An Exoneuromusculoskeleton for Self-Help Upper Limb Rehabilitation After Stroke

Chingyi Nam; Wei Rong; Waiming Li; Chingyee Cheung; Wingkit Ngai; Tszching Cheung; Mankit Pang; Li Li; Junyan Hu; Honwah Wai; Xiaoling Hu

doi:10.1089/soro.2020.0090

. 2022 Feb 14;9(1):14–35. doi: 10.1089/soro.2020.0090

An Exoneuromusculoskeleton for Self-Help Upper Limb Rehabilitation After Stroke

Chingyi Nam ¹, Wei Rong ¹, Waiming Li ¹, Chingyee Cheung ¹, Wingkit Ngai ², Tszching Cheung ², Mankit Pang ², Li Li ³, Junyan Hu ³, Honwah Wai ², Xiaoling Hu ^1,^✉

PMCID: PMC8885439 PMID: 33271057

Abstract

This article presents a novel electromyography (EMG)-driven exoneuromusculoskeleton that integrates the neuromuscular electrical stimulation (NMES), soft pneumatic muscle, and exoskeleton techniques, for self-help upper limb training after stroke. The developed system can assist the elbow, wrist, and fingers to perform sequential arm reaching and withdrawing tasks under voluntary effort control through EMG, with a lightweight, compact, and low-power requirement design. The pressure/torque transmission properties of the designed musculoskeletons were quantified, and the assistive capability of the developed system was evaluated on patients with chronic stroke (n = 10). The designed musculoskeletons exerted sufficient mechanical torque to support joint extension for stroke survivors. Compared with the limb performance when no assistance was provided, the limb performance (measured as the range of motion in joint extension) significantly improved when mechanical torque and NMES were provided (p < 0.05). A pilot trial was conducted on patients with chronic stroke (n = 15) to investigate the feasibility of using the developed system in self-help training and the rehabilitation effects of the system. All the participants completed the self-help device-assisted training with minimal professional assistance. After a 20-session training, significant improvements were noted in the voluntary motor function and release of muscle spasticity at the elbow, wrist, and fingers, as indicated by the clinical scores (p < 0.05). The EMG parameters (p < 0.05) indicated that the muscular coordination of the entire upper limb improved significantly after training. The results suggested that the developed system can effectively support self-help upper limb rehabilitation after stroke. ClinicalTrials.gov Register Number NCT03752775.

Keywords: stroke rehabilitation, robot, pneumatic muscle, neuromuscular electrical stimulation, exoskeleton

Introduction

Upper limb motor deficits are noted in >80% of stroke survivors,^1,2 who require continuous long-term physical rehabilitation to reduce upper limb impairments.^3,4 Restoration of poststroke limb function requires intensive repeated training of the paralyzed limb^5,6 with maximized voluntary motor effort^7,8 and minimized compensatory motions in close-to-normal muscular coordination.^8,9 However, long-term poststroke rehabilitation is challenging because of the expanding stroke population and insufficiency of professional staff worldwide.^10,11 Effective rehabilitation methods with potential for self-help training by stroke survivors are urgently required to improve the independency of stroke survivors and decrease the burden on the health care system. Suitable technologies for these methods are currently lacking.^11,12

Various rehabilitation robots have been developed to assist the labor-intensive process of physical poststroke training, with main advantages of higher dosage and lower cost compared with traditional “one-to-one” manual physical therapy.¹³ However, these robots are large equipment powered by alternating current (AC) that require professional operation in a clinical environment with limited access to outpatients. Mobile exoskeletons are an emerging technology with wearable application. These exoskeletons are powered by portable batteries and have potential for user-independent self-help rehabilitation that can be accessed anytime, even in unconventional environments (e.g., at home).^12,14,15 However, currently available upper limb exoskeletons, which are composed of rigid materials and actuated by electrical motors, are constrained by their heavy weight and low torque-to-weight ratio, which limit their user-independent applications. These exoskeletons require high-power consumption because their actuations must generate sufficient torque to support paralyzed limbs as well as the weight of the system worn on the body. Thus, most exoskeletons require AC supply,^11,15,16 which triggers electrical safety concerns for user-independent usage.

Furthermore, the body/device integration is neither stable nor comfortable in current rigid exoskeletons, with misalignment or migration occurring during repeated practice mainly because of the non-negligible weights mounted onto the paretic limb.^11,14 Misalignments with additional loads deteriorate abnormal muscular coordination in the paralyzed upper limb, which undermines the rehabilitative potential of the aforementioned systems.^17,18 Therefore, most rigid exoskeletons for poststroke upper limb rehabilitation are still used under the close assistance of professionals in clinical environments, and their rehabilitation effects in user-independent operations are unclear.

