Introduction

The ability to measure respiratory muscle function is clinically important in the assessment of neuromuscular1 and respiratory disease2. In the respiratory system, respiratory muscle force is usually estimated as pressure, and shortening as lung volume change or displacement of chest wall structures3. Moreover, respiratory muscle force is tightly related to the level of activation of the muscles3,4, which can be assessed by electromyography. The diaphragm is the main inspiratory muscle5,6, and so it is important to be able to measure diaphragm function accurately. Accurate assessment of diaphragm pressure generation and activation, however, requires invasive procedures, such as the balloon-catheter technique, to measure transdiaphragmatic pressure (Pdi)7, or crural diaphragm electromyography using a multipair oesophageal electrode (oesEMGdi)2,8,9,10. These invasive tests can be uncomfortable for patients and require some skill from the operator involved. They are therefore rarely carried out in clinical practice. The development of novel, non-invasive indices of diaphragm force output and activation would therefore represent a significant advance in the assessment of patients with disordered respiratory muscle function.

Transcutaneous assessment of muscle fibre vibration during contraction provides the opportunity to non-invasively quantify an index of the force exerted by the underlying contracting skeletal muscle11. In common with other skeletal muscles11, diaphragm muscle fibres vibrate laterally during contraction12. These vibrations, related to the mechanical activation of diaphragm muscle fibres, can be non-invasively recorded as sound, using microphones (phonomyography), to provide a non-invasive index of electromechanical coupling, recruitment, and Pdi generation by the diaphragm12. Alternatively, inspiratory muscle fibre vibration can be recorded by mechanomyography, using skin-surface accelerometers positioned on the chest wall over the lower intercostal spaces, proximal to the zone of apposition of the diaphragm (sMMGlic)13,14,15. The surface mechanomyogram (sMMG) is considered to be the mechanical counterpart of motor unit (MU) electrical activity as measured by surface electromyography (sEMG)16, and confers several important advantages. Unlike sEMG, sMMG is not influenced by skin preparation, bioelectrical interference from other muscles or by power line interference. Despite this, the use of sMMG to assess inspiratory muscle function has been relatively unexplored, and sMMGlic has not been compared with gold standard invasive measures of Pdi, during voluntary respiratory manoeuvres.

Measurements of both sEMGlic and sMMGlic are, however, contaminated by cardiac artefacts, especially using the average rectified value or root mean square (RMS) as standard. Recently, fixed sample entropy (fSampEn) has been proposed as a means to estimate the respiratory muscle effort from sMMGlic13,14 and sEMGlic17 signals, with less interference from cardiac artefacts. However, fSampEn has not previously been applied to analysis of gold standard invasive oesEMGdi measures.

The principal aim of the study was, therefore, to investigate the use of sMMGlic as a novel non-invasive index of inspiratory muscle force, by examining its relationship with the invasive gold standard, Pdi, in healthy subjects during an incremental inspiratory threshold loading protocol. We hypothesized that there would be close positive linear relationships between mean fSampEn |sMMGlic| and mean Pdi, and between the time-integrals of these signals. The relationship between invasive oesEMGdi and non-invasive sEMGlic was also investigated. In addition, we aimed to compare the utility of RMS- and fSampEn-based techniques to analyse oesEMGdi signals, hypothesizing that there would be a close positive relationship between RMS- and fSampEn-derived measures of oesEMGdi.

Experimental

Ethical approval

This study was approved by the NHS Health Research Authority (NRES Committee London – Dulwich 05/Q0703) and the experiments conformed to the standards of the Declaration of Helsinki. All subjects were fully informed of any risk associated with the study and provided their written consent before participation.

Subjects

Adult subjects, familiar with physiological studies, with no history of cardiorespiratory or neuromuscular disease were recruited.

Measurements

Both invasive and non-invasive measurements of inspiratory muscle force and activation were obtained simultaneously from all subjects. Unless specified, all measures were recorded continuously during all stages of the protocol.

Invasive measurements

Pdi was measured as the difference between gastric and oesophageal pressure obtained using a dual-pressure transducer tipped catheter (CTO-2; Gaeltec Devices Ltd., Dunvegan, UK) and associated amplifier (S7d; Gaeltec Devices Ltd., Dunvegan, UK), as previously described18,19,20.

