Generalizing Upper Limb Force Modeling With Transfer Learning: A Multimodal Approach Using EMG and IMU for New Users and Conditions.

In the field of EMG-based force modeling, the ability to generalize models across individuals could play a significant role in its adoption across a range of applications, including assistive devices, robotic and rehabilitation devices. However, current studies have predominately focused on intra-subject modeling, largely neglecting the burden of end-user data acquisition. In this work, we propose the use of transfer learning (TL) to generalize force modeling to a new user by first establishing a baseline model trained using other users' data, and then adapting to the end-user using a small amount of new data (only 10% , 20% , and 40% of the new user data). Using a deep multimodal convolutional neural network, consisting of two CNN models, one with high-density (HD) EMG and one with motion data recorded by an Inertial Measurement Unit (IMU), our proposed TL technique significantly improved force modeling compared to leave-one-subject-out (LOSO) and even intra-subject scenarios. The TL approach increased the average R squared values of the force modeling task by 60.81%, 190.53%, and 199.79% compared to the LOSO case, and by 13.4%, 36.88%, and 45.51% compared to the intra-subject case for isotonic, isokinetic and dynamic conditions, respectively. These results show that it is possible to adapt to a new user with minimal data while improving performance significantly compared to the intra-subject scenario. We also show that TL can be used to generalize on a new experimental condition for a new user.

Tutorial. Surface electromyogram (sEMG) amplitude estimation: Best practices.

This tutorial intends to provide insight, instructions and "best practices" for those who are novices-including clinicians, engineers and non-engineers-in extracting electromyogram (EMG) amplitude from the bipolar surface EMG (sEMG) signal of voluntary contractions. A brief discussion of sEMG amplitude extraction from high density sEMG (HDsEMG) arrays and feature extraction from electrically elicited contractions is also provided. This tutorial attempts to present its main concepts in a straightforward manner that is accessible to novices in the field not possessing a wide range of technical background (if any) in this area. Surface EMG amplitude, also referred to as the sEMG envelope [often implemented as root mean square (RMS) sEMG or average rectified value (ARV) sEMG], quantifies the voltage variation of the sEMG signal and is grossly related to the overall neural excitation of the muscle and to peripheral parameters. The tutorial briefly reviews the physiological origin of the voluntary sEMG signal and sEMG recording, including electrode configurations, sEMG signal transduction, electronic conditioning and conversion by an analog-to-digital converter. These topics have been covered in greater detail in prior tutorials in this series. In depth descriptions of state-of-the-art methods for computing sEMG amplitude are then provided, including guidance on signal pre-conditioning, absolute value vs. square-law detection, selection of appropriate sEMG amplitude smoothing filters and attenuation of measurement noise. The tutorial provides a detailed list of best practices for sEMG amplitude estimation.

Home-based upper limb stroke rehabilitation mechatronics: challenges and opportunities.

Interest in home-based stroke rehabilitation mechatronics, which includes both robots and sensor mechanisms, has increased over the past 12 years. The COVID-19 pandemic has exacerbated the existing lack of access to rehabilitation for stroke survivors post-discharge. Home-based stroke rehabilitation devices could improve access to rehabilitation for stroke survivors, but the home environment presents unique challenges compared to clinics. The present study undertakes a scoping review of designs for at-home upper limb stroke rehabilitation mechatronic devices to identify important design principles and areas for improvement. Online databases were used to identify papers published 2010-2021 describing novel rehabilitation device designs, from which 59 publications were selected describing 38 unique designs. The devices were categorized and listed according to their target anatomy, possible therapy tasks, structure, and features. Twenty-two devices targeted proximal (shoulder and elbow) anatomy, 13 targeted distal (wrist and hand) anatomy, and three targeted the whole arm and hand. Devices with a greater number of actuators in the design were more expensive, with a small number of devices using a mix of actuated and unactuated degrees of freedom to target more complex anatomy while reducing the cost. Twenty-six of the device designs did not specify their target users' function or impairment, nor did they specify a target therapy activity, task, or exercise. Twenty-three of the devices were capable of reaching tasks, 6 of which included grasping capabilities. Compliant structures were the most common approach of including safety features in the design. Only three devices were designed to detect compensation, or undesirable posture, during therapy activities. Six of the 38 device designs mention consulting stakeholders during the design process, only two of which consulted patients specifically. Without stakeholder involvement, these designs risk being disconnected from user needs and rehabilitation best practices. Devices that combine actuated and unactuated degrees of freedom allow a greater variety and complexity of tasks while not significantly increasing their cost. Future home-based upper limb stroke rehabilitation mechatronic designs should provide information on patient posture during task execution, design with specific patient capabilities and needs in mind, and clearly link the features of the design to users' needs.