With the introduction of soft materials in mechanical actuation, soft robotic equipment has been designed using easily deformable materials with light and flexible actuators that conform to human body contours^19–22 so as to achieve superior body/device integration to that provided by rigid robotic equipment. Three main types of actuation systems, namely cable, hydraulic, and pneumatic systems, are used in current wearable soft robots.²¹ Cable systems used cables with desired tension attached to a target limb for flexion/extension.^11,23 The cable-driven upper limb exoskeletons usually have a lightweight design with low inertia in the wearable part accommodating possible joint misalignment between the paretic limb and the exoskeleton.²³ However, the cable is driven by electric motors with gears/pulleys, leading to an increment of complexity and overall weight of the whole assembly.²³ Hydraulic systems are powered by hydraulic pressure, and able to produce greater torque compared with the actuators in cable and pneumatic systems.^11,23,24 However, few hydraulic systems have been developed for upper limb, because they are relatively heavy and complex in the design, requiring additional space to accommodate the fluid and to prevent leakages under pressure.^11,16,23

In contrast, pneumatic systems (pneumatic muscles) are the most commonly adopted actuation for the upper limb.^21,23 Pneumatic exoskeletons have high torque-to-weight ratios because of the low weight of the wearable part actuated by air.^21,25–29 However, pneumatic systems are usually bulky and slow in power transmission from pressure to torque during air inflation by compressors for needed air volume and pressure compared with electrical motor actuation in rigid exoskeleton to achieve equivalent mechanical outputs (e.g., joint torque).^23,30 Large and high-power compressors connected to the pneumatic muscles constrain these devices for user-independent applications.²¹ Thus, a novel lightweight mechanical design is required to achieve optimized body/device integration with fast power transmission, high torque-to-weight ratios, and low-power consumption for user-independent self-help rehabilitation.

Neuromuscular electrical stimulation (NMES), proposed for upper limb rehabilitation,^31,32 can activate the contraction of impaired muscles to generate limb movement^31,32 and effectively enhance the muscle force and sensory feedback for motor relearning after stroke.³³ However, controlling motion kinematics, such as the range of motion (ROM) and trajectory, by using NMES alone is difficult because of the limited stimulating precision in fine motor control.³⁴ Recently, NMES has been combined with mechanical robots in poststroke training.³⁵ The combined NMES-robot treatment is more effective than treatment involving the use of only NMES or only a robot in upper limb rehabilitation, particularly in improving muscular coordination by reducing muscular compensation.³⁶ The integration of NMES into a robot can trigger the biological actuation of target muscles to reduce the demand of mechanical support from the robot part.¹¹ However, little has been done on the integration of NMES with mobile exoskeletons or soft robots.

In this study, we designed a multi-integrated robotic system that combines the NMES, soft pneumatic muscle, and exoskeleton techniques, namely exoneuromusculoskeleton, for upper limb rehabilitation after stroke. Mechanical integration between rigid exoskeleton and pneumatic muscle (i.e., exomusculoskeleton) can enable high torque-to-weight ratios with a compact size and fast power transmission. By combining NMES with the exomusculoskeleton (i.e., exoneuromusculoskeleton), the mechanical scale and power requirement of the entire system can be reduced due to the evoked muscular effort. In addition, NMES and mechanical assistance enable the achievement of close-to-normal muscular coordination with minimized compensatory motions. To optimize therapeutic outcomes, electromyography (EMG) of the paralyzed limb has been used to indicate voluntary intentions³⁷ to maximize voluntary motor effort during practice for better improvements in voluntary motor functions with longer sustainability compared with those with passive limb motions.³⁸

In this study, we designed an EMG-driven exoneuromusculoskeleton to assist the upper limb physical practice at the elbow, wrist, and fingers. The assistive capability of the designed system was evaluated on patients with chronic stroke. The designed system's feasibility of self-help operation and rehabilitation effects were also investigated through a pilot single-group trial.

Methods

The designed exoneuromusculoskeleton (Fig. 1) could be worn on the paretic upper limb of a stroke survivor. The designed system comprised two wearable parts: the elbow (158 g) and wrist/hand (50 g). Both parts were connected to a pump box (80 g) mounted on the upper limb. Moreover, a control box (358 g) that included system control circuits and a rechargeable 12-V battery could be carried on the waist. The developed system can assist a stroke survivor to perform sequential arm reaching and withdrawing tasks, namely (1) elbow extension, (2) wrist extension with the hand open, (3) wrist flexion with the hand closed, and (4) elbow flexion. Real-time control and wireless communication between the control box and a mobile application (app) were achieved on a smartphone through a microprocessor and Bluetooth module.

System control platform

Figure 2a depicts the system control diagram of the exoneuromusculoskeletal system. The system comprised a microcontroller unit (MCU), a musculoskeletal unit, an NMES compartment, a channel switch module, and an EMG preprocessing module. The MCU (PIC18F46K22; Microchip Technology, Inc., Chandler, AZ) coordinated with a musculoskeletal unit, 4-channel NMES stimulator, 4-channel EMG preprocessing, and wireless communication with the developed app through a Bluetooth module (Bluetooth HC-05; JMoon Technologies, New Delhi, India).

FIG. 2. — **(a)** The schematic diagram of the control in the EMG-driven exoneuromusculoskeleton, and **(b)** the controlling workflow of the assistance in phasic and sequential limb tasks. BIC, biceps brachii; ECU, extensor carpi ulnaris; ED, extensor digitorum; FCR, flexor carpi radialis; FD, flexor digitorum; TRI, triceps brachii.

