A Flexible Wireless sEMG System for Wearable Muscle Strength and Fatigue Monitoring in Real Time

The detection of surface electromyography (sEMG) signals on the skin has attracted increasing attention because of its ability to monitor muscle conditions in a noninvasive manner and thus possesses great application potential to assess athletic status and training efficiency in real time or to evaluate postoperative muscle rehabilitation conveniently. Here, a flexible wireless sEMG monitoring system that consists of a stretchable sEMG epidermal patch and a flexible printed circuit board to provide real‐time evaluation of muscle strength and fatigue is reported. The epidermal patch is designed to have good stretchability and permeability and optimized to ensure a low contact impedance with the skin and minimized background noise for sEMG signal acquisition with high fidelity. Six commonly used time‐domain and two frequency‐domain features extracted from sEMG signals are systematically analyzed, and a strategy for feature selection and pattern identification is proposed that eventually enables the real‐time assessment of muscle strength and fatigue by using an integrated system in a wearable form.


A Flexible Wireless sEMG System for Wearable Muscle Strength and Fatigue Monitoring in Real Time
Qibei Gong, Xijun Jiang, Yuxuan Liu, Min Yu, and Youfan Hu* DOI: 10.1002/aelm.202200916 indicator that is measured at regular intervals to determine a patient's progress during physical rehabilitation. In a clinical environment, muscle strength is subjectively graded by a practitioner via manual muscle testing or objectively assessed by using dynamometers with specific setups, which is still largely dependent upon tester experience. More objective, quantitative and convenient assessments of muscle strength and fatigue are in high demand to benefit both clinical rehabilitation and daily athletic training.
Recently, surface electromyographic (sEMG) signals acquired on skin have gained great attention for obtaining information on muscle status in a noninvasive manner. Wearable sensor systems for sEMG monitoring have been well demonstrated but have mainly focused on humanmachine interface (HMI) applications. [3,4] It has been shown that time-and frequency-domain features of sEMG signals can be used to evaluate muscle strength or fatigue. [5][6][7][8][9] However, these investigations rely on bulk or rigid systems and are not suitable for long-term or daily monitoring purposes. Different from HMI applications, extracting this feature information from sEMG signals requires higher signal quality and stability, which makes constructing a wearable system for real-time muscle strength and fatigue assessment difficult, especially when a body is in motion.
Here, we report a flexible wireless sEMG monitoring system constructed with a stretchable sEMG epidermal patch and a flexible printed circuit board (FPCB) band for wearable muscle strength and fatigue monitoring. The sEMG epidermal patch was designed to enable conformal contact with the skin and optimized to maintain a low contact impedance with the skin and capture minimized noise from the environment to ensure high-fidelity signal acquisition. By analyzing six time-domain and two frequency-domain features, a timely signal-analyzing strategy was proposed, and the integrated system demonstrated real-time muscle strength and fatigue assessment in a wearable form with corresponding information wirelessly transmitted to and instantly displayed on a cellphone.

Flexible Wireless sEMG Monitoring System
The flexible sEMG monitoring system consists of a stretchable epidermal patch attached to the skin for sEMG signal acquisition The detection of surface electromyography (sEMG) signals on the skin has attracted increasing attention because of its ability to monitor muscle conditions in a noninvasive manner and thus possesses great application potential to assess athletic status and training efficiency in real time or to evaluate postoperative muscle rehabilitation conveniently. Here, a flexible wireless sEMG monitoring system that consists of a stretchable sEMG epidermal patch and a flexible printed circuit board to provide real-time evaluation of muscle strength and fatigue is reported. The epidermal patch is designed to have good stretchability and permeability and optimized to ensure a low contact impedance with the skin and minimized background noise for sEMG signal acquisition with high fidelity. Six commonly used time-domain and two frequency-domain features extracted from sEMG signals are systematically analyzed, and a strategy for feature selection and pattern identification is proposed that eventually enables the real-time assessment of muscle strength and fatigue by using an integrated system in a wearable form.