Crural oesEMGdi was recorded using a multipair oesophageal electrode catheter (Yinghui Medical Equipment Technology Co. Ltd., Guangzhou, China). The catheter consisted of nine consecutive recording electrode coils, which formed five pairs of electrodes2,21. The pressure transducer and electrode catheters were inserted transnasally and once correctly positioned, taped to the nose to prevent movement during the study.

Non-invasive measurements

sMMGlic was recorded using two triaxial accelerometers (TSD109C2; BIOPAC Systems Inc., CA, USA), and associated interface (HLT100C) and isolated power supply (IPS100C) modules (BIOPAC Systems Inc., CA, USA). The accelerometers were attached bilaterally to the skin with adhesive rings as close as possible to the surface EMG electrodes along the seventh or eighth intercostal space, over the anterior axillary line14.

sEMGlic was recorded bilaterally using two pairs of disposable surface Ag/AgCl electrodes (H124SG; Covidien Kendall) placed on the skin over the seventh or eighth intercostal spaces, between the mid-axillary and the anterior axillary lines8,12,17,22. Electrode pairs were spaced 2 cm apart and a ground electrode was placed on the clavicle. The skin was appropriately prepared prior to electrode application.

Respiratory airflow was measured using a pneumotachograph (4830; Hans Rudolph Inc., KS, USA) connected to a differential pressure transducer (DP45; Validyne Engineering, CA, USA) and amplifier (CD72; Validyne Engineering, CA, USA). Mouth pressure (Pmo) was measured from a side port on the pneumotachograph using a second differential pressure transducer (MP45; Validyne Engineering, CA, USA) attached to the amplifier.

Signal Acquisition

The oesEMGdi and sEMGlic signals were amplified (gain 100), high-pass filtered at 10 Hz, and AC-coupled before acquisition (CED 1902; Cambridge Electronic Design Limited, Cambridge, UK). All signals were acquired using a 16-bit analogue-to-digital converter (PowerLab 16/35; ADInstruments Ltd., Oxford, UK) and displayed on a laptop computer running LabChart software (Version 7.2, ADInstruments Pty, Colorado Springs, USA) with analogue to digital sampling at 100 Hz (flow and pressures), 2000 Hz (sMMGlic), and 4000 Hz (oesEMGdi and sEMGlic).

Maximal volitional manoeuvres

Three maximal volitional inspiratory manoeuvres were performed initially: maximal static inspiratory pressure from functional residual capacity3 (PImax), maximal sniff pressure, and maximal inspiration to total lung capacity2,9. These manoeuvres were performed sitting upright in a chair with a noseclip in place (except for maximal sniffs) and were repeated several times to ensure maximal volitional effort. Pdi, oesEMGdi, sEMGlic, and sMMGlic were recorded continuously during the manoeuvres and peak values determined for subsequent normalization of oesEMGdi recorded during the inspiratory loading protocol (see below). Each participant’s PImax value was used to determine the inspiratory threshold loads used in their individual incremental inspiratory threshold loading protocol.

Inspiratory threshold loading protocol

All participants performed an inspiratory threshold loading protocol at five inspiratory threshold loads set at 12% (L1), 24% (L2), 36% (L3), 48% (L4), and 60% (L5) of the subject’s PImax. Inspiratory threshold loads were generated using an electronic inspiratory muscle trainer (POWERbreathe K5; POWERbreathe International Ltd., Southam, UK) attached to the distal end of the pneumotachograph. The inspiratory muscle trainer had an electronically controlled resistance valve that provided a pressure threshold resistance, which was set using the associated software (Breathe-Link, POWERbreathe International Ltd., Southam, UK). Subjects were seated and breathed through the pneumotachograph via a mouthpiece with a noseclip in place. Baseline measurements were recorded during a minimum of 2 minutes of quiet tidal breathing, following which the inspiratory muscle trainer was attached to the pneumotachograph and the series of threshold loads was imposed. Subjects were not provided with any specific instructions to adopt a certain duty cycle, and were free to choose their own breathing frequency. Subjects were, however, informed that effort was needed to overcome the threshold loads, and they were therefore encouraged to focus on using their diaphragm, to perform quick deep inspirations and to ensure that expiration was complete before making their next inspiratory effort. Each load consisted of 30 breaths followed by a resting period to allow all objective and subjective respiratory measures to return to baseline. Participants were asked to score breathlessness intensity at the end of each load using the modified Borg scale (mBorg)23. Participants were coached to anchor responses to mBorg 0 (no breathlessness), mBorg 10 (maximum breathlessness intensity imaginable) and mBorg 5 (severe, half maximal).

