Wearable Movement Sensors for Rehabilitation: A Focused Review of Technological and Clinical Advances

Abstract

Recent technological advancements have enabled the creation of portable, low-cost, and unobtrusive sensors with tremendous potential to alter the clinical practice of rehabilitation. The application of wearable sensors to movement tracking has emerged as a promising paradigm to enhance the care provided to patients with neurological or musculoskeletal conditions. These sensors enable quantification of motor behavior across disparate patient populations and emerging research shows their potential for identifying motor biomarkers, differentiating between restitution and compensation motor recovery mechanisms, remote monitoring, tele-rehabilitation, and robotics. Moreover, the big data recorded across these applications serve as a pathway to personalized and precision medicine. This paper presents state-of-the-art and next generation wearable movement sensors, ranging from inertial measurement units to soft sensors. An overview of clinical applications is presented across a wide spectrum of conditions that have potential to benefit from wearable sensors, including stroke, movement disorders, knee osteoarthritis, and running injuries. Complementary applications enabled by next-generation sensors that will enable point-of-care monitoring of neural activity and muscle dynamics during movement are also discussed.

I. INTRODUCTION

Rapid advancements in electronics and computing have created an opportunity and responsibility¹ to translate these technological advances to the field of rehabilitation. In particular, wearable sensors have emerged as a promising technology with substantial potential to benefit a wide range of individuals, from patients living with mobility deficits to high performance athletes recovering from an injury. Wearable sensors provide precise, quantitative measurements of human movement, enabling tracking of the effects of disease or injury through their influence on the movement system. Importantly, the portability of wearable sensors allow their use in free-living environments, thus providing more ecologic and rich data related to health and disability. Wearable sensors provide an opportunity for the collection of big data across clinical and real-world settings, enabling the growth of personalized and precision medicine².

The field of wearable sensors has seen exponential growth during the last decade; however, widespread clinical use of this promising technology has yet to be realized. Clinical applications of wearable sensors include remote monitoring³, mobile health^3,4, and expansion of health metrics beyond traditional clinical settings⁵. This focused review begins with a summary of the state-of-the-art in wearable movement sensors and their current applications to neurologic and orthopedic rehabilitation, followed by emerging clinical applications. We conclude with an overview of next-generation sensor technologies that expand motion sensing through hybrid sensors, neural interfaces, and soft sensors.

2. LITERATURE SELECTION

To characterize the (i) state-of-the-art, (ii) emerging, and (iii) next generation wearable sensor technologies utilized in the fields of neurologic and orthopedic rehabilitation, a literature search was performed using Medline, Pubmed, and CINAHL databases. Studies published from 2013 – 2018 were the focus of this search. Our search delimiters included studies published in English and studies with adult human participants. Discussion was steered toward stroke and movement disorders to exemplify applications in neurological rehabilitation, and knee osteoarthritis and running to exemplify applications in orthopedic rehabilitation. Sample keywords and their combinations included: sensors, rehabilitation, stroke, Parkinson’s disease (PD), Huntington’s disease (HD), osteoarthritis (OA), and running.

3. REVIEW OF EVIDENCE

Recovery of motor function is a major goal of neurologic and orthopedic rehabilitation. Rehabilitation interventions facilitate motor learning by leveraging repetitive, progressive, and task-specific motor practice provided in sensory-enriched environments⁶—treatment parameters that enhance activity-dependent plasticity in the central nervous system⁷. Precise measurements of motor behavior over different timescales may assist in exploring and optimizing motor learning. Wearable motion sensors enable the objective measurement of body orientation, motion, direction, and physiological state during movement in ecological settings⁸, thus providing clinicians with data that can be used to guide and enhance rehabilitation activities.

State-of-the-art technology

Force-based sensors are commonly integrated with footwear to measure the interaction of the body with the ground during walking⁹. These sensors include load-sensitive switches or force-sensitive resistors that characterize gait based on the configuration of the sensors. A single sensor attached to the heel allows detection of heel strike and heel off phases of gait, while multiple sensors within an insole enable examination of walking strategies¹⁰, center of pressure translations¹¹, and the estimation of vertical ground reaction forces throughout the gait cycle¹². Force-based sensors are also used to drive auditory¹³ and visual¹⁴ biofeedback during gait training^13,14. Limitations of force-based sensors include their susceptibility to mechanical wear over time, limited direct measurements to events during stance phase⁹, and potential drift secondary to humidity and temperature inside the shoe¹⁵ that may influence data quality.