Considerations for at-home upper-limb rehabilitation technology following stroke: Perspectives of stroke survivors and therapists.

Introduction: This study investigated the needs of stroke survivors and therapists, and how they may contrast, for the design of robots for at-home post stroke rehabilitation therapy, in the Ontario, Canada, context. Methods: Individual interviews were conducted with stroke survivors (n = 10) and therapists (n = 6). The transcripts were coded using thematic analysis inspired by the WHO International Classification of Functioning, Disability, and Health. Results: Design recommendations, potential features, and barriers were identified from the interviews. Stroke survivors and therapists agreed on many of the needs for at-home robotic rehabilitation; however, stroke survivors had more insights into their home environment, barriers, and needs relating to technology, while therapists had more insights into therapy methodology and patient safety and interaction. Both groups felt a one-size-fits-all approach to rehabilitation robot design is inappropriate. Designs could address a broader range of impairments by incorporating household items and breaking activities down into their component motions. Designs should incorporate hand and wrist supports and activities. Designs should monitor trunk and shoulder motion and consider incorporating group activities. Conclusion: While therapists can provide insight in the early stages of design of rehabilitation technology, stroke survivors' perspectives are crucial to designing for the home environment.

Deep Multi-Scale Fusion of Convolutional Neural Networks for EMG-Based Movement Estimation.

EMG-based motion estimation is required for applications such as myoelectric control, where the simultaneous estimation of kinematic information, namely joint angle and velocity, is challenging and critical. We propose a novel method for accurately modelling the generated joint angle and velocity simultaneously under isotonic, isokinetic (quasi-dynamic), and fully dynamic conditions. Our solution uses two streams of CNN, called TS-CNN to learn informative features from raw EMG data using different scales and estimate the generated motion during elbow flexion and extension. The experimental results show the robustness of our approach in comparison to conventional CNN as well as some methods used in the literature. The best obtained R2 values, are 0.81±0.06, 0.71±0.06, and 0.80±0.13 for joint angle estimation and 0.78±0.05, 0.79±0.07, and 0.71±0.13 for the velocity estimation, during isotonic, isokinetic, and dynamic contractions, respectively. Additionally, our results indicate that the experimental condition can have an impact on the model's performance for motion prediction. EMG-based velocity estimation obtains higher performance than joint angle estimation under isokinetic conditions. Under dynamic conditions, joint angle estimation is more accurate than velocity estimation, and there is no difference between joint angle and velocity estimation in the isotonic case.

Convolutional Neural Network Approach for Elbow Torque Estimation during Quasi-dynamic and Dynamic Contractions.

Accurate torque estimation during dynamic conditions is challenging, yet an important problem for many applications such as robotics, prosthesis control, and clinical diagnostics. Our objective is to accurately estimate the torque generated at the elbow during flexion and extension, under quasi-dynamic and dynamic conditions. High-density surface electromyogram (HD-EMG) signals, acquired from the long head and short head of biceps brachii, brachioradialis, and triceps brachii of five participants are used to estimate the torque generated at the elbow, using a convolutional neural network (CNN). We hypothesise that incorporating the mechanical information recorded by the biodex machine, i.e., position and velocity, can improve the model performance. To investigate the effects of the added data modalities on the model accuracy, models are constructed that combine EMG and position, as well as EMG with both position and velocity. R2 values are improved by 2.35%, 37.50%, and 16.67%, when position and EMG are used as inputs to the CNN models, for isotonic, isokinetic, and dynamic cases, respectively compared to using only EMG. The model performances improves further by 2.29%, 12.12%, and 20.50% for isotonic, isokinetic, and dynamic conditions, when velocity is added with the EMG and position data.