Data analysis

LabChart data were exported as Matlab files, and analysed offline in the widely available Matlab R2014a software. All signal processing and data analysis procedures described below were automated.

Signal filtering and pre-processing

sMMGlic signals were resampled at 200 Hz and filtered with a 4th-order zero-phase Butterworth band-pass filter between 5 and 35 Hz. Each accelerometer simultaneously provided three sMMGlic signals (sMMGlic X, sMMGlic Y, and sMMGlic Z), representing acceleration of muscle fibre vibration along all three spatial directions. A new signal, representing the total acceleration of muscle fibre vibration measured by each accelerometer, was arithmetically calculated as the norm of the vector formed by the three sMMGlic signals, as follows:

$$|sMM{G}_{{\rm{lic}}}|=\sqrt{sMM{G}_{{\rm{lic}}}\,{X}^{2}+sMM{G}_{{\rm{lic}}}\,{Y}^{2}+sMM{G}_{{\rm{lic}}}\,{Z}^{2}}$$

oesEMGdi and sEMGlic signals were resampled at 2000 Hz, and filtered with a 4th-order zero-phase Butterworth band-pass filter between 5 and 400 Hz and with four 10th-order zero-phase notch filters to remove the power line interference at 50 Hz and its harmonics at 150, 250, and 350 Hz.

Respiratory cycle segmentation and selection

Flow and pressure signals were segmented into inspiratory and expiratory segments by means of a zero-crossing detector on the flow signal, as previously described24. After segmentation all cycles were visually inspected and those either containing artefacts within the EMG and MMG signals or having an unusual Pdi pattern were rejected. The following parameters were then calculated for each respiratory cycle: inspiratory time, total breath time, and mean Pdi. The median values of all respiratory cycles during resting breathing and threshold loading were then calculated and 10 cycles that contained the three parameters nearest to the median values were automatically selected, resulting in a total of sixty respiratory cycles for each subject.

Pdi parameters

The level of inspiratory muscle force during each respiratory cycle was calculated as the mean of the inspiratory Pdi signal. Transdiaphragmatic pressure-time product (PTPdi), the time-integral of Pdi25,26, was also calculated for each respiratory cycle by multiplying the area under the curve of the inspiratory Pdi signal by the respiratory frequency, and had units of cmH2O·s·min−1 26. Both mean Pdi and PTPdi parameters were calculated after removal of the baseline from the inspiratory Pdi signal, which was determined for each respiratory cycle as the minimum level observed from the start of inspiration to the start of expiration (i.e. between points of zero flow).

Quantification of oesEMGdi signals using RMS

An additional 4th-order zero-phase Butterworth high-pass filter at 20 Hz was applied to oesEMGdi signals in order to reduce the P and T waves of electrocardiographic artefacts, and the low-frequency, large amplitude deflections in signal baseline produced by electrode motion and oesophageal peristalsis27. oesEMGdi signals were converted to RMS using a moving window of 50 ms with a one-sample step. The RMS peak values of oesEMGdi of a subject’s sixty respiratory cycles were then determined by manually analysing inspiratory oesEMGdi signal segments falling between QRS complexes of the electrocardiographic noise2. For each respiratory cycle, the highest value obtained across all five bipolar electrode pairs was selected (peak RMS oesEMGdi). As previously described2,9, the per-breath RMS peak values of oesEMGdi were expressed as percentages of the largest RMS peak value of oesEMGdi obtained throughout the inspiratory threshold loading protocol and the three maximal volitional manoeuvres (peak RMS oesEMGdi%max).

Quantification of EMG and MMG signals using fSampEn

Sample entropy is a measure of the complexity of a time-series signal28, and depends on the regularity of a signal, so that the higher the regularity, the lower the complexity, and the lower the entropy of a signal. For a given signal, regularity is calculated as the probability given by the ratio A/B, where A and B are the number of pairs of signal segments of length m + 1 and m, respectively, that are similar, that is with a maximum sample-by-sample difference less than a predefined tolerance parameter (r). The input parameter m is commonly set at 2 samples. However, r is usually expressed as a function of the standard deviation of the signal analysed. In fSampEn, the sample entropy is calculated within a moving window, instead of over the whole signal, using a fixed r value13. In this way, the fSampEn of a signal is a time-series whose values are higher not only when the signal is more complex, but also when the signal includes a wider range of amplitudes.