Gyroscopes measure the rate of change of angular motion by detecting the Coriolis forces that act on a moving mass in a rotating reference frame. These forces are proportional to the rate of angular rotation of the limb. Gyroscopes are secured on to body segments in line with the plane of the movement that is being measured¹⁶, and tri-axial gyroscopes allow three-dimensional measurements. Particular strengths of gyroscope sensors are that their measurements are not influenced by gravitational forces¹⁷ and vibrations during heel strike do not distort the signal¹⁸.

Accelerometers measure body movements based on the rate of change of speed. The measurement principle underlying accelerometry is commonly explained by a mass-spring system¹⁹. Based on the displacement of the mass element, the resultant acceleration is derived¹⁹. While there are several classes of accelerometers, the most commonly used in rehabilitation research are strain gauge, capacitive, piezoresistive, and piezoelectric¹⁹. Accelerometers used in rehabilitation commonly have 1 to 3 sensing axes, which allow motion detection in one- to three-dimensional space. Accelerometers are commonly used for continuous monitoring of gait, mobility, and activities of daily living. Accelerometer signals can be used to compute position or velocity; however, drift from integration reduces data quality¹⁸. Additional limitations associated with the use of accelerometers include poor reliability when measuring non-dynamic events²⁰ and the influence of gravity on the acceleration signal⁹. Various signal processing strategies are being developed to improve data quality⁹.

Magnetometers are devices that detect the earth’s gravitation vector. Their measurements provide compass heading information and a reference measure for body orientation relative to gravity⁹. Because magnetometers are insensitive to acceleration during dynamic movements, their use alongside accelerometers allow separation of gravitational components from kinematic acceleration data. Moreover, given the qualities and limitations of gyroscopes, accelerometers, and magnetometers, these sensor types are often combined in self-contained devices called inertial measurement units (IMUs) to optimize measurement capabilities. Force-based sensors offer additional insight into a wearer’s interaction with the environment, and have been used alongside IMUs as well. By and large, limitations in the quality of individual sensor signals can be addressed with advanced processing and intelligent algorithms²¹. The following section overviews applications of these sensors across the neurological and orthopedic domains.

State-of-the-art clinical applications

Wearable sensors are portable, low-cost, and unobtrusive tools that provide objective, quantitative, and continuous information about motor behavior in a range of environments. Clinically, wearable sensors have been utilized for assessment, including the instrumentation of common mobility tests²², identification of pathological movement^23,24, characterization of disease stage²⁵, falls management^26,27, and activity recognition (AR). They have also been used to augment treatments, such as enabling biofeedback-based gait training^12,28,29. This section cites specific examples of these clinical applications (Table 1).

Table 1.

Clinical applications of state-of-the-art technology in select neurological and orthopedic populations.

Clinical application		Sensor (model), associated technology	Findings
Assessment	Clinical Instrumentation	IMU (Physilog®, GaitUp, Lausanne, Switzerland)	High reliability and low measurement error for most measures taken when used for instrumented TUG test in individuals post-stroke31
	Falls Management	Phone-based IMU (Xperia Ray SO-03C, Sony Mobile Communications Inc., Tokyo, Japan)	Able to identify differences in kinematic gait variables in those post-stroke with and without a history of falls35
		IMU (Opal, APDM Inc., Portland, Oregon, USA)	Able to identify differences in dynamic gait stability between stroke cohort and control, and variables that may play a significant role in increased fall risk36
	Identification of Pathologic Motor Features	IMU (Kinesia ONE™, Great Lakes NeuroTechnologies Inc., Cleveland, OH)	High test-retest reliability and sensitivity in measuring bradykinesia, hypokinesia and dysrhythmia in those with PD32
		iPod-based IMU (iPod®, Cupertino, California, USA)	Able to detect significant differences in trunk control during static activities in people with HD compared to controls; found amplitude of thoracic and pelvic trunk movements was significantly greater in participants with HD33
	Activity Recognition	IMU (Physilog®, GaitUp, Lausanne, Switzerland	Excellent ability to classify (90.4%) basic activities common to daily life in individuals post-stroke (e.g. lying, sitting, standing, walking, stair walking and taking an elevator)37
		Step Watch Activity Monitor™ (Orthocare Innovations, Seattle Washington, USA)	Able to characterize activity levels without relying on self-report data or clinician opinion38; assess real-world performance39; and guide community-based treatments using goal setting40 for individuals post-stoke
		Phone-based IMU (Blackberry Z10, Waterloo, Ontario, CAN)	Good sensitivity and specificity in detecting immobile (standing, sitting, laying) versus mobile (walking) states, but poor ability to classify more complicated movements (walking up stairs and other small movements)4 in people with PD
	Characterization of Disease Stage	IMU (Opal inertial sensors, APDM, Inc., Portland, OR, USA)	High correlation between disease severity and turning velocity, duration, and step number in those with PD tracked over 7 days34
Treatment	Biofeedback	Force sensor Smart Shoes, custom made IMUs Smart Pants, custom made	Significant improvements in balance, mobility, strength, and range of motion comparable to improvements seen with therapist cueing only; suggesting the potential use in at home training for those with PD and post-stroke17
		IMU (TecnoBody®, Dlamine BG, ITL)	Improved BBS score, and reduced mediolateral sway during standing in participants with PD who received biofeedback with Gamepad during training42
		Force sensor (not specified) Pager motor (not specified)	Reduced KAM by 14.2% in people with OA43