Automated Channel Selection in High-Density sEMG for Improved Force Estimation.

Accurate and real-time estimation of force from surface electromyogram (EMG) signals enables a variety of applications. We developed and validated new approaches for selecting subsets of high-density (HD) EMG channels for improved and lower-dimensionality force estimation. First, a large dataset was recorded from a number of participants performing isometric contractions in different postures, while simultaneously recording HD-EMG channels and ground-truth force. The EMG signals were acquired from three linear surface electrode arrays, each with eight monopolar channels, and were placed on the long head and short head of the biceps brachii and brachioradialis. After data collection and pre-processing, fast orthogonal search (FOS) was employed for force estimation. To select a subset of channels, principal component analysis (PCA) in the frequency domain and a novel index called the power-correlation ratio (PCR), which maximizes the spectral power while minimizing similarity to other channels, were used. These approaches were compared to channel selection using time-domain PCA. We selected one, two, and three channels per muscle from the original seven differential channels to reduce the redundancy and correlation in the dataset. In the best case, we achieved an approximate improvement of 30% for force estimation while reducing the dimensionality by 57% for a subset of three channels.

PCA-Based Channel Selection in High-Density EMG for Improving Force Estimation.

In this paper, a method for selecting channels to improve estimated force using fast orthogonal search (FOS) has been proposed. Surface electromyogram (sEMG) signals acquired from linear surface electrode arrays, placed on the long head and short head of biceps brachii, and brachioradialis during isometric contractions are used to estimate force induced at the wrist using the FOS algorithm. The method utilizes principle component analysis (PCA) in the frequency domain to select the channels with the highest contribution to the first principal component (PC). Our analysis demonstrates that our proposed method is capable of reducing the dimensionality of the data (the number of channels was reduced from 21 to 9) while improving the accuracy of the estimated force.

An Investigation of Dimensionality Reduction Techniques for EMG-based Force Estimation.

In this paper, extracted features in time and frequency domain, from high-density surface electromyogram (HD-sEMG) signals acquired from the long head and short head of biceps brachii, and brachioradialis during isometric elbow flexion are used to estimate force induced at the wrist using an artificial neural network (ANN). Different hidden layer sizes were considered to investigate its effect on the model accuracy. Also, we applied two dimensionality reduction techniques, principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE), on the feature set and investigated their effects on force estimation accuracy.

Non-invasive detection of low-level muscle fatigue using surface EMG with wavelet decomposition.

Median frequency (MDF) is widely used for detection and tracking of muscle fatigue using surface electromyography (sEMG). However, MDF does not behave consistently or accurately distinguish fatigued from non-fatigued states. In this paper, we study the concept of low-level fatigue and propose increasing average ratio (IAR) and trigger pattern index (TPI) based on discrete wavelet transform (DWT) for distinguishing low-level muscle fatigue. We recorded sEMG using an 8-electode linear monopolar array during isometric contractions from brachioradialis (BRD), biceps brachii long head (BBL), and biceps brachii short head (BBS) muscles of different subjects when performing force exertion. We then calculated the proposed parameters for characterizing low-level fatigue. The analysis indicated that the proposed approach is more consistent and stable when distinguishing low-level muscle fatigue and sheds light on the behavior of sEMG in frequency domain with respect to low-fatigue force exertion.

Force Modelling of Upper Limb Biomechanics Using Ensemble Fast Orthogonal Search on High-Density Electromyography.

An important quality of upper limb force estimation is the repeatability and worst-case performance of the estimator. The following paper proposes a methodology using an ensemble learning technique coupled with the fast orthogonal search (FOS) algorithm to reliably predict varying isometric contractions of the right arm. This method leverages the rapid and precise modelling offered by FOS combined with a univariate outlier detection algorithm to dynamically combine the output of numerous FOS models. This is performed using high-density surface electromyography (HD-SEMG) obtained from three upper-arm muscles, the biceps brachii, triceps brachii and brachioradialis. This method offers improved performance over other HD-SEMG and SEMG based force estimators, with a substantial reduction in the number of channels required.