The musculoskeletal unit comprised elbow module and hand module for providing mechanical assistance. Each module included a related musculoskeleton, connected to a respective miniature air compressor (P54A02R; Oken Seiko Co., Ltd., Tokyo, Japan) with a valve and pressure sensor (BMP series; Adafruit, Inc., New York City, NY). The air compressor was used to inflate the musculoskeleton, which would deflate when the valve was opened. The inflated musculoskeleton provided mechanical torque to a joint during extension and it deflated passively during flexion.

The NMES compartment provided electrical stimulation (square pulse with adjustable pulse width of 0–300 μs, 70 V, 40 Hz)³⁹ to the muscle of the biceps brachii (BIC) during elbow flexion, muscle of the triceps brachii (lateral head, TRI) during elbow extension, muscle union of the flexor carpi radialis (FCR) and the flexor digitorum (FD) during wrist flexion with the hand closed, and muscle union of the extensor carpi ulnaris (ECU) and the extensor digitorum (ED) during wrist extension with the hand open.⁴⁰ The activation of the musculoskeletons and NMES were controlled by the EMG signals detected for the BIC muscle, TRI muscle, FCR-FD muscle union, and ECU-ED muscle union in different motion phases. In this study, EMG detection and NMES delivery to a target muscle were performed using a pair of surface electrodes (5 × 5 cm², PALS Neurostimulation Electrodes; Axelgaard Manufacturing Co., Ltd., Fallbrook, CA) connected to an EMG-NMES channel. An electrode pair was placed on the motor point at the muscle belly for achieving effective EMG capture and NMES delivery, as achieved in Muraoka's work.⁴¹ Because of the close anatomical proximity between the FCR and FD muscles and between the ECU and ED muscles, electrode pairs were located in the common area of the motor point of the two muscle bellies of the FCR-FD and ECU-ED muscle unions.⁴⁰ A channel switch circuit was integrated into each EMG-NMES channel and used to alter the functions between the input of the EMG detection and the output of the NMES through the same electrode pair. This circuit also protected the EMG amplification circuit from the high stimulation voltage.^41,42 A reference electrode (2 × 3 cm², Blue Sensor N; Ambu, Inc., Ballerup, Denmark) was attached to the skin surface of the olecranon for reducing the common mode noise.

The EMG signals captured using the surface electrodes were first amplified 1000 times (preamplifier: INA 333; Texas Instruments, Inc., Dallas, TX) and filtered from 10 to 500 Hz. These amplified and filtered signals were then sampled using an analog-to-digital converter (AD73360; Analog Devices, Inc., Norwood, MA) with a sampling frequency of 1000 Hz for each EMG channel. Digitized EMG data were retrieved using a digital signal controller (dsPIC33F, 16bit; Microchip Technology, Inc.) for further manipulation by the MCU. After digitization, the EMG signals were full-wave rectified and moving-averaged with 100-ms window to obtain the EMG levels.

EMG-driven control

EMG-triggered control was adopted in this study.^43,44 In other words, voluntary EMG from a target driving muscle was used to initiate assistance from the developed system. Once the EMG level of a driving muscle or muscle union reached a preset threshold, exoneuromusculoskeletal assistance (i.e., musculoskeleton and NMES) was initiated and continuously provided during an entire motion phase. In each motion phase, a patient was also required to exert the residual voluntary effort, together with the exoneuromusculoskeletal assistance to achieve the desired motion. The controlling workflow of the EMG-driven exoneuromusculoskeleton assisted in phasic and sequential limb tasks is shown in Figure 2b. The assistance scheme of the developed system was defined as follows for the coordinated multijoint limb tasks:

\begin{matrix} E x o - n e u r o - m u s c u l o - s k e l e t a l a s s i s t a n c e = \\ \{\begin{matrix} V_{E M G, T R I} ∙ A s s i s t a n c e_{e l b o w} (N M E S + M u s c u l o s k e l e t o n I n f l a t i o n), i n e l b o w e x t e n s i o n \\ V_{E M G, E C U - E D} ∙ [A s s i s t a n c e_{w r i s t} (N M E S) + A s s i s t a n c e_{f i n g e r s} (N M E S + M u s c u l o s k e l e t o n I n f l a t i o n)], i n w r i s t e x t e n s i o n h a n d s o p e n \\ V_{E M G, F C R - F D} ∙ [A s s i s t a n c e_{w r i s t} (N M E S) + A s s i s t a n c e_{f i n g e r s} (N M E S + M u s c u l o s k e l e t o n d e f l a t i o n)], i n w r i s t f l e x i o n h a n d c l o s e d \\ V_{E M G, B I C} ∙ A s s i s t a n c e_{e l b o w} (N M E S + M u s c u l o s k e l e t o n D e f l a t i o n), i n e l b o w f l e x i o n \end{matrix} \end{matrix}

(1)

where three times the standard deviation (SD) above the EMG baseline in the resting state was set as a threshold level in each motion phase. When the EMG level of a driving muscle or muscle union m reached a preset threshold, the value of $V_{E M G, m}$ was 1 and assistance was simultaneously triggered from both the musculoskeleton and NMES to assist the extension or flexion of the related joint. When the EMG level did not reach the preset threshold, the value of $V_{E M G, m}$ was 0.

