Layout 1 Thematic Section: Advances in Musculoskeletal and Neuromuscular Rehabilitation | Maccarone & Masiero Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 Human muscle activity quantification is a vital com- ponent in biomedical engineering research.1,2 In mo- tion control, lower limb muscle activity yields information on timing, patterns, and coordination, which illumines movement disorder mechanisms.3,4 The evaluation of changes in muscle activity in cases of motor control ab- normalities or movement disorders elucidates the mech- anisms underlying these disorders.5,6 In rehabilitation, this quantification provides insights into the overall activity and response speeds, facilitating the measurement of re- habilitation progress and contributing to plan optimiza- tion.7,8 However, although electromyography (EMG) is widely employed for human muscle activity quantifica- tion, it necessitates placement of electrodes on the skin surface, which makes it unsuitable for long-term or in- tense physical activity measurements. To overcome these limitations, Isezaki et al. developed a sock-type wearable electromyograph.9 Nevertheless, issues such as noise from sock sizing and sweatinduced interference between the electrodes and skin still require mitigation. In this study, we shifted the focus to the tendons near the ankle joint and the biosignals they emit to circumvent the shortcomings of EMG signals. Tendons linked to muscles are known to emit signals associated with mechanical ac- tivities, such as mechanomyographic signals, and physio- logical tremors linked to the nervous system, both of which are useful for muscle activity quantification.10,11 The ankle joint, which is a convergence point for tendons, such as the Achilles joint, and has minimal interference from fat and muscle, is an ideal site for biosignal acquisition. Consequently, we hypothesized that biosignals from the tendons near the ankle would be effective for quantifying lower limb muscle activity. Preliminary experiments using high-sensitivity piezoelectric film sensors placed on the Achilles tendon during isometric exercises indicated a cor- relation between the exercise intensity and sensor ampli- tude. Attempting to directly identify the mechanism behind these results would necessitate medical and phys- iological experimentation, which would involve consid- erable costs and risks. Hence, in this study, we posit a hypothesis that biosignals related to muscle activity orig- inate from tendons near the ankle joint. Our aim is to in- directly corroborate this hypothesis through data science methods, allowing its plausibility to emerge. We conducted data acquisition experiments and analyzed data from 63 participants. A data acquisition system Abstract We collected biosignals from 63 participants and extracted the features corresponding to each level of exerted muscle force. Data were classified into typical and atypical patterns. Data analysis was performed using the Linear Latent Curve Model (LCM) and the Conditional Linear LCM. The typical patterns demonstrated a high degree of fit. Factors, such as ankle circumference and muscle mass, influenced the model intercept. A larger ankle circumference indicated attenuation of signal transmission from the tendon to the skin surface, leading to lower biosignal values. These results indicate that biosignals from the tendons near the ankle can be captured using piezoelectric film sensors. There are studies that define biosignals originating from tendons as mechanotendography. It has been demonstrated that the relationship between biosignals originating from tendons and the exerted muscle force can be explained linearly. Insights from this study may facilitate individualized approaches in the fields of motion control and rehabilitation. Physiological studies to elucidate the mechanisms underlying biosignal generation are necessary. Key Words: lower limb muscle activity, tendon biosignals, longitudinal data, latent curve model, piezoelectric film sensor. Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Tatsuhiko Matsumoto,1,2 Yutaka Kano1 1Graduate School of Engineering Science, Osaka University, Osaka; 2Murata Manufacturing Co., Ltd., Kyoto, Japan. This article is distributed under the terms of the Creative Commons Attribution Noncommercial License (CC BY-NC 4.0) which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited. - 15 - Non -co mmerc ial us e o nly Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 equipped with high-sensitivity piezoelectric film sensors was used to detect the biosignals at the ankle joint. The participants performed isometric exercises at four levels of muscle force exertion. The features of the piezoelectric film sensor that were correlated with the exerted muscle force were identified from the collected data. Longitudinal datasets, each consisting of four data points representing different exerted forces, were compiled for all the partic- ipants. Statistical models were fitted to evaluate the rela- tionship between the identified features and exerted muscle force. In addition, we investigated the factors con- tributing to individual variations and their underlying mechanisms. Materials and Methods Experimental methods Participants A total of 63 healthy individuals aged 21– 58 years (31 males and 32 females) participated in the experiment. Prior to the experiment, the physical measurements of the participants (age, height, weight, body fat percentage, skeletal muscle mass, ankle circumference, and average grip strength) were recorded. The body fat percentage and skeletal muscle mass were measured using a bioelectrical impedance analysis scale (InBody S10; InBody, Tokyo, Japan). Ankle circumference was measured in the upper ankle. The average grip strength was calculated as the mean of four measurements (two from each hand). Data acquisition system We developed a system for acquiring biometric signals near the ankle joint. The system uses a biodegradable pi- ezoelectric film sensor (Picoleaf®, Murata Manufacturing Co., Ltd.)12 made of polylactic acid and is characterized by nonpyroelectric properties, making it less susceptible to body temperature. This feature allows for the generation of electric charges from material strain, enabling more ac- curate measurements. Furthermore, the sensor, owing to its high sensitivity and flexibility, could detect minute skin displacements and vibrational changes caused by tendon movements, regardless of body contours, making it valu- able for biological applications. The system was designed such that a sensor was positioned on the side of the Achilles tendon.13 The sampling frequency of the sensor was set to 1000 Hz. Exercise protocol The potential noise sources in this experiment are defined as follows: i) Experimental noise (noise that may occur during the experiment, such as body movements, system positioning, and participants not performing the exercises as intended); ii) Biological noise (noise that may arise owing to day-to-day variations in the physical and neuro- logical properties of the participants). The experimental protocol was designed such that biomet- ric signals were obtained from ankle joints with minimal experimental and biological noise. A Cybex isokinetic dy- namometer (Cybex NORM®, HUMAC, CA, USA) was used14 to suppress experimental noise caused by body movements. The participants were seated in the dynanom- eter with the knee rotation axis (the lateral epicondyle of the femur) aligned with the machine axis. The knee angle was fixed at 90°, and the shin pad was secured just above the external malleolus. The ankle angle was fixed at 90°. The right foot was used for the measurements, regardless of the participant’s dominant foot or hand, and was se- cured accordingly. The dynanometer was calibrated ac- cording to the manufacturer’s recommendations before clinical trials. The experiment was structured into three phases. The warm-up phase was initially conducted with two objectives, to prevent physical abnormalities in par- ticipants during the test and to homogenize any heteroge- neous physical and neurological properties, thereby reducing biological noise. In operation training phase, the researcher provided instructions to the participants. To en- sure stable data acquisition, the participants were first se- cured to a Cybex machine and provided ample practice time (approximately 1 min) to become familiar with the operation. This was done to reduce the experimental noise that could arise from participants’ failure to perform the exercises as intended. Data acquisition phase was placed after warm-up and operational training. Let i (i = 1, ... ,4) denote the exertion force number, j (j = 1, ... ,63) the par- ticipant ID, and pij (pij = 25,50,75,100) the exertion