In this study, the oesEMGdi, sEMGlic, and |sMMGlic| signals were converted to fSampEn using a moving window of 750 ms with a 50-ms step and m = 2. The tolerance parameter, r, was set at 0.1 and 0.5 times the mean standard deviation EMG and MMG values, respectively, of respiratory cycles of the upper half loads. Values of fSampEn parameters were selected in accordance with the guidelines proposed by Estrada et al.29.

The level of inspiratory muscle force and activation during each respiratory cycle was calculated as the mean inspiratory fSampEn of the MMG and EMG signals, respectively.

Analogous to PTPdi, a novel index, the “entropy-time product (ETP)”, was calculated by multiplying the area under the curve of the inspiratory fSampEn of the MMG (ETP |sMMGlic|) signals by the respiratory frequency. Thus, ETP had units of s·min−1.

Statistical analysis

All data are expressed as median (interquartile range). Correlation coefficients were determined to measure the relationships between the recorded signals. Normality of all the parameters calculated for each subject was tested using Lilliefors tests. Since not all parameters had a normal distribution, and a linear relationship between them could not be assumed a priori, the degree of association between parameters was measured using Spearman’s rank correlation (ρ). The significance level for all correlations was set at 0.05. Statistical differences in breathing pattern (inspiratory time and respiratory rate), pressures (Pmo and Pdi) and breathlessness (mBorg) between first and last loads of the loading protocol were analysed using non-parametric Wilcoxon signed rank tests.

Within-subject correlation coefficients were calculated over the 60 respiratory cycles of each individual. A group mean correlation coefficient of the 12 participants was also calculated for each pair of parameters using the Fisher z-transform. After applying the Fisher z-transform to the correlation coefficients of the 12 participants, the transformed z-values were averaged, and the inverse Fisher z-transform was applied to the average z-value to convert it back to a group mean correlation coefficient30,31.

The strength of correlation coefficients was interpreted according to Evans’ classification32, where correlation coefficients between 0.2 and 0.39 represent a weak correlation, coefficients between 0.4 and 0.59 represent a moderate correlation, coefficients between 0.6 and 0.79 represent a strong correlation, and coefficients of 0.8 and above represent a very strong correlation.

Results

Twelve subjects (6 male, age 33 (30–38) years, BMI 22.2 (20.6–24.2) kg/m2, FEV1 98.0 (94.8–105.5) % of predicted, FVC 105.0 (91.5–110.2) % of predicted, FEV1/FVC 82.0 (74.1–83.9) %) were recruited and completed the incremental inspiratory loading protocol.

Breathing pattern, pressure generation and breathlessness intensity during incremental inspiratory threshold loading

Representative recordings from a single subject during resting breathing and loads 1 to 5 of the inspiratory threshold loading protocol are shown in Fig. 1.

Figure 1
figure 1

Measurements recorded during the inspiratory threshold loading protocol in a healthy subject. Two respiratory cycles are shown for quiet resting breathing and inspiratory threshold loads equivalent to 12%, 24%, 36%, 48%, and 60% of PImax (left to right). The oesEMGdi signal corresponds to the electrode pair 1.

Full size image

The group median (interquartile range) PImax was 87.0 (78.0–116.5) cmH2O equivalent to 109.0 (89.5–126.5) % of predicted values33. The inspiratory threshold loads increased from 11 (10–14) cmH2O during load 1 to 52 (47–70) cmH2O during load 5. Pmo increased from 0.3 (0.3–0.4) cmH2O during resting breathing to 53.7 (49.5–68.4) cmH2O (P < 0.001) at the end of load 5. Pdi increased from 11.1 (7.1–11.7) cmH2O at rest to 64.4 (46.1–67.9) cmH2O (P < 0.001) at the end of load 5. Inspiratory time ranged from 1.7 (1.4–2.1) to 2.2 (1.8–2.6) s (P = 0.11), and respiratory rate ranged from 13 (11–15) to 12 (10–13) breaths.min−1 (P = 0.27). mBorg breathlessness intensity increased from 0.5 (0.0–0.8) during load 1 to 4.0 (2.5–5.5) by the end of load 5 (P < 0.001).