Introduction
Muscle fatigue, which is a decline in the maximal force or power capacity of muscle due to prolonged stimulation or exertion, is a major direct cause of athletic injuries, [1] and long-term fatigue may lead to chronic pain, musculoskeletal disorders, and even disability. [2] Improving muscle strength by training can prolong the time to fatigue, and muscle strength itself is also an important www.advelectronicmat.de and an FPCB band worn on the arm for power management, signal processing, and wireless data transmission (Figure 1a). The sEMG epidermal patch has a multilayer structure (Figure 1b) in which three Au mesh electrodes (ground electrode, reference electrode, and measurement electrode) are sandwiched between two polyimide (PI) films and covered with a sterile dressing. To improve the stretchability, flexibility, and air permeability of the sEMG epidermal patch and achieve comfortable adhesion to the skin, a serpentine layout and neutral mechanical plane design were introduced into the sandwich structure, and the sterile dressing was hollowed out with many holes. Figure 1c illustrates the block diagram and information flow of the system. The sEMG signal, measured as a differential input between the measurement and reference electrodes from the epidermal patch, was first processed by an instrumentation amplifier (common-mode rejection ratio of 80 dB at 10 kHz). In addition, to further improve the commonmode rejection ratio, our design adds a driven-right-leg circuit as a floating ground and adopts the AC coupling method to ultimately eliminate low-frequency noise. Then, a bandpass filter extracts the useful frequency band signals (sEMG: 10-500 Hz), and back-end circuits that consist of amplifiers, notch filters, and level lift circuits remove 50 Hz power line noises and improve anti-interference ability. Finally, a microcontroller unit (MCU) with a 12-bit analog-to-digital converter (ADC) and universal synchronous and asynchronous receivertransmitter (USART) serial interface handles the processed sEMG signals and communicates wirelessly with the user interface through a Bluetooth module. The user interface provides a real-time display of muscle strength and fatigue information during exercise or rehabilitation. The detailed circuit schematic diagram is shown in Figure S1, Supporting Information.

Design of Au Serpentine Mesh Electrodes and Characterization of the sEMG Epidermal Patch
As shown in Figure 2a, epidermal electrodes for signal recording were first fabricated on a Si substrate ( Figure S2, Supporting Information) and then released and transferred to a hollowed-out sterile dressing by using water-soluble tape and an electrochemical delamination method. [10,11] Au serpentine mesh structure (Figure 2b) was adopted to construct the electrodes, taking advantage of its stretchability, biocompatibility, and stability. [12] The geometry of the serpentine mesh, including the width and curvature radius of the serpentine wire, was first optimized to achieve a balance among effective contact areas with the skin, structure stretchability, and gas permeability, as shown in Figure S3, Supporting Information. The thickness of the serpentine mesh is approximately 3.5 µm to ensure good conformal contact with the skin, [3] as shown in the inset of Figure 2b. Key parameters of the electrode design that affect the quality of the obtained sEMG signal are analyzed, which include the area size of electrodes and center-to-center inter-electrode distance. First, the contact impedance between the electrode and skin, which directly determines the transmission amplitude of the epidermal electrophysiological signal, is

www.advelectronicmat.de
measured with different electrode area sizes from 10 mm 2 to 100 mm 2 when attached to the skin. As shown in Figure 2c, the impedance decreases as the size of the electrode increases and is in the range of 10-35 kΩ, which is excellent for dry electrodes that generally have a contact impedance in the range of 30 kΩ-1 MΩ. [13,14] Then, sEMG signals were recorded from the flexor carpi radialis of the right forearm during isometric contractions (see method for details), and typical data are shown in the inset of Figure 2d. The recorded signal before implementing isometric concentration clearly reveals that there are background noises recorded from the electrodes. Six sets of time-domain features are extracted from these signals for further electrode evaluation in terms of noise level and SNR. These features include root mean square (RMS), mean absolute value (MAV), integrated electromyogram (iEMG), variance (VAR), log-detector (logDetect), and ν-order (νOrder), which all www.advelectronicmat.de have been used for muscle strength assessment [15,16] and are defined as follow: where x represents the sEMG signal in an analysis time window with N samples, x i is the i-th sample in this analysis window, ν is 3, and t is 0.25. According to the recorded background noise magnitude (Figure 2d,e) and the calculated SNR of the signal (Figure 2f) with different area sizes and inter-electrode distances for these six features and with the consideration of spatial resolution, electrodes with an area size of 50 mm 2 (length: 10 mm, width: 5 mm) and an inter-electrode distance of 2 cm were selected, ensuring a noise level < 65 mV, an SNR > 22 dB and a whole signal capture area < 2.5 cm × 1 cm. The impedance change of the Au serpentine mesh electrode is measured and compared with controls (circle-type, moontype, and rectangle-type commercial electrodes, Figure S4, Supporting Information) in the frequency range between 1 Hz and 1 MHz, as shown in Figure 2g. The contact impedance of the Au serpentine mesh electrode is significantly lower than that of three other commercial electrodes in the low-frequency range (1-800 Hz), which means better signal fidelity with lower transmission loss. Figure 2h shows that when the patch is attached to the skin, it can sustain a compression of 73%, a stretch of 26%, and a twist of 57%, which ensures a stable adhesion of the patch on the skin where the epidermis can elastically deform up to 10%-20%. [17,18] It should be mentioned that the sterile dressing with a hollowed-out structure prepared by laser cutting not only improves the stretchability of the dressing ( Figure S5, Supporting Information), but also improves the breathability of the patch. To evaluate biocompatibility and adhesiveness of our epidermal patch, the device was attached to the subject's forearm ( Figure S6, Supporting Information). After 40 min of running at 7 km h −1 , the epidermal patch remained strongly attached to the skin, indicating that the adhesion was not affected by the movement and sweat. Also, there was no obvious skin response in the area where the epidermal patch was attached. And, our results further showed that, due to the good breathability, the contact impedance of an epidermal patch with a hollowed-out sterile dressing recovered much faster than an epidermal patch with a normal sterile dressing after the exercise ( Figure S7, Supporting Information).