force expressed as a percentage of each participant’s MVC. Iso- metric plantar flexion movements of the ankle were per- formed, and data were acquired in the following order: MVC (100% MVC), 25% MVC load (25% MVC), 50% MVC load (50% MVC), and 75% MVC load (75% MVC). Data acquisition was conducted in three sets, each lasting for 5s. The dynamometer-measured torque values were displayed in real time on a monitor used by the par- ticipants to control the exertion force. The participants were instructed to gradually increase their exertion force over 10s with the goal of reaching the specified value at the 10-s mark. The experimenter also monitored and con- firmed that the exerted force had reached a designated value before the measurement. A 20-s rest period was pro- vided between each measurement set. The data acquisition system and procedure are illustrated in Figure 1. Data preprocessing Feature extraction The amplitude of the sensor signal and the exerted muscle strength exhibited a directly proportional relationship. In fields such as acoustics and vibration, the magnitude of the amplitude is quantified using the root-mean-square (RMS) value.15 We chose RMS as the feature for quanti- fying the amplitude. Let t (1 ≤ t ≤ Tij) denote the time at which the raw data from the piezoelectric film sensor are acquired. If sijt represents the value of the piezoelectric film sensor signal at time t and RMSij is the RMS of the signal, the RMS can be defined by Equation (1). (1) - 16 - Non -co mmerc ial us e o nly Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 Furthermore, in the fields of acoustics and vibration, the values are often expressed in decibels relative to a stan- dard reference value.16 A similar approach was adopted in the present study. The reference value RMS0 was defined as the minimum RMS value obtained from the steady- state data of all participants, as shown in Equation (2). (2) The RMS0 value obtained from the experimental data was 2.44 × 10−5. Therefore, the magnitude of the biosignal, ex- pressed in dB, Vij was calculated using Equation (3): (3) Clustering After preprocessing, the data were structured into longitu- dinal datasets with each participant j having a biosignal Vij corresponding to the exerted muscle force pij. The dif- ference in Vij from the muscle force i to i + 1 was defined as ΔVkj (k = 1,...,3). Curve patterns were classified based on ΔVkj using the following procedure: i) Using density- based spatial clustering of applications with noise (DBSCAN), the patterns were classified into typical and atypical curve patterns based on the differences ΔV1j (be- tween 25% MVC and 50% MVC) and ΔV2j (between 50% MVC and 75% MVC);17 ii) After excluding typical curve patterns, the remaining curve patterns were reclassified using DBSCAN based on ΔV1j and ΔV2j, and it was man- ually determined which belong to typical curve patterns and which do not; iii) Typical curve patterns and others were classified using DBSCAN based on the difference ΔV3j (between 75% MVC and 100% MVC). The classification based on ΔV1j and ΔV2j was separated from that based on ΔV3j because the former was part of an experiment in which participants controlled their exerted muscle force, whereas the latter was used for the maxi- mum torque value assessment at 100% MVC. Thus, they potentially involved different types of noise. Figure 2 il- lustrates the clustering procedure and types of atypical patterns. Data analysis methods Structural equation modeling (SEM) is a versatile theo- retical framework extensively used in various research fields, such as social sciences and medicine, to analyze causal models involving latent and observed variables.18 The fundamental procedure in SEM involves creating models based on hypotheses and then selecting the opti- mal model by comparing the information criteria and fit - 17 - Figure 1. Illustration of the biometric signal acquisition system, device placement on the body and procedure. Non -co mmerc ial us e o nly Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 indices, thereby identifying the true hypothesis. In this study, we evaluated four measures: the adjusted goodness of fit index (AGFI), comparative fit index (CFI), RMSE of approximation (RMSEA), and standardized root mean square residual (SRMR). AGFI and CFI values greater than 0.9 generally indicate a good model fit.19,20 Similarly, RMSEA and