The parameter Assistance_elbow (NMES + Musculoskeleton Inflation) is the assistance provided during elbow extension, including the NMES (with a threshold pulse width to evoke visible elbow extension)⁴⁵ applied to the TRI muscle and the mechanical extension torque provided to the elbow joint by the inflated elbow musculoskeleton. The parameter $A s s i s t a n c e_{w r i s t} (N M E S) + A s s i s t a n c e_{f i n g e r s} (N M E S + M u s c u l o s k e l e t o n i n f l a t i o n)$ is the assistance provided during wrist extension with the hand open, including the NMES (with a threshold pulse width to evoke maximal wrist extension with full-finger extension)⁴⁵ applied to the ECU-ED muscle union and the mechanical extension torque provided to the fingers by the inflated hand musculoskeleton.

The parameter $A s s i s t a n c e_{w r i s t} (N M E S) + A s s i s t a n c e_{f i n g e r s} (N M E S + M u s c u l o s k e l e t o n d e f l a t i o n)$ is the assistance provided during wrist flexion with the hand closed, including the NMES (with a threshold pulse width to evoke maximal wrist flexion with full-finger flexion)⁴⁵ applied to the FCR-FD muscle union, and the hand musculoskeleton could be deflated passively during the aforementioned assistance. Most stroke survivors could perform voluntary finger flexion, but most of them cannot extend their fingers.⁴⁶ The residual voluntary effort from the finger flexors of the paretic limb would facilitate the release of the air from the musculoskeleton in deflation. The parameter $A s s i s t a n c e_{e l b o w} (N M E S + M u s c u l o s k e l e t o n D e f l a t i o n)$ represents the assistance provided during elbow flexion, including the NMES (with a threshold pulse width to evoke visible elbow flexion)⁴⁵ applied to the BIC muscle, and the elbow musculoskeleton could be deflated passively during the aforementioned assistance. Meanwhile, the residual voluntary effort from the elbow flexors of the paretic limb would facilitate the release of the air from the musculoskeleton in deflation.

Mechanical structure of the musculoskeletons

The elbow musculoskeleton, which had a length of 24 cm (the detailed dimensions are presented in Fig. 1b), was composed of one piece of pneumatic muscle (polyvinyl chloride [PVC] membrane, 1 mm thick) in the middle and an exoskeletal extension at each end. The musculoskeleton provided extension torque when the pneumatic muscle was inflated by the air compressor. A three-dimensional (3D)-printed plastic (photopolymer) extension (height: 4 cm, width: 6 cm, and thickness: 1 cm) was connected and sealed at each end of the pneumatic muscle. The connections were pressed with two aluminum plates and fastened using rivets. The elbow musculoskeleton was attached to the ventral side of the elbow, with its geometric center located at the joint on the paretic arm around which an elastic sleeve-like bracing (spandex) was wrapped. The musculoskeletons were integrated inside of an elastic bracing (spandex) with the purpose to achieve an average pressure applied to the skin surface from 1279 to 2860 Pa during the inflation and deflation of the pneumatic muscles for the needed mechanical assistance, as well as stable and comfortable wearing experience.^47,48

The hand musculoskeleton (Fig. 1b) comprised five pneumatic finger muscles (PVC membrane, 1-mm thick), one for each digit (width of 1.6 cm for each pneumatic muscle; thumb length = 12.5 cm, index finger length = 17.5 cm, middle finger length = 18.5 cm, ring finger length = 18.0 cm, and little finger length = 14.5 cm), that converged to a 3D-printed exoskeletal connector (photopolymer) at the end of the musculoskeleton, which connected to the air compressor. Each pneumatic finger muscle generated extension torque for the fingers during inflation. The hand musculoskeleton was embedded in an elastic glove-like bracing (spandex) and fixed on the palm during hand opening or closing movements, with the exoskeletal connector located near the bottom of the palm.

The maximal inner pressures of the pneumatic muscles of both musculoskeletons were set at <100 kPa to maintain the stability of the musculoskeletons under repeated inflations and deflations. The proportions and lengths of the musculoskeletons were selected according to mean values of the upper limb anthropometrics for Asian adults.^49,50

Pressure/torque transmission of the musculoskeletons

The pressure/torque transmission properties of the musculoskeletons were quantified by determining the relationship between the inner pressure and extension torque of the musculoskeletons. The pressure/torque transmission rate was determined as the response time of each musculoskeleton for achieving a preset maximal inner pressure. The musculoskeletons for the elbow and hand were evaluated separately in this study.

Elbow module

The experimental setup for measuring the pressure/torque transmission of the elbow musculoskeleton during extension is depicted in Figure 3a. One end of the skeletal extension was fixed on a platform, with half the length of the musculoskeleton falling outside the platform. The configuration in Figure 3a was used to evaluate the extension torque provided to the elbow joint, when the elbow musculoskeleton was attached to the elbow and extended around the center located at the elbow joint.