IMU = Inertial Measurement Unit; TUG = Timed Up and Go Test; PD = Parkinson’s Disease; HD = Huntington’s Disease; BBS = Berg Balance Scale; OA = Osteoarthritis; KAM = Knee Adduction Moment

Stroke

Advanced signal processing approaches have enabled IMU instrumentation of popular clinical tests such as the 10-meter walk test²⁷ and the Timed Up-and-Go test²², providing clinically-relevant data on movement quality in addition to the traditional outcome of “time to complete”. Moreover, advanced AR algorithms have enabled IMU data to be used to identify and quantify gross movements with high sensitivity and specificity³⁰. For example, data extracted from IMUs located in mobile phones have differentiated stroke survivors who are fallers from those who are not based on an estimate of inter-stride variability²⁶. These analyses, however, have been limited when used to quantify more complex movements⁴, motivating further work in this area. Accelerometer-based step activity monitors have also been used to monitor physical activity in the home and the community, providing ecologically-valid mobility data for the development of treatment-based classifications³¹, the assessment of real-world performance³², and to guide community-based treatment programs³³.

Wearable sensors have also enabled novel gait training approaches, such as biofeedback-based interventions. For example, a custom body-worn sensor system comprised of force sensors and IMUs was used to provide kinematic biofeedback during gait training, leading to improvements in balance, mobility, strength, and range of motion that were comparable to the treatment benefits obtained through therapist-directed gait training¹². These results demonstrate the potential for wearable sensors to provide effective gait intervention without direct oversight by a clinician (e.g., in real world settings).

Parkinson’s disease (PD)

Like in stroke, AR algorithms have enabled IMU data to be used to identify pathological motor features characteristic of PD. For example, periods of motor fluctuations between mobile and immobile states (i.e., on-off periods) in levodopa-treated individuals were detected using IMU data analyzed with an advanced AR algorithm³⁴. Other studies have demonstrated how IMUs may be useful in tracking primary physical symptoms of PD, such as tremor³⁵, dyskinesias, and bradykinesia²³, and in tracking disease progression²⁵. For example, IMUs have been used to differentiate between tremor-dominant and non-tremor-dominant patients with PD³⁵. Mancini et. al. tracked features of turning performance (e.g., velocity, duration, and step number) for seven days, revealing a high correlation between disease severity and turning mobility²⁵. Additional studies have shown that IMU-enabled continuous monitoring of baseline gait metrics can predict disease progression and gait decline 1- and 2-years later³⁶. Moreover, a recent large study of 190 patients with PD and 101 age-matched controls shows the feasibility for large-scale clinical trials to use IMUs to robustly track spatiotemporal parameters of gait³⁷.

Like in stroke, sensor-enabled biofeedback interventions have gained popularity as noninvasive training tools in PD rehabilitation. For example, wearable sensors have been used to facilitate the delivery of rhythmic auditory or haptic cues during gait training, an approach shown to enhance motor learning in persons with PD³⁸. Similarly, IMUs have been used effectively to provide haptic and visual biofeedback related to kinematic data during balance and gait training in persons with PD²⁸.