SRMR values less than 0.08 are considered good, whereas values greater than 0.1 indicate a poor fit.21,22 These indices define the fit differently; therefore, we report all of them here. We based our analysis on the Latent Curve Model (LCM), which is well suited for longitudinal data.23 The LCM is a statistical model designed to model longitudinal phenom- ena. It can estimate the trajectory of intra-individual changes over time and the inter-individual differences in these changes. Thus, it estimates parameters associated with time-variant general group tendencies and estimates the extent to which these parameters vary among individ- uals within a population simultaneously. (4) (5) (6) (7) (8) (9) In this model, β0j represents the random intercept, which, in our study, denotes the population mean of the biosig- nal magnitude for the participant j at 25% MVC exerted muscle strength pij. β1j is the random slope indicating the average change in the dependent variable Vij when the exerted muscle strength pij increases by 1%. The term rij is the residual error that represents the deviation of the measured value Vij for each exerted muscle strength pij of participant j from β0j + β1jpij. Equation (5) indicates that the deviation rij follows a normal distribution with zero mean and variance σ2. This model includes various as- sumptions when approached using SEM. For example, explanatory variables for the random intercept denoted as x(l) 0j and those for the random slope denoted as x(m) 1j, with l and m representing the number of explanatory variables, can be added to equations (6) and (7) to trans- form them into equations (10) and (11), respectively, re- sulting in a conditional LCM. Equations (10) and (11) - 18 - Figure 2. Clustering procedure and illustration of atypical patterns. Non -co mmerc ial us e o nly Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 allows us to focus on explaining interindividual differ- ences.24,25 (10) (11) Results Linear LCM Path diagram is presented in Figure 3. Equalvariance con- straints are applied to the variances of V1j, V2j, and V3j, whereas V4j is treated separately. This decision is based on the different types of noise involved, as discussed in Sec- tion II-B.2. Table 1 displays the fit indices and the parameters using only typical pattern data. Overall, these indices indicate a good fit. The CFI and AGFI reflect the overall fit of the proposed model to the observed data, while the RMSEA and SRMR highlight the discrepancies, uncertainties, and error rates between the actual observations and predic- tions. The results were favorable based on both sets of in- dices. Additionally, the estimated average intercept γ00 was 53.392, and the average slope γ10 was 0.083. The 95% con- fidence interval for the slope was positive, indicating that the biosignal Vij tended to increase monotonically on aver- age, as the exerted muscle strength pij increased. This monotonic increase was quantified in increments of 0.083 for each 1% increase in pij. Moreover, the variance in the slope τ11 was 0.001. The slope distribution was assumed to follow a normal distribution, as indicated in Equation (5), suggesting that the biosignal Vij increased monotoni- cally with the exerted muscle strength pij, regardless of in- terindividual differences. Conditional linear LCM A conditional linear LCM was constructed using partici- pant-specific information to estimate intercepts and slopes. Participant-specific information was selected after removing multicollinearity through a stepwise variable in- crease and decrease method based on the Akaike infor- mation criterion. Although the SRMR value was slightly high (0.092), the AGFI, CFI, and RMSEA indices indicated a good fit. This implies that although the model fit the data well, there is potential for improvement through structural adjustments or consideration of additional variables. Overall, the model demonstrated high reliability. The estimated value of the explained variance, PVE(τ00) was 0.421, which in- dicates that the explanatory variables for the intercept could explain 42.1% of the individual variance. Therefore, the selected explanatory variables contributed signifi- cantly to the estimation of the biosignal value at 25% MVC. Table 2 lists the explanatory variables selected for the in- tercepts and slopes. The selected explanatory variables for the intercept were the