Comparison between RMS and fSampEn for myographic signal analysis

RMS and fSampEn time-series of a representative subject’s myographic signals are shown in Fig. 2.

Figure 2
figure 2

RMS and fSampEn time-series of EMG and MMG signals shown in Fig. 1. Two respiratory cycles are shown for quiet breathing and 12%, 24%, 36%, 48%, and 60% of PImax (left to right).

Full size image

There was a very strong positive group mean correlation between mean fSampEn oesEMGdi and oesEMGdi%max (ρ = 0.81) (Fig. 3). Individual correlation coefficients are shown in Table 1. All correlations were statistically significant (P < 0.05).

Figure 3
figure 3

Relationship between RMS- and fSampEn-derived measures of oesEMGdi. Data points represent median and interquartile range of the 120 respiratory cycles of the twelve study subjects for each load. The group mean correlation coefficient, ρ, of the twelve subjects was calculated using the Fisher z-transform. Dashed lines show the order of execution of the inspiratory threshold loads.

Full size image
Table 1 Spearman’s rank correlation coefficients (ρ) between oesEMGdi%max and mean fSampEn oesEMGdi.
Full size table

The RMS peaks corresponding to cardiac noise were clearly identifiable and greater in the RMS than in the fSampEn time-series (Fig. 2). This was quantified by calculating the level of cardiac noise as the average ratio, in decibels, between the peak value of segments with and without cardiac noise. This ratio was calculated for the expiratory phases of resting breathing and the final load of all RMS and fSampEn time-series shown in Fig. 2 (Table 2).

Table 2 Level of cardiac noise, in decibels, of the RMS and fSampEn time-series shown in Fig. 2.
Full size table

In light of the strong positive group mean correlation between mean fSampEn oesEMGdi and oesEMGdi%max, and the robustness of fSampEn to cardiac noise, from here on EMG and MMG data will be presented as the mean fSampEn value only.

Patterns of inspiratory muscle activation during incremental inspiratory threshold loading

In general, mean Pdi, mean fSampEn sEMGlic, and mean fSampEn |sMMGlic| increased progressively during the inspiratory threshold loading protocol. However, mean fSampEn oesEMGdi increased relatively less at higher than at lower loads (Fig. 4).

Figure 4
figure 4

Measures of inspiratory muscle force and activation during inspiratory threshold loading. Data points represent median and interquartile range of the 120 respiratory cycles of the twelve study subjects for each load. Dashed lines show the order of execution of the inspiratory threshold loads.

Full size image

Mean fSampEn sEMGlic behaved similarly to mean fSampEn oesEMGdi up to 36% of PImax, but continued to increase at inspiratory threshold loads above this level.

Correlations between invasive and non-invasive indices of inspiratory muscle force

Strong to very strong positive group mean correlations were obtained between mean Pdi and mean fSampEn |sMMGlic| left (ρ = 0.76, Fig. 5a) and right (ρ = 0.81, Fig. 5b), and between PTPdi and ETP |sMMGlic| left (ρ = 0.78, Fig. 5c) and right (ρ = 0.83, Fig. 5d). Individual correlation coefficients are shown in Table 3, and individual relationships between mean Pdi and mean fSampEn |sMMGlic|, and between PTPdi and ETP |sMMGlic| are shown in the Supplementary Fig. S1. Individual and group mean correlations between the time-integrals of Pdi and fSampEn |sMMGlic| signals are shown in the Supplementary Table S1. All correlations were statistically significant (P < 0.05). These strong positive group mean correlations were observed despite an increase in the slope of the relationship between mean fSampEn |sMMGlic| and mean Pdi values at loads L2–L5 relative to loads L0–L2 (Fig. 5a,b, Supplementary Table S2).

Figure 5
figure 5

Relationship between invasive and non-invasive measures of inspiratory muscle force output recorded from the left and right sides, calculated as the mean (a,b) and time-integral (c,d) of the Pdi and fSampEn |sMMGlic| signals. Data points represent median and interquartile range of the 120 respiratory cycles of the twelve study subjects for each load. The group mean correlation coefficients, ρ, of the twelve subjects were calculated using the Fisher z-transform. Dashed lines show the order of execution of the inspiratory threshold loads.