FIG. 3. — Experimental setup for the evaluation of the pressure/torque transmission properties of the musculoskeleton for the **(a)** elbow and **(b)** hand. MCU, microcontroller unit.

The musculoskeleton was inflated by the compressor with the fully opened valve till the inner pressure reached 95 kPa, at which the elbow musculoskeleton was fully extended to 180°. Then, a weighed loading (sandbag) was hanged to the unfixed end of the exoskeletal extensions. The total hanging weight increased until the musculoskeleton flexed at the joint position with an angle of 170° because most chronic stroke patients with muscle spasticity can reach 170° of the elbow joint passively.⁵¹ When adding the load, the inner pressure was maintained at <100 kPa. The change in the joint angle was measured using a protractor whose midpoint was aligned with the rotation center of the musculoskeleton. The total weight of loading and the corresponding reading of the inner pressure were recorded. The inner pressure of the musculoskeleton was then decreased in steps of 5 kPa with an error within 1 kPa. The measuring scale was set from 5 to 95 kPa without loading in this study because a minimum inner pressure of 5 kPa was required to achieve a joint angle of 180° for the elbow musculoskeleton under free loading. The measurement was repeated three times for each scaling step. A similar evaluation method was adopted in a study on a pneumatic elbow sleeve.⁵² The produced output torque related to each measured pressure was calculated as follows:

T o r q u e = L \cdot c o s 1 0^{\circ} \cdot W

(2)

where $L$ is the length between the axis of rotation and the endpoint of exoskeletal extension with loading and W is the weight of the loading.

The response time of the elbow musculoskeleton during inflation was recorded under free loading. The response time was used to measure the baseline performance of the elbow module before the module was used to provide joint assistance to humans. The musculoskeleton was fixed on a platform with the configuration depicted in Figure 3a. The musculoskeleton was then inflated from an inner pressure of 0 to <100 kPa without external loading with the fully opened valve. The measurements were repeated thrice.

Hand module

The pressure/torque transmission of the hand module was assessed using the middle finger as a representative finger for the measurement of the pressure/torque relationship (Fig. 3b). The middle finger was considered the representative finger because it has the longest extended length and the largest air volume among the fingers in the evaluation of the finger extension torque of the musculoskeleton. The configuration in Figure 3b was used to evaluate the extension torque provided to the metacarpophalangeal (MCP) joint of the middle finger when the musculoskeleton was attached to the palm and extended around the center located near the MCP joint, which was the primary joint when the hand was open.⁵³ The exoskeletal connector of the hand module was fixed on the platform with the pneumatic middle finger, which was present in the palm area inside the platform. The length of the middle finger musculoskeleton inside the platform was 11 cm, which represents the mean length of the wrist joint and MCP joint of the middle finger on the palmar side of Asian adults.⁵⁰

The hand musculoskeleton, including all the pneumatic fingers, was inflated and deflated simultaneously during the experiment. The procedures for the pressure/torque measurement performed on the middle finger musculoskeleton and the response time measurement performed on the hand musculoskeleton having the configuration depicted in Figure 3b were similar to those used to assess the pressure/torque relationship and response time of the elbow musculoskeleton.

Evaluation of joint assistance by the EMG-driven exoneuromusculoskeleton

The assistive capability of the EMG-driven exoneuromusculoskeleton was evaluated on patients with chronic stroke by using four assistance schemes, as presented in Table 1, to understand the different assistance contributions of NMES and the musculoskeleton to the upper limb movements. After obtaining ethical approval from the Human Subjects Ethics Subcommittee of the Hong Kong Polytechnic University, 10 participants with chronic stroke were recruited for the evaluation. The demographic data of the participants in the evaluation are presented in Table 2. Written informed consents were obtained from all the recruited participants in this study. The inclusion criteria were as follows: (1) at least 1 year after the onset of a singular and unilateral brain lesion due to stroke; (2) the spasticity at the elbow, wrist, and fingers was ≤3 as measured by the Modified Ashworth Scale (MAS)⁵⁴; (3) motor impairments in the affected upper limb range from severe to moderate according to the Fugl-Meyer Assessment (FMA; 15 < FMA <45, with a maximal score of 66 for the upper limb)⁵⁵; (4) presence of no visual deficit and the ability to understand and follow simple instructions, as assessed by the Mini-Mental State Examination (MMSE >21)⁵⁶; (5) presence of detectable voluntary EMG signals from the driving muscle on the affected side (three times the SD above the EMG baseline); and (6) presence of passive ROM for the wrist from 45° extension to 60° flexion, presence of passive ROM for the elbow from 30° to 170°, and ability of the MCP finger joints to be passively extended to 170°.

Table 1.

Notations for the Different Assistance Schemes of the Exoneuromusculoskeleton

Notation of assistance schemes	Description
N0M0	No assistance from either the musculoskeleton or the NMES
N1M0	Assistance from the NMES only
N0M1	Assistance from the musculoskeleton only
N1M1	Assistance from both the musculoskeleton and the NMES

Open in a new tab

NMES, neuromuscular electrical stimulation.

Table 2.