sensor value at steady state (0% MVC), maximum torque value at 100% MVC, ankle cir- cumference, and average grip strength. All the variables contributed significantly to the intercept estimation when a two-tailed t-test is conducted at the 5% significance level. The most influential variable was the maximum torque value at 100% MVC, with a regression coefficient of 0.063, indicating that the intercept fluctuated by ±2.64 based on this variable. The other variables also signifi- cantly influenced the intercept: the sensor value at the steady state (±2.27), ankle circumference (±1.78), and average grip strength (±2.03). However, none of the vari- ables significantly affected the slope. Comparison of typical and atypical patterns Figure 4 presents a box plot that illustrates the distribution of the physical and experimental data for all participants and the information of the participants classified as having atypical patterns. The box represents the interquartile - 19 - Figure 3. Path diagram for the linear LCM and conditional linear LCM. Non -co mmerc ial us e o nly Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 range of the dataset and the whiskers extend to cover the remainder of the distribution. We denote the first quartile by Q1, the third quartile by Q3, and the interquartile range by IQR = Q3 − Q1. The lower and upper whiskers extend to the minimum value within Q1 − (1.5 × IQR) and maxi- mum value within Q3 + (1.5 × IQR), respectively. Data points outside these ranges were plotted as outliers. The overall distribution of the participants did not show sig- nificant differences in terms of age between males and fe- males. The participant with atypical pattern A, showing a mono- tonic decrease, was a male with short stature and a lean body type. Participants with atypical pattern B, showing a larger increase in the biosignal value relative to the ex- erted muscle strength, included two males and one female. One of the males had a higher height, weight, body fat percentage, and maximum torque at 100% MVC. Partic- ipants with atypical pattern C, in which the biosignal values tended to decrease during 100% MVC, included two males and four females. One male patient had an ankle circumference notably larger than his physique. Among the females, one had a larger ankle circumference, higher BMI, and body fat percentage. The participants, re- gardless of gender, exhibited high maximum torque values at 100% MVC. Additionally, four of the five participants with pattern C had notable experimental notes, including - 20 - Table 1. Parameters and Fit Indices of Linear LCM and Conditional Linear LCM. Linear LCM Conditional linear LCM. Model Estimate Standard 95% confidence Estimate Standard 95% confidence parameter error intervals error intervals γ00 53.392 0.501 [52.400, 54.383] 41.784 5.991 [29.921, 53.646] γ10 0.083 0.005 [0.0073, 0.0092] 0.083 0.005 [0.073, 0.092] τ00 11.079 - - 6.414 - - τ10=τ01 -0.041 - - -0.037 - - τ11 0.001 - - 0.001 - - σ2 1-3 1.377 - - 1.386 - - σ2 1-3 2.125 - - 2.069 - - PVE(τ00) - - - 0.421 - - Fit indices for the linear LCM Fit indices for the linear LCM χ2 df p-value AGFI CFI RMSEA SRMR χ2 df p-value AGFI CFI RMSEA SRMR 8.39 7 0.298 0.999 0.993 0.061 0.068 27.65 19 0.090 0.998 0.963 0.052 0.092 Table 2. Explanatory variables selected for the intercept and slope of the conditional LCM. Explanation variablefor intercept Intercept Variance Estimate Standard error p-value x(l) 0j Sensor value at steady [dB] 49.577 13.427 0.317 0.089 0.000 Torque value at MVC [N·m] 53.660 458.074 0.063 0.019 0.001 Ankle circumference [cm] 21.586 3.294 -0.501 0.197 0.011 Average grip strength [kg] 33.610 108.262 0.100 0.040 0.013 x(l) 1j All variables non-significant based on variable selection Non -co mmerc ial us e o nly Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 changes in foot fixation settings or unstable torque values during the experiment. Discussion and implications Factors contributing to inter-individual differences When examining the explanatory variables for the inter- cept, it was observed that a larger ankle circumference re- sulted in a lower intercept value. This outcome aligns with the hypothesis that biosignals generated by tendons atten- uate before reaching the skin surface and that this attenu- ation is proportional to the distance from the tendon to the skin surface. This hypothesis is consistent with