Full size image
Table 3 Spearman’s rank correlation coefficients (ρ) between invasive (Pdi) and non-invasive (sMMGlic) measures of inspiratory muscle force.
Full size table

Correlations between invasive oesEMGdi and non-invasive sEMGlic measures

Very strong positive group mean correlations were obtained between mean fSampEn oesEMGdi and mean fSampEn sEMGlic left (ρ = 0.84, Fig. 6a) and right (ρ = 0.83, Fig. 6b). Individual correlation coefficients are shown in Table 4. All correlations were statistically significant (P < 0.05).

Figure 6
figure 6

Relationship between invasive and non-invasive measures of inspiratory muscle electrical activation recorded from the left (a) and right (b) sides, calculated as the mean of the fSampEn oesEMGdi and sEMGlic signals. Data points represent median and interquartile range of the 120 respiratory cycles of the twelve study subjects for each load. The group mean correlation coefficients, ρ, of the twelve subjects were calculated using the Fisher z-transform. Dashed lines show the order of execution of the inspiratory threshold loads.

Full size image
Table 4 Spearman’s rank correlation coefficients (ρ) between invasive (oesEMGdi) and non-invasive (sEMGlic) signals.
Full size table

Discussion

In this study, sMMGlic, Pdi, sEMGlic and oesEMGdi have been measured simultaneously for the first time, allowing investigation of the relationship between non-invasive and invasive indices of inspiratory muscle force, and electrical activation, respectively. The main finding of this study was the observation of strong to very strong positive correlations between sMMGlic, a novel non-invasive index of inspiratory muscle force output, and the invasive gold standard measure of diaphragm pressure generation, Pdi. Very strong correlations were also observed between non-invasive sEMGlic and invasive oesEMGdi measures. Furthermore, this study is the first to demonstrate strong correlations between RMS- and fSampEn-derived measures of oesEMGdi, and, importantly, the superior signal-to-noise ratio of fSampEn over RMS analysis of oesEMGdi signals, including attenuation of the cardiac artefact. This supports the use of the fSampEn technique to analyse respiratory muscle electromyogram signals.

The peak force and power output of skeletal muscle depends upon numerous factors, including muscle and fibre size and length, muscle fibre type, force-velocity relationship and the force-frequency relationship34. All of these factors are relevant to respiratory muscle function7,35,36,37,38. It is not possible to measure the force produced by the diaphragm muscle directly, and the best index of this force is Pdi, which can be recorded either during voluntary manoeuvres39,40 or in response to phrenic nerve stimulation3,40,41. Pdi is defined as the difference between intrapleural and abdominal pressure42 and, in practice, is generally equated to the arithmetic difference between gastric and oesophageal pressure3. Measurement of Pdi therefore requires insertion of balloon-catheters7,41 or a solid-state dual-pressure transducer tipped catheter20, as used in the present study.