The role of electromechanical delay in modelling the EMG-force relationship during quasi-dynamic contractions of the upper-limb.

There is a discontinuity in published electromechanical delays (EMD) in upper-limb muscles and the state-of-the-art in modelling end-point force from electromyographic signals collected from one or more muscles. Published values are typically in the range of 10 to 30ms, depending on the nature of the contraction. In published literature where the EMG-force relationship is modelled, generally a delay of 100ms or more is induced during linear enveloping to match the EMD. The implications of EMD on end-point force prediction were considered using inter-session end-point force modelling with a support-vector-regression model. The delays were estimated using the first-order cross-correlation and the force and EMG signal were temporally aligned. The results show the delays vary by 20ms or more but did produce a notable trend based on elbow joint angle. We conclude that for upper-limb biomechanics modelling, the best practice is to align the force and EMG signals based on the induced delay during linear enveloping.

Enhanced dynamic EMG-force estimation through calibration and PCI modeling.

To accurately estimate muscle forces using electromyogram (EMG) signals, precise EMG amplitude estimation, and a modeling scheme capable of coping with the nonlinearities and dynamics of the EMG-force relationship are needed. In this work, angle-based EMG amplitude calibration and parallel cascade identification (PCI) modeling are combined for EMG-based force estimation in dynamic contractions, including concentric and eccentric contractions of the biceps brachii and triceps brachii muscles. Angle-based calibration has been shown to improve surface EMG (SEMG) based force estimation during isometric contractions through minimization of the effects of joint angle related factors, and PCI modeling captures both the nonlinear and dynamic properties of the process. SEMG data recorded during constant force, constant velocity, and varying force, varying velocity flexion and extension trials are calibrated. The calibration values are obtained at specific elbow joint angles and interpolated to cover a continuous range of joint angles. The calibrated data are used in PCI models to estimate the force induced at the wrist. The experimental results show the effectiveness of the calibration scheme, combined with PCI modeling. For the constant force, constant velocity trials, minimum %RMSE of 8.3% is achieved for concentric contractions, 10.3% for eccentric contractions and 33.3% for fully dynamic contractions. Force estimation accuracy is superior in concentric contractions in comparison to eccentric contractions , which may be indicative of more nonlinearity in the eccentric SEMG-force relationship.

Shouldering the load: A review of Joan Stevenson's work on occupational lifting and design evaluation of load carriage equipment.

OBJECTIVE: In this paper, Dr. Joan Stevenson's work on assessment of the effects of lifting, supporting and transporting loads is reviewed. A defining attribute of this work is the use of objective, biomechanical measures as the basis from which a fuller understanding of all factors affecting worker performance can be obtained, and how such performance should be measured and evaluated. METHODS: The central objectives and conclusions of Dr. Stevenson's research programs spanning the years from 1985 through 2012 are summarized and discussed in terms of an overall research trajectory. CONCLUSIONS: The guiding principle of Dr. Stevenson's work is to reduce the potential harm to which workers are exposed through the development of bona fide occupational standards, a better understanding of risk factors leading to low back pain, and the establishment of an enhanced objective design process for functional load-bearing clothing and equipment.

Surface EMG force modeling with joint angle based calibration.

In this paper, a calibration method to compensate for changes in SEMG amplitude with joint angle is introduced. Calibration factors were derived from constant amplitude surface electromyogram (SEMG) recordings from the biceps brachii (during elbow flexion) and the triceps brachii (during elbow extension) across seven elbow joint angles. SEMG data were then recorded from the elbow flexors (biceps brachii and brachioradialis) and extensors (triceps brachii) during isometric, constant force flexion and extension contractions at the same joint angles. The resulting force at the wrist was measured. The fast orthogonal search method was used to find a mapping between the system inputs - estimated SEMG amplitudes and joint angle - and the system output - measured force, for both calibrated and non-calibrated SEMG data. Models developed with calibrated data yielded a statistically significant improvement in force estimation compared to models developed with non-calibrated data, suggesting that the calibration method can compensate for changes in the SEMG-force relationship with changing joint angle. It was also found that the number of non-linear, joint angle-dependent terms used in the SEMG-force model was reduced with calibration. Additionally, initial inter-session analysis performed for four subjects suggests that calibration values can be used for subsequent recording sessions, and different output force levels.