the finding that higher steady-state sensor values lead to higher inter- cepts. Furthermore, higher maximum torque values at 100% MVC were associated with higher intercept values. Gen- erally, individuals with higher maximum torque values are more likely to have greater muscle mass. At 25% MVC, individuals with a greater muscle mass can mobilize more muscles. Consequently, more muscle activation leads to higher biosignal quantities, which explains why individ- uals with higher maximum torque values at 100% MVC had higher intercept values. This hypothesis is also con- sistent with the finding that higher average grip strength leads to higher intercept values. Factors contributing to atypical patterns Further analysis was conducted for participants with atypi- cal pattern C, who showed a decreasing trend, especially at p4j. Among the five participants identified as exhibiting pattern C, participants with IDs 109 and 114 displayed ex- ceptionally high maximum torque values at 100% MVC. Such high values indicate that the participants used syn- ergistic muscles (such as quadriceps) in addition to the primary muscles during the dynamometer test. In this sce- nario, the combined torque values were recorded as the maximum torque at 100% MVC. Consequently, the torque settings for p1j to p3j were set to be higher than the actual abilities of the participants, resulting in higher values of V1j to V3j. However, the activation of synergistic muscles can reduce the activity of the primary muscles.26 If V4j rep- resents the activity of the primary muscles, it exhibits lower values. The average grip strength of these two par- ticipants with IDs 109 and 114 supports the hypothesis that the maximum torque values at 100% MVC include the force from the synergistic muscles. Furthermore, changes in the fixation methods during an experiment can alter the contribution of synergistic muscles.27 The control of the primary muscles also affects the ratio of the primary and synergistic muscle contributions. These phenomena were observed in the experimental notes of four of the five participants, suggesting that the high maximum torque - 21 - Figure 4. Box plot of participant’s physical information and participants information for atypical patterns. Non -co mmerc ial us e o nly Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 values at 100% MVC included forces from the synergistic muscles. Thus, atypical pattern C can be attributed to ex- perimental noise owing to the unintended use of synergis- tic muscles during the 100% MVC tests. This also implies the validity of excluding pattern C from the model. Relationship between biosignals and mechanotendography The analysis revealed that the slopes, including the 95% confidence intervals, are positive for both the latent growth model and the conditional latent growth model. Specifically, the positivity of the slopes in the conditional latent growth model suggests that, for both genders, the biosignal Vij increases monotonically on average with the increase in muscle strength pij. This study assumes the presence of biosignals originating from tendons, which Schaefer et al. have defined and reported as mechanoten- dography.28,29 Although they assumed biosignals akin to acoustic signals, they considered lower frequencies and used the same piezoelectric sensors as we did. The bio- signals dealt with in this study are likely related to mech- anotendography as studied by Schaefer et al. To date, no study has verified the trends of mechanotendography under multi-level isometric contraction loads. This study suggests that mechanotendography can contribute not only to the detection of muscle activity levels but also to their quantification. Future research needs to investigate the relationship with the mechanotendography defined by Schaefer et al. Conclusions In this study, the biosignals originating from the tendons near the ankle joint were used for limb muscle activity quantification. We conducted data acquisition experiments and analyses on 63 participants to capture biosignals from the ankle joint and extract features corresponding to the exerted muscle forces. For typical patterns, modeling was performed using a linear LCM and a conditional linear LCM. The linear LCM for typical patterns showed high potential for linearly explaining the relationship between the exerted muscle forces and biosignals. Conditional