Knee Osteoarthritis

Wearable sensors have been used to understand population-level behavior in individuals with OA. Based on the Osteoarthritis Initiative (OAI), a large epidemiological study on knee OA that utilized wearable sensors to track physical activity in 1,111 adults, only 12.9% of men and 7.7% of women with knee OA met aerobic physical activity guidelines³⁹. The study revealed that in people with knee OA, more sedentary behavior was associated with worse physical function⁴⁰ and greater risk of future functional decline⁴¹. The Multicenter Osteoarthritis Study (MOST), another large epidemiological study enabled by wearable activity trackers, revealed that disease severity and knee pain were not predictive of physical activity levels⁴² and that older adults with high risk of knee OA did not meet Physical Activity Guidelines despite walking ≥ 10,000 steps per day⁴³. These wearable sensor-enabled studies have yielded critical insights into the factors related to reduced physical activity in persons with knee OA and the effects of reduced physical activity on health.

For individuals with knee OA, the most common therapeutic application of wearable sensors is directed toward altering kinematics to reduce knee joint loading during walking. People with medial tibiofemoral OA walk with greater medial compartment loading compared to individuals with knee OA⁴⁴. Greater medial compartment loading is implicated in more rapid disease progression⁴⁵. There is thus significant interest in interventions that can reduce medial compartment loading. The knee adduction moment (KAM) during walking, measured using 3D motion capture, is commonly used as a surrogate for medial compartment loading⁴⁴. There are several examples in the literature of wearable sensors being used to reduce KAM. Dowling et al., for example, developed an active feedback system fitted inside a shoe²⁹. The system delivered haptic feedback if the pressure on the lateral aspect of the shoe exceeded a specific threshold, with the goal being to produce a subtle medial shift in weight bearing to reduce KAM. Use of this innovative biofeedback system led to a mean reduction of 14.2% in KAM. Although encouraging, this study was performed in healthy individuals, in a controlled lab environment, using expensive motion analysis instruments, and with a prototype version of the device. Significant work is needed to translate these systems to free-living conditions among people with knee OA.

Running

Up to 79% of runners are injured in a given year⁴⁶. There is emerging interest in the role that impact mechanics may play in running injuries. Accumulating evidence shows associations between impact loading, as measured with a force plate, and injuries in runners. Indeed, vertical loadrates during the impact phase of running are associated with tibial stress fractures⁴⁷. Runners with diagnosed injuries also have higher vertical loadrates compared to those who have never been injured⁴⁸. Similarly, vertical loadrates are related to other common running injuries such as patellofemoral pain and plantar fasciitis⁴⁸. While it is vertical loadrates that are related to running injuries, peak tibial acceleration during landing has been shown to be related to these loadrates⁴⁹. Therefore, peak tibial acceleration, which can be measured with an accelerometer, has become a surrogate measure for vertical loadrates (Fig. 1A).

Figure 1.

(A) Vertical ground reaction force (dark line) with the tibial acceleration (light line) overlaid to demonstrate the similarity in timing of its peak with the vertical impact peak. (B) Vertical ground reaction force curves of a rearfoot strike (RFS) pattern and a forefoot strike (FFS) pattern. Note the distinct impact peak of the RFS pattern. (C) Tibial acceleration pattern for a RFS runner and a FFS runner. Pattern recognition can be used to distinguish footstrike pattern from these traces. Authors’ original work.

Wearable sensors may also assist in examining other gait characteristics that may contribute to running injuries, such as cadence and strike pattern. Among elite runners, achieving cadences near 180 steps per minute is thought to optimize performance⁵⁰. Increased cadence has other benefits such as reduction in hip and knee energy absorption, patellofemoral stress, and hip adduction^51,52. Further, increasing habitual cadences have demonstrated small reductions in vertical loadrates⁵³. Strike pattern, on the other hand, influences the ground reaction forces applied to the body. Rearfoot strike (RFS) results in a very distinct impact peak in the vertical ground reaction force that is absent during forefoot strike (FFS)⁵⁴ (Figure 1B). Transitioning to a FFS pattern has been shown to resolve chronic patellofemoral pain⁵⁵ as well as chronic anterior compartment syndrome⁵⁶. These distinct impact features can be seen in accelerometer data and can be used to differentiate between a RFS and a FFS pattern (Figure 1C).