The strong relationship between sMMGlic and Pdi in the present study suggests that sMMGlic could provide a reliable non-invasive index of inspiratory muscle force output. The MMG signal recorded at the skin surface records the summation of the mechanical activity of single MUs as pressure waves generated by the active muscle fibres, and does not, therefore, provide a direct measure of muscle force. However, the pressure waves generated by muscle fibre activity have been suggested to reflect the mechanical aspects of muscle contraction: the gross lateral movement related to the overall change of muscle geometry at the beginning of contraction, the smaller subsequent vibrations at the resonance frequency of the muscle, and the dimensional changes in the active muscle fibres43. As such, sMMG has been extensively evaluated in non-respiratory muscle groups as a tool to investigate MU recruitment strategies44,45, and as indirect indices of muscle function, including muscle force output and fatigability46. Studies of locomotor47,48, upper limb49,50,51 and intrinsic hand muscle50,52 function, have reported curvilinear relationships between the RMS sMMG and force output expressed as a % of maximum voluntary contraction (%MVC). sMMG has been observed to plateau, or even subsequently decrease, at higher activation levels in a manner that differs from muscle to muscle. Since the amplitude of MMG signals is related to both the tension increase and the velocity of tension increase during twitch and tetanic contractions53, it has been postulated that a decrease in MMG amplitude at higher firing rates reflects fusion of the mechanical activity of MUs in relation to the elevated firing rate50,53. It has therefore been proposed that the variation in RMS-sMMG/%MVC relationships between muscles corresponds to differences in MU activation strategy (MU recruitment vs frequency modulation) due to differences in muscle size and histochemical type50. RMS-sMMG/%MVC relationships have also been reported to vary with age49, sex51, and joint angle54. Single MU recordings of the diaphragm have previously shown that the diaphragm muscle predominantly employs a strategy of frequency modulation, rather than progressive recruitment of MUs throughout contraction55. In this way, the behaviour of the diaphragm is, interestingly, more similar to that of intrinsic hand muscles than that of large proximal muscles56. RMS sMMG recordings contributed to by diaphragm muscle fibres alone would therefore be expected to plateau at a relatively low %PImax, akin to RMS-sMMG/%MVC relationships observed in intrinsic hand muscles50. In the present study, however, both mean fSampEn sEMGlic and mean fSampEn |sMMGlic|, but not mean fSampEn oesEMGdi, increased progressively as %PImax and Pdi increased, without an observable plateau. The lack of progressive increase in mean fSampEn oesEMGdi likely reflects the increasing contribution of extradiaphragmatic muscle activity to non-invasive |sMMGlic| and sEMGlic signals, but not to invasive oesEMGdi signals, recorded at the highest inspiratory threshold loads57.

Few previous studies have investigated the utility of sMMG in respiratory muscle physiology. Petitjean and Bellemare12, using condenser microphones, reported significant linear relationships between the amplitude of the right and left phonomyogram and Pdi peak values, elicited by phrenic nerve stimulation. Recently, Sarlabous et al.15 used uniaxial accelerometers to record sMMGlic in patients with chronic obstructive pulmonary disease and healthy subjects during an incremental inspiratory flow protocol. Very strong correlations were observed between sMMGlic and peak inspiratory Pmo. There was also a strong correlation between sMMGlic and FEV1 in patients, suggesting the potential utility of sMMGlic as a novel non-invasive index of the mechanical load on the respiratory muscles. Distinct to this previous work, the present study employed triaxial accelerometers to record sMMGlic, allowing inspiratory muscle vibrations to be measured in all three spatial directions. Uniaxial accelerometers detect muscle mechanical activity in one spatial direction only and can therefore lead to an underestimation of muscle mechanical activation. Moreover, inspiratory threshold loading allowed the sMMGlic to be measured over a wider range of respiratory effort than the inspiratory flow protocol employed previously by Sarlabous et al.15.

Invasive measurement of Pdi and oesEMGdi is technically complex, requires some skill from the operator involved, may be time-consuming, and most importantly can be uncomfortable for study participants. In this regard, non-invasive techniques would facilitate the assessment of inspiratory muscle force and activation in physiological, and clinical, studies at scale. In contrast to sMMG, sEMG has been used extensively to assess respiratory muscle activation17,58,59,60,61,62. Reilly et al.20 reported a very strong correlation between non-invasive parasternal intercostal muscle EMG (sEMGpara) and oesEMGdi during an incremental cycle exercise test to exhaustion in healthy subjects and in cystic fibrosis patients with chronic airway obstruction. PTPdi increased with increasing sEMGpara in healthy subjects, whereas PTPdi in cystic fibrosis patients plateaued at submaximal oesEMGdi and sEMGpara levels reflecting neuromechanical uncoupling as a consequence of impaired respiratory mechanics in chronic lung disease. Simultaneous recordings of sEMGlic and sMMGlic, as described in the present study, could therefore provide the potential to assess the relationships between electrical and mechanical activation of inspiratory muscles in a wholly non-invasive manner. The oesEMGdi is increasingly recognized to provide an index of neural respiratory drive which, together with measures of respiratory muscle pressure generation, are facilitating a greater understanding of the neurophysiology of breathlessness perception in health and in respiratory disease10,63,64. sMMGlic and sEMGlic could therefore provide non-invasive estimates of neural respiratory drive to further this work at scale. A combination of sMMGlic and sEMGlic could also provide clinically applicable non-invasive indices of respiratory muscle function that could be particularly useful in the following settings: monitoring of acute exacerbations of chronic obstructive pulmonary disease65,66, weaning from mechanical ventilation in critical care settings67, monitoring of diaphragm weakness as a prognostic factor in motor neuron disease1, and monitoring patients who cannot reliably perform lung function68. The increase in the slope of the relationship between the non-invasive measures (sMMGlic and sEMGlic) and invasive measures (Pdi and oesEMGdi) at the highest inspiratory threshold loads, which we suggest is due to increased extradiaphragmatic respiratory muscle activation, occurred at relatively high levels of diaphragm activation (median(IQR) oesEMGdi%max was 50.7% (39.3%-61.7% at load L2). This level of oesEMGdi%max is higher than the values typically recorded at rest even in severe COPD patients2, and approaches end-exercise values in health and in respiratory disease10,20,64. This change in slope is therefore unlikely to significantly limit the utility of these non-invasive correlates of inspiratory muscle activation within typical clinical settings.