EMG-force modeling using parallel cascade identification.

Measuring force production in muscles is important for many applications such as gait analysis, medical rehabilitation, and human-machine interaction. Substantial research has focused on finding signal processing and modeling techniques which give accurate estimates of muscle force from the surface-recorded electromyogram (EMG). The proposed methods often do not capture both the nonlinearities and dynamic components of the EMG-force relation. In this study, parallel cascade identification (PCI) is used as a dynamic estimation tool to map surface EMG recordings from upper-arm muscles to the induced force at the wrist. PCI mapping involves generating a parallel connection of a series of linear dynamic and nonlinear static blocks. The PCI model parameters were initialized to obtain the best force prediction. A comparison between PCI and a previously published Hill-based orthogonalization scheme, that captures physiological behaviour of the muscles, has shown 44% improvement in force prediction by PCI (averaged over all subjects in relative-mean-square sense). The improved performance is attributed to the structural capability of PCI to capture nonlinear dynamic effects in the generated force.

Enhanced multi-site EMG-force estimation using contact pressure.

A modification method based on integrated contact pressure and surface electromyogram (SEMG) recordings over the biceps brachii muscle is presented. Multi-site sEMGs are modified by pressure signals recorded at the same locations for isometric contractions. The resulting pressure times SEMG signals are significantly more correlated to the force induced at the wrist (FW), yielding SEMG-force models with superior performance in force estimation. A sensor patch, combining six SEMG and six contact pressure sensors was designed and built. SEMG, and contact pressure data over the biceps brachii and induced wrist force data were collected from 5 subjects. Polynomial fitting was used to find a mapping between biceps SEMG and wrist force. Comparison between evaluation values from models trained with modified and non-modified SEMG signals revealed a statistically significant superiority of models trained with the modified SEMG.

Joint angle-based EMG amplitude calibration.

A calibration method is proposed to compensate for the changes in the surface electromyogram (SEMG) amplitude level of the biceps brachii at different joint angles due to the movement of the muscle bulk under the EMG electrodes for a constant force level. To this end, an experiment was designed, and SEMG and force measurements were collected from 5 subjects. The fast orthogonal search (FOS) method was used to find a mapping between SEMG from the biceps and force recorded at the wrist. Comparison between evaluation values from models trained with calibrated and non-calibrated SEMG signals revealed a statistically significant superiority of models trained with the calibrated SEMG.

Wavelet frequency-temporal relative phase pattern analysis for intermuscular synchronization of dynamic surface EMG signals.

Cross-correlation is often used as the primary technique to compare two biological signals. Cross-correlation is an effective means to measure the synchronization of two signals assuming the relative phases of all frequencies are distributed linearly, that is, a group delay. The group delay assumption imposes an unfavorable restriction on signals with varying relative phase correlation at different frequencies. The traditional Fourier technique provides phase information for each frequency component, but it is not suitable for biological signals with non-stationary statistics. The application of a wavelet based phase analysis technique is discussed in this study. The frequency decomposition and temporally localized nature of the wavelet transform provides localized phase-frequency information for two signals. The merits and weaknesses of using the wavelet relative phase pattern for determining the synchronization of surface electromyographic signals from two muscle sites is discussed.

Dynamic modeling of EMG-force relationship using parallel cascade identification.

Parallel cascade identification (PCI) is used as a dynamic estimation tool to map surface electromyography recordings from upper-arm muscles to the elbow-induced force at the wrist. PCI mapping is composed of parallel connection of a cascade of linear dynamic and nonlinear static blocks. Experimental comparison between PCI and previously published orthogonalization scheme has shown superior force prediction by PCI. The improved performance is attributed to the structural capability of PCI in capturing nonlinear dynamic effects in the generated force.