Wearable sensors present an exciting opportunity both in the prevention and treatment of running-related injuries by affording the ability to provide real-time feedback to the individual. Many of the commercial IMUs provide information on cumulative loads which can be extremely helpful in preventing overload injuries in runners. Given the range of gait characteristics that can be measured (e.g., strike pattern, lower extremity angles, tibial shock, etc.) a wide variety of gait deviations can be addressed. Once the faulty aspect of gait is identified by the physical therapist, the runner can be instructed in how to alter the gait pattern. The therapist can then set audible signals to remind the patient to attend to their gait when it begins to degrade beyond a certain threshold. Feedback can then be gradually removed with time. Runners can first practice these gait changes in the clinic; however, wearable sensors allow the runner to translate the gait changes from the clinic into their natural running environment. This provides greater ecologic validity to the treatment, as well as can reduce the number of clinical visits needed, thereby reducing overall healthcare costs.

IMUs have important limitations to note when assessing running. Impact magnitudes during running can often exceed 16gs, which is the limit of some commercial devices. Similarly, accelerations during running include high frequency components that require adequate sampling frequencies (500–1000hz). These factors need to be considered when choosing IMU-based devices for running studies.

Clinimetric properties of sensors

The use of wearable sensors to inform neurologic and orthopedic rehabilitation practice warrants careful consideration of their clinimetric properties, which vary among devices⁵⁷, conditions, measures, and environments⁵⁸. Information on reliability, validity, and sensitivity is available for some devices, but not all. For example, wearable sensors used for running have been shown to provide acceptable, valid, and reliable values for some measures⁵⁹; however, IMU-derived measures of tibial acceleration magnitudes and determinations of strike patterns require validation. For PD, a recent review of sensor characteristics concluded that only 9 out of the 73 devices considered could be recommended based on the availability and acceptability of their clinimetric properties⁵⁷. Continued examination of the clinimetric properties of wearable sensor measurements may improve the standardization of data processing, definition of variables, and development of population-specific algorithms^57,58.

Emerging clinical applications of commercially available technology

Emerging clinical applications using existing sensor technologies include their use (i) to identify biomarkers of disease onset and progression, (ii) to differentiate between restitution and compensatory mechanisms of motor recovery, (iii) to provide opportunities for tele-rehabilitation and big data collection, and (iv) in next-generation robotics.

Biomarkers

Tracking disease onset and progression is particularly valuable for those with chronic diseases. As such, there are increasing research efforts directed toward identification of biomarkers. A biomarker is a measurable characteristic that represents either a normal biologic process, pathologic process, or response to an intervention⁵⁸. For example, there is emerging research on identifying motor biomarkers in genetic neurodegenerative diseases. The unobtrusive nature of wearable sensors, coupled with their ability to measure subtle changes in mobility in ecological settings, makes them a highly promising tool for detecting subclinical motor changes that can signal disease onset and progression. Evidence for this emerging application follows.

PD is characterized by dopamine depletion in the basal ganglia, which results in motor disturbances such as tremor, postural instability, bradykinesia, and gait impairment. While most cases of PD are idiopathic, a subset of the population can be explained by genetic factors, of which the most common mutation is leucine-rich repeat kinase 2 (LRRK2) – G2019S⁵⁹. Accelerometers fixed on the low back have been used to identify increased stride time variability⁶⁰, arm swing asymmetry, and trunk axial jerk in asymptomatic carriers of the LRRK2-G2019S mutation (i.e. at-risk for PD) during dual-task walking compared to healthy controls⁶¹.

The identification of motor biomarkers in HD is also an emerging area where wearable sensors have strong potential. HD is an autosomal-dominant neurodegenerative disease that is characterized by a combination of hyperkinetic and hypokinetic motor features⁶². Pharmaceutical and rehabilitative interventions are being developed to delay the clinical onset or slow down progression of HD⁶³. These efforts, however, are attenuated due to limited knowledge of optimal clinical endpoints that are needed for clinical trials. There is emerging evidence for the use of wearable sensors to identify alterations in motor control, which may serve as a worthwhile endpoint. As in PD, an IMU fixed on the low back of individuals with pre-manifest HD and healthy controls was effective in detecting subclinical decrements in the sensory modulation of postural control⁶⁴ and variability in trunk movement during walking⁶⁵. Similarly, wearable iPOD sensors (IMU-based) fixed on the trunk and low back were able to detect abnormal trunk movements in persons with manifest HD compared to controls²⁴. Despite these exciting preliminary findings that support the use of wearable sensors to identify and monitor biomarkers of disease onset and progression in movement disorders, larger, multi-site, and longitudinal studies are needed to catalyze this application.