A major novel feature of this study is the use of the fSampEn to analyse the simultaneously recorded oesEMGdi, sEMGlic, and sMMGlic signals. The fSampEn of a signal is a time-series whose values not only depend on the signal amplitude, but also on the signal complexity. Like RMS, fSampEn can track amplitude changes evoked by EMG and MMG activity. However, since cardiac artefacts are much more regular (less complex) than EMG and MMG signals, which are random in nature, fSampEn is less influenced by cardiac artefacts than RMS, as previously described for sMMGlic13 and sEMGlic signals17. In this study, the potential of fSampEn to reduce cardiac artefacts has been shown, for the first time, in oesEMGdi signals, which are more affected by cardiac cross-talk than sEMGlic and sMMGlic signals. Based on the advantageous properties of fSampEn, the entropy-time product has been proposed in this study as a new index to measure inspiratory muscle force output from sMMGlic signals. The ETP parameter allows inspiratory muscle activity to be analysed over the whole inspiratory phase of a respiratory cycle regardless of cardiac noise which can be markedly elevated depending on recording site. Accordingly, no relevant differences have been found between correlation values of left and right sides in healthy subjects. However, an RMS-based analysis, which has been the conventional approach to analyse EMG signals, implies prior rejection of signal segments that contain cardiac noise, and therefore provides only coarse estimates of inspiratory muscle activity, since only the RMS peak value of the signal generated during muscle activation is analysed. Moreover, this is a subjective and time-consuming task when performed manually. Although some automatic algorithms have been previously proposed to remove cardiac noise from EMG signals69,70,71, these algorithms involve the recording of an extra electrocardiographic channel for QRS complexes detection.

Despite the potential for using sMMGlic and sEMGlic to assess inspiratory muscle force and activation non-invasively, we acknowledge limitations of our study, which invites further investigation. Firstly, although our findings suggest that sEMGlic and sMMGlic are useful indices of inspiratory muscle function, these measures are not specific for the diaphragm, since, as discussed above, we cannot exclude the contribution of extradiaphragmatic chest wall and abdominal musculature to these non-invasive signals, particularly during loaded breathing72,73. There is no consensus standard for optimum surface EMG electrode positioning during non-invasive assessment of diaphragm activation, with several recording sites having been suggested for recording electrode placement8,59,74,75,76,77,78,79. There is also no consensus on methods to maintain electrode orientation with respect to the muscle fibres or to control for the influence of variable muscle-to-electrode distance, such as subcutaneous fat or chest wall deformities that produce variable muscle-to-electrode filtering effects3. The participants in our study cohort were twelve healthy subjects with BMI values within the normal range and the effect of BMI in sMMGlic and sEMGlic measures should therefore be a focus of future research. Future work should also investigate the reproducibility of the sMMGlic measures in health and disease.

Regarding fSampEn, although this parameter tracked amplitude changes of related physiological signals in this and in previous studies14,15,17, fSampEn values have no units and the true physical meaning remains unclear. Therefore, definition of a normal range of values and a normalization process for fSampEn, as that described by Jolley et al.2 for oesEMGdi, is another important area of research.

In summary, we found strong correlations between non-invasive sMMGlic and sEMGlic measures and invasive Pdi and oesEMGdi measures, respectively, in healthy subjects. The strong correlation between sMMGlic and Pdi suggests that sMMGlic could provide a novel non-invasive correlate of inspiratory muscle force through which to further the study of the physiology, and pathophysiology, of inspiratory muscle function in health and disease.