Identification of Robust Features and Patterns for Real-Time sEMG Signal Analysis
Previous investigations have shown that time-domain features that reflect the magnitude of sEMG signals represent a linear correlation with muscle strength [19][20][21] and that the frequencydomain features of sEMG signals can reflect the fatigue state of muscle, [22][23][24] but the test conditions vary in different investigations. To identify robust features and patterns for real-time sEMG signal analysis, we carried out grip strength and isometric contraction tests of the right hand with the sEMG epidermal patch attached to the skin over the flexor carpi radialis of the forearm, as shown in Figure 3a. Figure 3b shows the recorded sEMG signals and the extracted time-domain features during intermittent force application in a grip strength test with different force levels, revealing a positive relationship among the force level, the amplitude of the sEMG signal and the magnitude of all the time-domain features. Figure 3c shows the dynamic response of these features when the force is exerted continuously with variations, presenting that the extracted time-domain features follow the same subtle changes as the exerted force, and thus, a proper algorithm design for efficient extraction of these features will support real-time analysis of sEMG signals. Figure 3d and e show the change in time-domain features in response to different exerted forces in grip strength and isometric contraction tests, respectively. First, a linear relationship was observed in both cases with a significance level of P < 0.05. However, a larger slope is obtained in the former, as the contraction of the flexor carpi radial muscle is more remarkable or more motor unit potentials are recruited in this force generation manner. Second, in the same force generation manner, the linearity, which is defined by the coefficient of determination R 2 , of different features is different. In the grip strength test, R 2 varies between 0.78 and 0.98 and is maximized with the feature of logDetect. In the isometric contraction test, R 2 varies between 0.8 and 0.87 and is maximized with the feature of νOrder. This means that the degree of linear correlation between the extracted features and muscle strength is dependent on the force generation manner. In addition, sEMG signals collected at different positions are different under the same muscle movement, so the location of the sEMG epidermal patch attachment is also a major influencing factor on the linear correlation degree. [25] Additionally, this linearity of different features varies from individual to individual or from time to time even for the same individual. Therefore, to avoid these indeterminacies during real-time monitoring, including the individual difference of subjects, the uncertainty of force generation manners, and the variation of patch attached positions, the feature with the highest R 2 was first picked up during a pretest before the real-time monitoring started to realize a high accuracy of muscle strength evaluation at a high speed based on only one feature estimation.
To evaluate the fatigue state of the muscle, the power spectrum of the sEMG signal ( Figure S8, Supporting Information) collected during rest and isometric contraction was analyzed. The change in frequency-domain features, mean frequency (MNF), and median frequency (MDF) with time is shown in Figure 3f. As muscle contraction continues, both features www.advelectronicmat.de continue to decrease (light blue area in Figure 3f). When the subject can no longer hold the load, which indicates the upper limit of endurance, the load is released; the total duration of the isometric contraction before the load is unable to be held is called endurance time. [26,27] Then, the features rapidly return to their level at the resting state (240-270 Hz) in a short period of time, which is referred to as the unloading period (faint yellow area in Figure 3f). As shown in Figure 3g, there is a linear relationship between these two frequency-domain features and the time during isometric contraction; both have a P < 0.05 and a linearity of 0.89 with slope = −0.47, R 2 = 0.89 for MNF and 0.82 with a slope = −0.5, R 2 = 0.82 for MDF. Based on multiple experiments, MNF was selected for the following real-time monitoring on account of its higher linearity and stability during the test. Figure 4a shows the construction of the wireless FPCB band, which can be comfortably attached to the arm (Figure 4b) with reduced signal crosstalk based on the rational modular structural design. To work on the body surface for real-time monitoring, high-quality signal filtering is very important to suppress environmental noise, such as motion artifacts, for high-fidelity signal acquisition. As shown in Figure S9, Supporting Information,