Motor restitution versus compensation

An emerging clinical application of wearable sensor technologies is in differentiating restitution from compensation when assessing the nature of motor recovery⁶⁶. Restitution refers to the re-appearance of movement patterns that were present prior to the injury, while compensation refers to the emergence of a new set of movement patterns post-injury resulting from substitution or adaptive mechanisms⁶⁶. Elucidation of the mechanisms by which recovery occurs during rehabilitation allows for the development of computational models that can organize biological and behavioral data to inform clinical decision making⁶⁶.

Researchers have also begun to use wearable sensor technologies and analytical techniques to look beyond gross functional and biomechanical recordings, with a focus on the neural control of movement. An example is the use of surface electromyography (sEMG) in the examination of motor modules during functional activities⁶⁷ to identify neuromechanical differences between healthy and pathological movement⁶⁸, evaluate the effects of neurorehabilitation intervention⁶⁹, and assess changes in neuromotor control resulting from robotic intervention⁷⁰. While more research is needed, this is a promising application of commercially available sensor technology. By differentiating between restitution and compensation mechanisms of recovery following neuromotor injury/dysfunction, sEMG analyses have potential to influence the prescription and evaluation of rehabilitative treatments.

Tele-rehabilitation

As our population ages and chronic disease rates and healthcare costs continue to rise, there is a demand for increased access to healthcare services and decreased costs. Tele-rehabilitation is a relatively new branch of telemedicine that prioritizes developing and optimizing telecommunication technologies for rehabilitation services (e.g., evaluation, monitoring, and treatment)⁷¹. The emerging use of wearable movement sensors to enable tele-rehabilitation services is both exciting and timely.

It is not the goal of tele-rehabilitation to replace health professionals; rather, it is to elevate the level of care⁷². The remote monitoring afforded by wearable sensors allows for real-time movement tracking in real world settings. This enables the continuous sampling of activity, rather than a finite series of collections taken during periodic clinic visits. Continuous remote monitoring of movement data may be used by clinicians to map progress and develop personalized interventions. The transmission of these data to clinicians via wireless communication systems could increase patient access to clinicians by bypassing the need to physically travel to a clinic. Similarly, for those with progressive neurological conditions, personalized biofeedback or tele-therapy can be administered in the comfort of home or community settings. These data, coupled with supported human-computer interactions, could also enable an assessment of quality of task practice, as well as patient engagement and compliance with home-based interventions (e.g., exercise programs). There is limited, moderate evidence showing that tele-rehabilitation results in comparable improvements with that of conventional therapy. Additional research is needed to extend the evidence base⁷³. Research is also needed to determine the reliability and validity of the wearable sensor data that might be utilized through tele-rehabilitation approaches⁷³. Furthermore, challenges in privacy and security of information exist, warranting consideration of enhanced security protections based on policy, regulatory⁷⁶, and security protocols⁷⁷.

Robotics

Wearable sensors have played an important role in enabling the development of next-generation assistive and rehabilitation robots. For example, during the last decade, portable rigid exoskeletons have emerged as an exciting tool to enable individuals who cannot walk, to walk again⁷⁴. These powerful systems utilize sensors such as encoders or potentiometers to measure their movement and provide an estimate of limb movement—information that is used to modulate the forces delivered to the wearer. Such sensors, however, are not compatible with a new class of wearable robots that are made from soft and compliant materials⁷⁵. IMUs and force sensors have been shown to be more easily integrated into these soft robotic exosuits, enabling their emergence for a variety of biomedical applications, including reducing the energy used during healthy walking⁷⁶ and running⁷⁷, and restoring more normal walking after stroke^78,79.