Demographic Characteristics of the Participants Recruited for the Range of Motion Measurements (n = 10)

Subjects no.	GenderFemale/male	Stroke typesHemorrhagic/ischemic	Side of hemiparesisLeft/right	Age (years)Mean ± SD	Years after onset of strokeMean ± SD	FMAMean ± SD	MAS elbowMean ± SD	MAS wristMean ± SD	MAS fingerMean ± SD
10	4/6	4/6	6/4	64.1 ± 5.89	5.60 ± 3.98	37.2 ± 11.6	1.46 ± 0.39	1.54 ± 1.19	1.44 ± 0.91

Open in a new tab

FMA, Fugl-Meyer Assessment; MAS, Modified Ashworth Scale; SD, standard deviation.

Evaluation

The evaluation comprised three sessions for the measurement of assistive performance of the developed system for the elbow, wrist, and finger joints. The performance of each joint was evaluated according to the ROM achieved under different assistance schemes (Table 1).

Elbow and wrist sessions

The ROMs related to the elbow and wrist joints were measured separately through motion capturing. In total, 25 spherical reflective markers (12 mm diameter for each) were attached to the skin with double-sided tape according to the upper limb model of the BodyBuilder model (Vicon Motion Systems, Oxford, United Kingdom).⁵⁷ The marker positions were captured through an eight-camera motion system (Vicon Motion Systems), with a sampling frequency of 250 Hz. A Vicon Workstation (Vicon Motion Systems) with 3D reconstruction software (Vicon Nexus and BodyBuilder, Oxford, United Kingdom) was used to anatomically label, filter, and apply the upper limb model.^57–59 The dynamic joint angles during motion were thus obtained during all trials.⁶⁰ The ROMs of the target joints in the extension phase were investigated because most patients with chronic stroke experience impairment in joint extension in the upper limb rather than in flexion.⁴⁶ Most of the poor limb performance (e.g., open hand to grasp) was related to an inability to activate extensor muscles on the upper limb.⁶¹

In the elbow session, the participants wore the elbow module on the affected limb and sat on a 45-cm-high straight-back chair in front of a 72-cm-high table (Fig. 4a). The tested arm was positioned using an upper arm fixer on a lifting shelf placed near the table edge. In the initial position, the forearm was pronated and the shoulder was positioned at 80° vertical abduction with ∼10° flexion. The participants' unaffected hand rested on their thigh. The participants were required to perform a task that involved placing their elbow at ∼90° initially and then extending their elbow to their maximal angle. The participants were instructed to complete the task at their natural speed after the experiment operator provided them an audio starting signal. The trial was completed when they reported that they had achieved their maximal elbow extension or when the inner pressure of the elbow musculoskeleton reached 100 kPa. All the participants reported the completion of the trial before the inner pressure reached 100 kPa within 25 s. The recorded trial lengths were sorted in ascending order from 0 to 21 s and used for comparing the response time in the elbow session because a stable value (defined as <1% change in the maximum value) was achieved within 21 s in all the trials.

In the wrist session, the participants wore the wrist/hand module on the affected limb and sat on the same chair as in the elbow session (Fig. 4b). The table used in the wrist session was also the same as that used in the elbow session. The tested arm was positioned using a forearm fixer and a splint that attached to the table edge. The shoulder was positioned at 20° lateral rotation with ∼30° vertical abduction. The elbow was positioned with a joint angle of 140°, and the unaffected hand rested on the thigh. The participants were required to conduct a wrist task that involved placing the wrist at ∼0° initially (i.e., the neural position for extension or flexion) and then extending the wrist to their maximal angle. After the experiment operator provided an audio starting signal, the participants were required to perform the task at their natural speed. The trial was completed when they reported that the maximal wrist extension was reached or when the inner pressure of the hand musculoskeleton reached 100 kPa. Although the wrist movement was only supported by NMES in this study, four assistance schemes (Table 1) were used in the evaluation of the ROM of the wrist joint because the assistance provided by the exoneuromusculoskeleton for the wrist was triggered in conjunction with that for the fingers in the same motion phases (i.e., wrist extension with the hand open and wrist flexion with the hand closed) during limb practice. The developed system assisted coordinated movements between the wrist and the fingers for the participants in this study. Moreover, a study reported that the wrist ROM can be affected by finger positions.⁶²

All the participants reported the completion of the trials within 16 s. The recorded trial lengths were sorted in ascending order from 0 to 13 s and used for comparing the response time in the wrist session because all trials reached a stable value in <13 s.

The ROM of each joint in the task was measured by comparing the final and initial joint angles. The participants performed the task three times with each assistance scheme (Table 1) in random order. Thus, each participant performed 12 trials in each session. A 1-min rest period was provided between two consecutive trials to prevent fatigue.

Finger session

The finger ROMs were obtained through manual goniometric measurements because the main impairment in the hand for patients with chronic stroke is hand opening, during which fingers flex passively due to spasticity in the finger flexors.⁴⁶ Attaching markers on the spastic finger joints of the recruited participants was not feasible.