www.advelectronicmat.de
our circuit design provides a better band-pass property for 10-500 Hz, which is the main frequency range of sEMG signal, [28,29] compared to commercial products. Through the integrated system with an sEMG epidermal patch attached to the skin over the flexor carpi radialis muscle of the forearm and a wireless FPCB band worn on the upper arm, the processed sEMG signal was wirelessly transmitted to a LabVIEW interface, and time-and frequency-domain features were extracted in a real-time manner via the embedded algorithm, as shown in Figure 4c. This system can also well serve at other locations for sEMG signal recording from different muscles ( Figure S10, Supporting Information).
Using a pretest, the RMS feature with the highest R 2 = 0.88 (P < 0.05) was picked up for muscle strength evaluation, as shown in Figure S11, Supporting Information. For muscle fatigue evaluation, our results reveal that different loads lead to different fatigue evolution characteristics. As shown in Figure 4d, when the loads of isometric contraction increase from 49 N to 98 N and 147 N, the frequency recorded at the upper limit of endurance (noted the frequency as the F-point) decreases from 152 Hz to 135 Hz and 113 Hz, and the endurance time also reduces quickly from 330 s to 77 s and 54 s, which is consistent with the acknowledgment that a heavier load leads to fatigue more quickly and severely. Furthermore, ten isometric contractions with a load of 98 N were performed at intervals, and the changes in endurance time during these tests are shown in Figure 4e. In the first seven isometric contractions, the intervals were set to 15 min, and the endurance time decreased continuously, which is proposed to contribute to the unrecovered fatigue from the previous isometric contraction.

www.advelectronicmat.de
Then, to confirm this hypothesis, we extended the interval to 20 min in the next three isometric contractions, and a clear recovery of the endurance time was observed. In addition, the ratio of the F-point in each interval to the average frequency of MNF features before the corresponding isometric contraction as a function of time is shown in the inset of Figure 4e. The ratio tends to a specific value of ≈45.8%, which can be used as a predicted reference point to evaluate muscle fatigue. Therefore, combined with the average frequency of MNF features before isometric contraction, the ratio value of 45.8%, and the linear decrease in frequency with time during loading, as revealed in Figure 3g, the F-point can be estimated before every test. Thus, the muscle fatigue state can be evaluated in real-time, which is defined as where MNF real is the extracted MNF feature in real-time. Studies have shown that in addition to relying on the body's own repair mechanism to passively alleviate or eliminate muscle fatigue (noted as passive recovery), appropriate electrical stimulation can serve as external assistance to actively eliminate muscle fatigue (noted as active recovery). [30,31] The normalized endurance time as a function of time under active and passive recovery is shown in Figure 4f. Low-frequency therapy equipment (HV-F311, Omron) was used to relieve muscle fatigue after each isometric contraction during the active-recovery testing. Under the same exercise intensity, active recovery (relieved by HV-F311) is more effective than passive recovery (natural rest) in fatigue elimination, which makes it possible to integrate muscle fatigue monitoring and treatment as feedback for the point-of-care system. Finally, with the strategy we proposed here, the flexible wireless sEMG system demonstrated that it can record sEMG signals and access the status of muscle strength and fatigue in a wearable form and in a real-time manner with typical data shown in Figure 4g. A mobile application was developed to communicate with the flexible wireless sEMG system with these data instantly displayed on a cellphone, as shown in Figure 4h and in Video, Supporting Information.

