Considering the review objective to identify technology suitable for biomechanical data gathering outside a laboratory environment (RQ1), the refined list then resulted in technology with sufficient gold standard validation (RQ2), with evaluation on technology level of remoteness (RQ3) and commercial availability (RQ4). With a considerable number of results and experimental technology under development, there is growing interest and feasibility for research in this field from numerous groups.
With almost 75% of the technology identified as participant wearable technology – body fixed, or shoe worn, this gives rise to both individual needs of data collection methods and types, as well as their range of potential in different uses. The remaining quarter focussed on camera technology with growing prevalence of marker-less MoCap (Kanko et al. 2021) being used, though predominantly still within laboratory settings but with a degree of portability. 75% of the technology identified is focussed on lower limb, as expected, with gait and SPT metrics as prevalent outcomes. Gait SPTs were the most common and therefore, expected to be the most valuable; less valuable were the 2 studies using EMG technology.
4.1 Technology Types
4.1.1 Wearables - IMU
The range of methods, protocols, population groups and overarching contexts used within the IMU studies indicate the extent of their potential as well as supporting previous observations stating their considerable commercial availability. Most studies using IMUs were interested in SPT outcome measures, in agreement with previous evaluation of IMUs for OA research (Kobsar et al. 2020), as a highly useful clinical evaluation tool. Mainstream use of IMU technology was found to be prevalent for studies collecting kinematic parameters in clinical, research and rehabilitation settings (Small et al. 2019), with a small number of IMU studies focussed on predicting joint angles and ROM measures (Oubre et al. 2020). These technologies were all developmental, apart from the use of Shimmer IMUs (Oubre et al. 2020). This indicates a field of research that is cumulatively gaining interest, however, requires increased complexities of computational predictive modelling to produce joint angle data due to the information required from 3-axes data and related biomechanical models when compared to camera-based produced measures.
Most IMUs were demonstrated as belt or strap worn devices, however, emerging in two studies (Ammann et al., 2020; Moon et al., 2017) is a commercially available skin adherent IMU; Biostamp (BiostampRC, M10 Inc. USA), found to be well suited to a variety of uses including potential granular monitoring of gait both inside and outside the clinic (Moon et al. 2017). A variety of experimental systems were found reporting on IMU data collection alongside a mobile application (Aqueveque et al 2020, Li et al., 2020) and visual user-feedback (Bell et al., 2019) with good accuracy and platform outputs for joint angle measurements, demonstrating strong potential for reliable home-based rehabilitation data collection.
Several studies revealed positive patient acceptance and usability for IMU wearables (Papi et al. 2015; Belsi et al. 2016; Bell et al. 2019). However, all IMUs identified in this review were considered suitable for part remote use only, requiring expert support due to specific requirements around their placement or connectivity. Therefore, despite the evidence that they provide solutions to remote data collection protocols, the level of support required for their set up and data acquisition means they are unlikely to be suitable or utilised for long-term at home data collection.
The variability of IMU data collection methods and lack of consensus in terms of standard measures, IMU positioning, or approach offers researchers flexibility for application however it decreases the ability to compare and utilise shared data and results (Aqueveque et al. 2020). Aqueveque provided recommendations for IMU data collection methods for SPT parameters towards the standardisation of IMU data collection methods, however, further pragmatic guidelines using validated methods are required to aid future remote gait assessment where environmental unknowns will complicate data interpretation (Celik et al. 2021).
4.1.2 Wearables - Other
This technology group is dominated by small consumer grade devices, e.g., activity monitors, VR headsets or consumer smartphones containing accelerometers, gyroscopes and cameras. The value of these technologies is their level of remoteness, since many of them are comprised of smartphone technology, as consumer grade devices with simple user interfaces. Six out of the seven ‘fully remote’ technologies were in this category and used smartphone application software. Therefore, this is likely to be a successful route for true remoteness of data collection methods. Though growing in popularity, the current ageing population is prevalent within the OA population requiring rehabilitation, thus usability and reliability may still be limited (Belsi et al., 2016). This is supported by mixed results for SPT measures in the studies of this category. Although good validity was found for SPT outputs from a smartphone application with camera tools (Kim et al. 2015), many unreliable results were found for smartphone 3-axes accelerometers (Silsupadol et al. 2020) and IMU SPT data acquisition (Brabandere 2020). This suggests that camera integrated systems have developed better for end user results. Although consumer grade technology is widely available, only Encephalog was commercially available as a complete solution for researchers gathering metrics of interest (Tchelet et al. 2019). This highlights that further progress is required for many of these fully remote and smartphone-based tools.
