributes were also defined for the consequent “LOW”, “MEDIUM” and “HIGH”. The consequent attributes were defined based on some prior pilot experiments. A full combination of four entries with three attributes gave 81 fuzzy rules where all had similar form. For example, If x ₁₁ was “HIGH” AND If x ₁₂ was “LOW” AND If x ₂₁ was “HIGH” AND If x ₂₂ was “LOW”, THEN y was “HIGH” where x ₁₁, x ₂₁ represented NRMS and x ₁₂, x ₂₂ represented PeSaEn of, respectively, the ipsi-lateral descending and clavicular parts. Finally, y represented the consequent of fuzzy inference.

Fig. 6

Membership functions prototype of the entries (system’s inputs) to fuzzy biofeedback system. The membership function defines a quantified extent to which each entry belongs to fuzzy sets (“LOW”, “MEDIUM” and “HIGH”) versus the range of input variation. τ ₁, τ ₂, τ ₃, τ ₄ and τ ₅ are membership functions parameters which have being adjusted based on history of stored inputs

However, some of the rules were not in line with our hypotheses (i.e. lower PeSaEn and higher RMS as risk factors) where two entries of a single EMG channel contradicted each other (measure contradiction) and/or the attributes from descending and clavicular parts contradicting each other (channel contradiction). This could occur, for example, with PeSaEn and NRMS of a single channel being both “HIGH” (measure contradiction) and/or PeSaEn of the descending part being “HIGH” while that of clavicular part was “LOW” (channel contradiction). In such cases, the rules were down weighed by halving the rule weight per respective contradiction and setting the consequent to “MEDIUM”. We hypothesized “LOW” PeSaEn and “HIGH” NRMS were both risk factors and the consequent should have set to “HIGH” on occurrence of these risk factors and eventually feedback should have been applied.

During computer work, triggered rules were probed and if the triggered attributes of each one of the entries were continuously the same, e. g. “HIGH” for more than 40 s (duration defined on the basis of pilot tests). Then, the parameters of membership function were readjusted to avoid continuous triggering (median of collected samples were used to readjust the membership parameters of a complementary attribute). If the output of fuzzy inference exceeded 0.7 as the rising point of “HIGH” attribute the feedback alarm could potentially set to “on”. During the first minute of recording, the feedback was set to silent to prevent applying feedback due to possible transient effect and re-adjustment of Fuzzy membership functions. Fuzzy inference was also blocked from generation of two successive alarms with a time interval <30 s.