Conclusion
We present a flexible wireless sEMG monitoring system that can noninvasively monitor muscle strength and fatigue status in real-time. A stretchable sEMG epidermal patch is designed and optimized to achieve conformal contact and low contact impedance with the skin and minimize noise to ensure that the sEMG signal is captured with high fidelity. A wireless FPCB band is designed to process the collected sEMG signal and transmitte the processed data wirelessly to a user interface.
With the epidermal patch attached to the skin and the FPCB band worn on the arm, the integrated system can characterize muscle strength and fatigue in real time by analyzing time-and frequency-domain features with the strategy we proposed here. Such a wearable system with constant muscle status monitoring and evaluation can provide timely advice for efficient and safe training for rehabilitation or fitness, and combined with prompt feedback therapy in the future, chronic pain, musculoskeletal disorders and even disability due to muscle fatigue can be prevented.

Experimental Section
Fabrication of the sEMG Epidermal Patch: The fabrication process began with the preparation of a Si wafer (Tianjin SEMI Tech. Res. Inst., resistivity of 0.009 Ω cm). A PI film with a thickness of 1.6 µm was spin-coated on the substrate and cured at 200 °C for 30 min and 350 °C for 1 h. Then, photolithography (EVG610, EV Group) was used to define the electrodes and interconnect structures, followed by the deposition of titanium (5 nm) and gold (80 nm) by electron beam evaporation and a standard lift-off step. Subsequently, a second PI layer with a thickness of 1.6 µm was spin-coated to encapsulate the entire structure. Then, oxygen plasma etching was conducted to etch the layers of PI to expose the electrodes and connection pads and clear the surrounding redundant PI by using a photolithography-defined Cu film as a mask. Next, the wafer was soaked in FeCl 3 solution for 10 s to remove the Cu film, cleaned with DI water, and dried with N 2 . Then, hollowed-out sterile dressings were prepared by using a laser cutting machine (PLS6MW, Universallaser system), as shown in Figure S12, Supporting Information. Finally, the device was transferred to a water-soluble tape by electrochemical delamination and then transferred to the hollowed-out sterile dressing after being released from the tape.
Contact Impedance Characterization of Au Serpentine Mesh Electrodes: Au serpentine mesh electrodes were attached to the skin over the flexor carpi radialis muscle of the forearm, which was treated with conductive paste (Ten 20, Weaver & Co.) to prepare the skin for attachment, eliminating the influence of hair and impurities on the skin. The impedance was measured between two electrodes by using an LCR meter (LCR6300, Gwinstek), the signal of which was configured with a frequency of 40 Hz and an amplitude of 0.5 Vrms. The measured two-electrode impedance is twice the contact impedance of a single electrode according to the Webster model.
Isometric Contraction and Grip Strength Tests: Isometric contractions are performed without joint motion, and the muscle length remains constant. The subject's hand lifts the load vertically and strictly maintains this position during the entire duration of the test. The load is a canvas bag containing lead blocks. For the grip strength test, a grip dynamometer was used to directly measure the maximum force produced by the subject's forearm muscles. All tests were performed on the right hand, and the site of patch electrode attachment was the skin over the flexor carpi radials muscle.
sEMG Signal Characterization: sEMG signals recorded from the sEMG epidermal patch were characterized by using a low-noise preamplifier (SR560, Stanford Research Systems), which was DC-coupled, filtered (bandpass filter with cutoff frequencies of 10 and 300 Hz), amplified (60 dB), and then sampled at 1 kHz with a data acquisition board (BNC-2120, National Instruments).
Time-and Frequency-Domain Feature Extraction: To analyze the sEMG signals with nonstationary properties, sliding windows were used and time-and frequency-domain features were extracted from each window. For time-domain feature extraction, a sliding window of 0.25 s with a coverage of 50% was adopted, while for frequency-domain extraction, a sliding window of 1 s with a coverage of 50% was adopted.
SNR Analysis: Assessment of the background noise performance of the patch electrode was analyzed through evaluation of the signal-tonoise ratio (SNR). The SNR was determined as the ratio of the root mean square (RMS) value of the raw EMG signal recorded during muscular contraction to the RMS value of the background noise recorded while the muscle was at rest.
The RMS of an sEMG signal segment was calculated as follows: where N is the total number of data points and x n is the nth data point in the sEMG signal segment. The SNR was calculated as follows: where RMS contraction and RMS background are the root mean square values of the sEMG signal during contraction and during the rest period before contraction, respectively.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.