Next-generation wearable sensors

Non-invasive monitoring of neural activity

Extending our discussion of sEMG-enabled assessment of motor module analyses, complementary sensor modalities are emerging that enhance movement measurement by monitoring underlying neural control mechanisms. Indeed, motor impairments arise from changes in both neural control and the degradation of mechanical properties of muscles, and the relative contribution of each may be unique for each individual. Inherent to the control of movement are the firings of individual motoneurons that propagate towards the neuromuscular junction where their activation and rate coding regulate muscle contraction force and quality of movement. Deficits in motoneuron control are known to underlie neurological^80,81 and musculoskeletal⁸² conditions, but have been difficult to discern using traditional techniques based on needle EMG recordings⁸³, which are invasive, yield the firings of relatively few motoneurons, and are not practical beyond monitoring highly constrained activities that result from isometric muscle contractions. With the advance of neural sensors and their underlying artificial intelligence concepts, methods for extracting motoneuron firing behavior from noninvasive sEMG during isometric contractions⁸⁴, and more recently during functional activities of everyday life⁸⁵, have been made possible (Figure 2A).

Figure 2:

(A) Schematic of advanced sEMG sensor technology that can extract the firings of individual motor unit activity during functional tasks which can be used to study the underlying mechanisms of human movement in health and disease as well as provide a non-invasive neural interface as a real-time controller of a prosthetic or similar robotic device. (B) A schematic illustrating the use of hybrid sensor technology to autonomously monitor changes in the presence and severity of body bradykinesia in response to dopamine replacement medication in a person with Parkinson’s disease. Authors’ original work.

Recent work in this area has shown that groups of motoneurons are regulated differently when multiple muscles function in synergy to perform a functional task⁸⁶ and that abnormal motoneuron firing behavior underlies motor impairments following stroke^80,81. Assessing motoneuron recruitment patterns across neurological and orthopedic populations could provide valuable insight in determining whether rehabilitation efforts that target abnormalities in movement also have a measurable effect on reversing underlying deficits in motoneuron firing behavior.

Another emerging application of this technology includes assessing activation patterns of motoneurons specific to different training interventions. For example, a recent study showed that subjects were able to selectively activate different populations of motoneurons and thereby exercise components of the muscle with greater fatigue-resistance capabilities⁸⁷. Subjects were able to increase the activation of relatively larger motoneurons that control higher-forces with respect to relatively smaller motoneurons that control lower-forces, in some cases by as much as 40%. Research is continuing to expand on these exciting preliminary findings to provide a basis for new strength training protocols to mitigate muscle weakness in patient populations with muscle atrophy from normal aging, musculoskeletal injury, or long-term bed-rest⁸⁸.

Hybrid Sensors for Monitoring Muscle Activity and Movement

Recent technological advancements have enabled integration of miniaturized sensor components into on-chip electronic systems with ultra-low power consumption. This has fostered the development of “hybrid” wearable sensors that combine in a single encapsulation (i) motion sensing and (ii) EMG sensing of muscle activity. Hybrid sensors can be particularly advantageous for monitoring quality of movement when assessing and treating motor impairments. Indeed, the ability to measure characteristics of both the wearer’s movement and the underlying muscle activity responsible for regulating the movement provides a more holistic assessment of movement dysfunction. Hybrid sensors in use today for movement monitoring include an EMG recording component and a motion component, such as an accelerometer or IMU^89,90

The feasibility of this technology was initially evaluated for automated detection of functional activities of daily living in individuals with stroke⁹¹. Using a minimal subset of four hybrid sensors (combined sEMG and accelerometer sensors located on both upper arms, one forearm, and one thigh), activities related to feeding, grooming, dressing, transferring, locomotion, and toileting were detected with a mean sensitivity of 95.0% and a mean specificity of 99.7%. Significant improvements in sensitivity and specificity resulted when both sEMG and accelerometer data were included, highlighting the value of a hybrid sensor approach for this application. Preliminary work in stroke demonstrated that a hybrid sEMG and accelerometer sensor could differentiate voluntary from spastic contractions⁹². Hybrid sensing has also been shown to be effective for the automated detection of the involuntary movements associated with PD during unscripted activities of daily living^89,90. Indeed, the use of one hybrid sensor (sEMG and accelerometer) per symptomatic limb was sufficient in achieving 94.9% sensitivity and 97.1% specificity for autonomous tracking of tremor and dyskinesia in that limb in response to levodopa treatment.