lin- ear LCM revealed that physical information, such as ankle circumference and average grip strength, influenced the intercept of the model. This study demonstrated that the relationship between the biosignals and exerted muscle forces can be explained by a simple linear structure, which is advantageous for esti- mating exerted muscle forces using biosignals. Ad- ditionally, participant-specific physical information can be used to adjust for individual biases. These results in- crease the feasibility of realizing a generalized model for estimating exerted lower limb muscle forces. We plan to analyze the relationship between the EMG sig- nals and biosignals captured in this study to further dem- onstrate the efficacy of biosignals. Further experiments will be conducted under different conditions to validate the applicability of this method to other muscles and areas of motion. Experiments to elucidate the mechanisms un- derlying biosignal generation should also be conducted from a physiological perspective. List of abbreviations EMG, Electromyography RMS, Root mean square MVC, Maximal voluntary contraction SEM, Structural equation modeling LCM, Latent curve model AGFI, Adjusted goodness of fit index CFI, Comparative fit index RMSEA, Root mean square error of approximation SRMR, Standardized root mean square residual PVE, Proportions of variance explained IQR, Interquartile range Conflict of interest The authors declare that they have no financial or non-fi- nancial conflicts of interest related to the content of this manuscript. Funding This study did not receive any funding in any form. Ethics approval and consent to participate The experimental protocol was performed according to the principles of the Declaration of Helsinki. Informed consent was obtained from all the participants prior to the experiment, and the study was approved by the Ethics Committee of Shikoku Medical School (Approval Number: R05-08-002). Consent for publication Not applicable. Availability of data and material The dataset used and/or analysed during the current study are available from the corresponding author on reasonable request. Acknowledgments We extend our deepest gratitude to Yuichi Motohisa and Nozomi Matsunaga of the Nagai Cardiovascular Internal Medicine Clinic and Chiharu Fujisawa from Shikoku Med- ical School for their invaluable cooperation and support. The expertise and dedication of these individuals were in- strumental in the successful completion of our experiments and fundamental to achieving the profound insights gained from this research. We would also like to express our ap- preciation to Atsushi Naito, Naoki Kawara, Yutaka Taka- maru, and Risako Yamashita from Murata Manufacturing - 22 - Non -co mmerc ial us e o nly Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 Co., Ltd. for their assistance and expertise, which contrib- uted significantly to our research. We would like to thank Editage (www.editage.jp) for the English language editing. Corresponding author Tatsuhiko Matsumoto, Osaka University, 1-3, Machika- neyama, Toyonaka, Osaka, Japan. ORCID ID: 0009-0003-4423-9718 E-mail: tatsuhiko.matsumoto@murata.com Yutaka Kano ORCID ID: 0000-0001-7323-1772 E-mail: kano.yutaka.es@osaka-u.ac.jp References 1. Powell KE, Paluch AE, Blair SN. Physical activity for health: What kind? how much? how intense? on top of what? Ann Rev Public Health 2011;32:349-365. 2. Howard RM, Conway R, Harrison AJ. A survey of sensor devices: use in sports biomechanics. Sports Biomechanics 2016;15:450-461. 3. Walter JP, Kinney AL, Banks SA, et al. Muscle syn- ergies may improve optimization prediction of knee contact forces during walking. J Biomech Engin 2014;136:021031. 4. Hansen C, Teulier C, Micallef JP, et al. Lower limb muscle activity during first and second tennis serves: a comparison of three surface electromyography nor- malisation methods. Sports Biomech 2023;1-12. 5. Schmitz A, Silder A, Heiderscheit B, et al. Differences in lower-extremity muscular activation during walking between healthy older and young adults. J Electro- myography Kinesiol 2009;19:1085-91. 6. Arendt-Nielsen L, Graven-Nielsen T, Svarrer H, Svensson P. Influence of low back pain on muscle ac- tivity and coordination during gait: A clinical and ex- perimental study. Pain 1996;64:231-240. 7. Farthing JP, Chilibeck PD. The effects of eccentric and concentric training at different velocities on muscle hypertrophy. Eur J Appl Physiol 2003;89:578-86. 