In the finger session, the participants wore the wrist/hand module on the affected limb and sat on a 45-cm-high straight-back chair in front of a 72-cm-high table. The tested hand was fixed 12 cm from the table edge in the midline of an acrylic shelf with straps. The participants' other hand rested on their thigh. The wrist was positioned at ∼0° during the evaluation of the MCP, proximal interphalangeal (PIP), and distal interphalangeal (DIP) joints of the index, middle, ring, and little fingers (Fig. 4c). When conducting measurements on the MCP and DIP joints of the thumb, the hand was pronated and the wrist was positioned at ∼0° (Fig. 4d). The aforementioned configurations were used to minimize the gravity effect on the finger movements.

During the measurements, the participants were required to perform their maximal flexion and extension for each joint at their natural speed in a trial. A trial was initiated when the participants reported that they had reached their maximal flexion and was completed when the participants reported that they had achieved their maximal extension or when the inner pressure of the hand musculoskeleton reached 100 kPa. A video camera was used during the measurement, and videos were recorded at a frame rate of 30 fps to confirm the movement timing in each trial. All the participants reported the completion of the trials within 12 s before the inner pressure reached 100 kPa. The finger joint angles were obtained manually by placing the axis of a finger goniometer on the dorsal part of each joint.⁶³ Each joint was measured thrice with the four assistance schemes in random order. Thus, 12 trials were performed for each joint. A 1-min rest period was provided between two consecutive trials to prevent fatigue. The ROM of each measured finger joint was recorded by measuring the angles between the beginning position (i.e., at a maximal flexion angle) and the final position (i.e., at a maximal extension angle) in a trial. In addition to the ROM of each measured finger joint, the ROM of each finger (SUM_ROM) was defined as the sum of the ROMs of its measured joints (i.e., the MCP, PIP, and DIP joints for the index, middle, ring, and little fingers and the MCP and DIP joints for the thumb).

Each participant was required to complete all the elbow, wrist, and finger sessions on the same day. A 20-min break was provided between two consecutive sessions to avoid fatigue. Figure 4e illustrates the evaluation protocol presented with the time line.

Self-help upper limb training assisted by the EMG-driven exoneuromusculoskeleton

A pilot clinical trial with a single-group design was conducted to investigate the feasibility and rehabilitation effects of self-help upper limb training assisted with the EMG-driven exoneuromusculoskeleton. A total of 15 participants with chronic stroke who met the same inclusion criteria as in the aforementioned evaluations were recruited in the pilot trial after obtaining ethical approval from the Human Subjects Ethics Subcommittee of the Hong Kong Polytechnic University. The demographic data of the participants in the pilot clinical trial are presented in Table 3. Written consent was obtained from each participant before clinical trial commencement.

Table 3.

Demographic Characteristics of the Participants Recruited for the Electromyography-Driven Exoneuromusculoskeleton-Assisted Self-Help Upper Limb Training (n = 15)

Subjects No.	Gender	Stroke types	Side of hemiparesis	Age (years)	Years after onset of stroke
Subjects No.	Female/male	Hemorrhagic/ischemic	Left/right	Mean ± SD	Mean ± SD
15	5/10	8/7	7/8	59.8 ± 8.20	6.07 ± 4.28

Open in a new tab

Training protocol

All the participants received self-help upper limb training assisted with the EMG-driven exoneuromusculoskeleton. The training comprised 20 sessions, with a training intensity of 3–5 sessions/week, within seven consecutive weeks.

Before the training, a tutorial session was provided to each participant on the device operation, electrode attachment (the electrode positions were marked on the skin by an experiment operator), wearing skills, and the training protocol. In the first three training sessions, professional assistance was provided in a rehabilitation laboratory at varying levels. The levels of support can be described as follows: (1) the operator supported the participants during the setup and supervised the entire training process in the first session (fully assisted session); (2) the participants mainly completed the session by themselves, with minimum assistance from the operator in the second session (semiassisted session); and (3) the participants completed the third session independently but with close observation by the operator (independent-with-observation session). Additional semiassisted sessions were offered to participants who were not ready for the independent-with-observation session; however, no additional semiassisted session was required by the participants in this study. In the remaining training sessions (i.e., the 4th to 20th sessions), the participants performed the required tasks independently in the laboratory without close supervision of the operator. The operator provided help if required (e.g., if an electrode lead was broken).

In each training session, the participants were seated at a table to maintain a vertical distance of 30–40 cm between the table surface and their shoulder. During the task, the participants' paretic upper limb with the wearable modules was lifted up to 80° vertical abduction of the shoulder with a hanging system (Fig. 5a). A smartphone was positioned on the table and placed in front of the participant with a horizontal distance of 60 cm. The participants were instructed through a visual indication on the mobile screen to perform device-assisted and repeated limb motions, namely (1) elbow extension, (2) wrist extension with the hand open, (3) wrist flexion with the hand closed, and (4) elbow flexion, at their natural speed (for totally 90 min in each session). A 15-min break was provided between two consecutive 30-min practice to prevent muscle fatigue. Figure 5b shows the training protocol presented with time line.

FIG. 5. — **(a)** Experimental training setup of the EMG-driven exoneuromusculoskeleton in the laboratory, and **(b)** the training protocol presented with time line.