Hybrid sensors that combine sEMG and IMU sensing hold even greater opportunities for wearable activity monitoring of movement disorders. The availability of angular velocity measurement in such a hybrid sensor proved highly effective in providing the first whole-body bradykinesia detector for PD (with an average accuracy of 95.0% for combined walking and non-walking activities) during unconstrained activities of daily living before and after levodopa therapy⁹⁰ (Figure 2B). Similar technology has been shown to be effective when assessing the quality of movement in stroke⁹³, and to monitor athletic performance for prevention of injury⁹⁴.

Soft sensors

Advances in materials science have enabled explorations into the development of soft sensors and their applications to rehabilitation. Soft sensors can be placed in locations not possible with current movement monitoring devices. For example, stretchy sensors can be placed on the arch of runners with plantar fasciitis. Because runners with plantar fasciitis often have weak intrinsic foot muscles⁹⁵, there is resultant flattening of the arch and increased strain on the plantar fascia. Stretchy sensors can provide feedback to the runner when their arch is lowering too much, reminding them to engage those muscles.

Because placement of sensors could be a source of imprecision in measurement⁹⁶, the prospect of soft, textile-based sensors that could be worn like clothing is very attractive. For example, ultra-thin, ultra-light, and stretchable sEMG sensors that look similar to a temporary tattoo and are mechanically unnoticeable to the user, are being tested for use in evaluating exercise performance during rehabilitation⁹⁷. In addition, elastomeric soft sensors have been integrated into a wearable sensing suit to measure hip, knee, and ankle kinematics⁹⁸. Such sensing garments could be used for continuous kinematic monitoring in the community. More recently, an alternative stretchable, capacitive sensor has been developed with conductive knit fabrics as the electrode layer and a dielectric layer made from silicone elastomer⁹⁹ (Figure 3 top). These sensors can be rapidly customized via a layered manufacturing process using a film applicator and laser cutting has demonstrated scalable, fast, low-cost production and arbitrary shaping of both strain⁹⁹ and pressure sensors¹⁰⁰. The textile-based nature of these sensors make them much more suitable for integration into apparel than existing sensor technologies. It has been demonstrated that these sensors can be integrated into a glove for measuring finger movements⁹⁹ (Figure 3, bottom left) and grip force¹⁰⁰. Additional promising initial results with other textile-compatible sensors demonstrate the ability to measure tension¹⁰¹ and applied pressure¹⁰² in wearable devices. Apart from making the transduction mechanism compatible with apparel, developments have also focused on creating conductive traces within textile materials to eliminate wiring and enable systems to be washable. Figure 3 (bottom right) highlights adaptations of this early work to develop an insole for measuring contact pressure.

Figure 3.

Preliminary work towards textile-based sensors. Top. Capacitive fabric-based stretch soft sensor for measuring joint kinematics. Bottom left, demonstration of sensors in soft robotic glove for measuring finger movement. Bottom right, ongoing work to develop pressure sensing insole using conductive textile traces and electrodes combined with a printed piezoresistive film. Authors’ original work.

Limitations

First, this review does not provide a comprehensive, systematic review of the literature, rather a focused discussion of the current and emerging sensor technologies and their clinical applications. Therefore, studies presenting similar technologies and clinical applications may not have been included. Second, the clinical applications discussed are limited to stroke, PD, HD, OA, and running populations. While these are only a few of the conditions that utilize and can benefit from wearable movement sensors, the conditions chosen illustrate the large spectrum of individuals with varying degrees of capabilities that may benefit from existing and emerging sensor technologies.

Conclusions

The central goal of physical rehabilitation is to facilitate the re-acquisition of movement abilities after injury or onset of disease. Motor behavior is viewed as an output of the movement system based on its encompassing interaction with cardiovascular, pulmonary, endocrine, integumentary, nervous, and musculoskeletal systems¹⁰³, thus movement data has high potential in examining health and disease across systems. Wearable sensors are a promising rehabilitation technology because of their precision, noninvasiveness, and easy deployment compared to other methods. Their complementary measurement of kinematic motion, neural activity, and muscle dynamics offers a targeted approach for assessing and treating a variety of neurologic and orthopedic conditions. Additionally, more widespread monitoring of movement in clinical and ecological settings and across different rehabilitation timescales may serve as a pathway to the development of computational models of recovery and precision medicine. Finally, advancements in materials science are allowing for the development of next-generation sensors that can record biologic movements from device interfaces that are more fully transparent to the wearer.