8. Ekstrom RA, Donatelli RA, Carp KC. Electromyo- graphic analysis of core trunk, hip, and thigh muscles during rehabilitation exercises. J Orthopaedic Sports Physical Ther 2007;37:754-62. 9. Isezaki T, Kadone H, Niijima A, et al. Sock-type wear- able sensor for estimating lower leg muscle activity using distal EMG signals. Sensors 2019;19:1954. 10. Stokes MJ, Dalton PA. Acoustic myography for inves- tigating human skeletal muscle fatigue. J Appl Physiol 1991;71:1422-6. 11. Koster B, Lauk M, Timmer J, et al. Central mech- anisms in human enhanced physiological tremor. Neu- roscience Letters 1998;241:135-8. 12. Ando M, Kawamura H, Kitada H, et al. Pressure-sen- sitive touch panel based on piezoelectric poly (l-lactic acid) film. Japanese J Appl Physics 2013;52:09KD17. 13. Jomyo S, Furui A, Matsumoto T, et al. A wearable finger-tapping motion recognition system using biode- gradable piezoelectric film sensors. In: 2021 43rd An- nual International Conference of IEEE Engineering in Medicine Biology Society (EMBC). IEEE, 2021. 14. Menegaldo LL, de Oliveira LF. Effect of muscle model parameter scaling for isometric plantar flexion torque prediction. J Biomech 2009;42:2597-2601. 15. Immovilli F, Bianchini C, Cocconcelli M, et al. Bear- ing fault model for induction motor with externally in- duced vibration. IEEE Transactions on Industrial Electronics 2012;60:3408-3418. 16. Tsypkin M. Induction motor condition monitoring: Vibration analysis technique-a twice line frequency component as a diagnostic tool. In: 2013 International Electric Machines Drives Conference, 2013. p. 117-124. 17. Ester M, Kriegel HP, Sander J, Xu X. A density-based algorithm for discovering clusters in large spatial da- tabases with noise. In: KDD, 1996. p. 226-231. 18. Hox JJ, Bechger TM. An introduction to structural equation modeling. In: Partial Least Squares Structural Equation Modeling (PLS-SEM) Using R: A Work- book, 2021. p. 1-29. 19. Hoelter JW. The analysis of covariance structures: Goodness-of-fit indices. Sociological Methods Re- search 1983;11:325-344. 20. Cole DA. Utility of confirmatory factor analysis in test validation research. J Consulting Clin Psychol 1987;5 5:584. 21. Steiger JH. Understanding the limitations of global fit assessment in structural equation modeling. Personal Individual Diff 2007;42:893-8. 22. Iacobucci D. Structural equations modeling: Fit in- dices, sample size, and advanced topics. J Consumer Psychol 2010;20:90-8. 23. McArdle JJ, Grimm KJ. Five steps in latent curve and latent change score modeling with longitudinal data. In: Longitudinal Research with Latent Variables, 2010. p. 245-273. 24. Bollen KA, Curran PJ. Latent curve models: a struc- tural equation perspective. Wiley and Sons, 2006. 25. Lance CE, Vandenberg RJ, Self RM. Latent growth models of individual change: The case of newcomer adjustment. Organizat Behav Human Decision Pro- cesses 2000;83:107-40. 26. Kubota K, Yokoyama M, Hanawa MT, et al. Muscle co-activation in the elderly contributes to control of hip and knee joint torque and endpoint force. Sci Rep 2023;13:7139. 27. Billot M, Simoneau EM, Ballay Y, et al. How the ankle joint angle alters the antagonist and agonist torques during maximal efforts in dorsi-and plantar flexion. Scandinavian J Med Sci Sports 2011;21:e273-81. 28. Schaefer LV, Bittmann FN. Mechanotendography: de- scription and evaluation of a novel method for inves- tigating the physiological mechanical oscillations of tendons using a piezo-based measurement system. Eur J f Translat Myol 2021;31:1. 29. Schaefer LV, Bittmann FN. Two forms of isometric muscle function: Interpersonal motor task supports a distinction between a holding and a pushing isometric muscle action. BioRxiv 2020;2020:08. - 23 - Non -co mmerc ial us e o nly mailto:tatsuhiko.matsumoto@murata.com mailto:kano.yutaka.es@osaka-u.ac.jp Longitudinal analysis of lower limb muscle activity and ankle tendon biosignals using structural equation modeling Eur J Transl Myol 34 (4) 12701, 2024 doi: 10.4081/ejtm.2024.12701 Disclaimer All claims expressed in this article are solely those of the authors and do not necessarily represent those of their af- filiated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher. Submitted: 1 June 2024. Accepted: 1 August 2024. Early access: 23 September 2024. - 24 - Non -co mmerc ial us e o nly