Clinical assessments

In this study, the training effects were assessed through clinical assessments of the FMA that the full score is 66 for the upper limb assessment and has been subscaled into shoulder/elbow (42/66) and wrist/hand (24/66),⁶⁴ the Action Research Arm Test (ARAT),⁶⁵ the Wolf Motor Function Test,⁶⁶ the Motor Functional Independence Measure,⁶⁷ and the MAS⁵⁴ at the elbow, wrist, and fingers. These clinical assessments were performed thrice in 2 weeks before the training for detection of the baseline stability. The aforementioned clinical assessments were also performed immediately after the last training session and 3 months after the training by a training-blinded assessor.

Cross-sessional evaluation through EMG

At the beginning of each training session, EMG recordings of the maximum voluntary contractions (MVCs) and a bare arm test (performed in previous studies^39,44) were performed.

Each participant first received an MVC test⁶⁸ for the following target muscle unions or muscles, that is, the ECU-ED, FCR-FD, TRI, and BIC. While conducting the MVC test on the ECU-ED and FCR-FD, participants were seated at a table and the paretic upper limb was placed on the table with the elbow joint extended to an angle of 130°, and the wrist was held by an experimental operator positioned around its neutral position. The finger positions were set by the operator to obtain an angle around 150° at the MCP joints of the index, middle, ring, and little fingers. During the isometric maximum voluntary extension (IMVE) of the wrist and the four fingers, the ECU-ED EMG signals were recorded; and during the isometric maximum voluntary flexion (IMVF) of the wrist and the four fingers, the FCR-FD EMG signals were captured. During the MVC test on the TRI and BIC, the paretic upper limb was positioned with the shoulder abducted at 70° and the elbow flexed at 90°. During the IMVE and IMVF of the elbow, the TRI and BIC EMG signals were recorded, respectively. The MVC test on each target muscle, or muscle union, was repeated twice and the contraction was maintained for 5 s. The variation of maximum EMG amplitude in the two repetitions was required to be within 10%, otherwise, the MVC test would be repeated. The largest EMG amplitude was then selected as the EMG amplitude of MVC for the target muscle union or muscle. A 2-min break was provided between two consecutive contractions to avoid muscle fatigue.

The bare arm test comprised horizontal arm reaching, hand grasping, and withdrawing motions, which were similar to the limb practice motions during the training task. The participants were required to use their paretic limbs (without assistance from the system) to repeat the test three times at their natural speed.

EMG electrodes (2 × 3 cm², Blue Sensor N; Ambu, Inc.) were attached to the skin surface of the aforementioned target muscle unions and muscles (the configuration specified in a previous study^40,69 was used). The collected EMG signals were amplified with a gain of 1000 (amplifier: INA 333; Texas Instruments, Inc.), band-pass filtered from 10 to 500 Hz, and then sampled with 1000 Hz for digitization for offline processing.^69,70 Two EMG parameters were calculated for quantitative session-by-session monitoring of the evolution of the muscle activation and coordination patterns: (1) the normalized EMG activation level of each target muscle and (2) the normalized EMG cocontraction index (CI) between muscle pairs.^71,72 The EMG activation level of a muscle was calculated as follows:

\bar{E M G} = \frac{1}{T} \int_{0}^{T} E M G_{i} (t) d t

(3)

where $\bar{E M G}$ refers to the average EMG envelope value of muscle i, $E M G_{i} (t)$ is the EMG envelope signal obtained after normalization with respect to the EMG MVC value of the muscle, and T is the length of the signal.

The CI between a pair of muscles can be expressed as follows:

C I = \frac{1}{T} \int_{0}^{T} A_{i j} (t) d t

(4)

where A_ij(t) is the overlapping activity of EMG linear envelopes for muscles i and j and T is the length of the signal. An increase in the CI value represents increased cocontraction of a muscle pair (broadened overlapping area), and a decrease in the CI value indicates decreased cocontraction of a muscle pair (reduced overlapping area). To obtain the tendency of the EMG parameters of an individual with normalized values (varying from 0 to 1) and to minimize the variations among different participants, a further normalization was applied to the aforementioned EMG parameters of individual participants with respect to the maximal and minimal values of the participants across the 20 training sessions.^39,69

Statistics

The normality tests on the ROMs, clinical scores, and EMG data were evaluated using the Lilliefors method with a significance level of 0.05.⁷³ The ROMs of the wrist and finger joints exhibited significance in the normality test (p < 0.05), and the ROMs of the elbow, the clinical score, and the EMG data exhibited nonsignificant probabilities (p > 0.05). Kruskal–Wallis one-way analysis of variance (ANOVA) with the Bonferroni post hoc test was used to evaluate the differences in the ROMs of the wrist and finger joints with the four assistance schemes. One-way ANOVA with the Bonferroni post hoc test was used to detect the differences in the ROMs of the elbow with the four assistance schemes, and to evaluate the differences in the clinical assessments across different time points (three pretraining assessments, a post-training assessment, and a 3-month follow-up assessment) and the EMG parameters (i.e., the normalized EMG activation levels and normalized CIs) across the 20 training sessions. The statistically significant level was set as 0.05 in this study. The significance levels at 0.01 and 0.001 are also indicated.

Results