Interestingly, studies that used more than one system at one time (Cui et al; Koiler et al) revealed a focus for technology fusion applied to future research data collection as the technology improves. Both studies demonstrated the value of using smartphone application software for data filtering, processing and outputs providing a successful, user-friendly tool for reduction of laboratory-based equipment. Cui et al. (2017) used portable EMG in conjunction with wearable technology to collect kinematic and kinetic data parameters. Though EMG data was collected, the main data used for functional parameters were force sensing and IMU units. Lack of EMG sensor data collection in study results also implies this parameter is only collected alongside other biomechanical parameters and less valuable information.
4.1.3 Insoles Platform
Most technologies in this category measured kinetic and SPT outcomes and took the form of an insole or device placed within a standard or customised shoe in common with widely available commercial products familiar to researchers. Generally, via force or pressure resistive sensors, when force is applied through the plantar surface of the foot (e.g. during stance phase of a gait cycle), a change in resistance allows phases of gait and pressure distribution on the plantar surface of the foot within a shoe/sock/insole to be calculated. This data is then used to determine SPT outputs and can be used in conjunction with kinematic based sensors and outputs (Amitrano et al. 2020; Chen et al. 2016; Cui et al. 2017; Haque et al. 2021) or camera-based technology (Bolaños et al, 2020; Bonnet et al. 2015). This kinetic data alongside kinematic joint angle data and could be paired with smartphone software, similar to other wearable/remote technologies reported (Yang et al., 2019).
Of specific note, are an experimental textile sock for analysis of gait and posture (Amitrano et al. 2020), that could overcome issues associated with insoles since they create an additional layer which can change the distribution of foot plantar loading (Oks et al. 2020). This would be more representative of laboratory-based activities that are usually undertaken barefoot. Also of note is SensFloor (SensFloor Gait, Future Shape GmbH, Germany) (Lanzola et al. 2020), a carpet product capable of recording basic SPT measures through identification of gait phases via the floor sensors which has shown good validity when compared to reference values. The carpet was identified as cost efficient and with good potential for patient rehabilitation monitoring, albeit limited to a defined environment.
The commercially available technology in this category reported were Loadsol (Renner and Queen, 2021) and OpenGo (Moticon Rego AG, Germany) (using a proprietary smartphone application) (He et al., 2019). Both are demonstrated as popular insole devices for gait data collection within the general market showing good usability features. Loadsol (Novel Electronics Inc. USA) insoles demonstrated high correlation values for vertical GRFs when compared to a gold standard instrumented treadmill. When used to detect gait impulse and loading rate, they could successfully identify various comparators such as age groups and degree of walking incline, thus providing an approachable technology for monitoring force and load information for patients’ gait. OpenGo demonstrated effective data acquisition, possible use as a rehabilitative tool with auditory cues and knee adduction moment calculations, a well-known measure for OA disease progression (He et al.; Maly et al., 2013). Auditory feedback was administered via the smartphone application and demonstrated promising use for both rehabilitation training and patient monitoring within a home-based environment. It could also be integrated with other wearable/remote technology tools for rehabilitation and data collection.
RGB-Depth cameras were found in a quarter of the results, many using the Microsoft Kinect skeletal tracker camera solution launched in 2010, with an upgraded version 2 launched in 2014. These cameras have the advantage of operating as a single camera system where multi-camera systems are not feasible (Albert et al. 2020), e.g., clinics, field test conditions and fitness centres, and they do not require body fixed components. Whilst widely used in research, Kinect has had mixed results in comparison with gold standard systems; showing limitations to SPT parameters and 3D kinematic accuracy (Xu et al. 2015; Guess et al. 2017; Xu et al. 2017; Vilas-Boas et al. 2019), but good degrees of accuracy for simple kinematic measures such as 3D ROM and movement velocity (Otte et al. 2016). If combined with other systems, accuracy may be improved (Bonnet and Venture 2015).
Marker-less motion capture software is growing in both research and industry settings and ‘Theia 3D Marker-less’ was found in this review as one of the prominent commercially available systems. By using the optical motion camera set-up, and thus limiting the remoteness of its use, the deep learning algorithm-based system, removes the need however for marker-based set-ups and increases the portability of its use. Though further testing is stated to be required to enable better sensitivities to environmental factors and subject characteristics, the promising comparably accurate results to marker-based motion capture, demonstrate its potential for improving feasibility and sample size of OA patient data collection.
Other commercially available camera and optical tracker components were found to give reliable results only for functional test outputs (Multi-Directional Reach Test, Timed Up and Go) (Moreno et al. 2017) and had limitations based on errors when compared to optical tracking systems (Niechwiej-Szwedo et al. 2018). Only 1 technology incorporating a depth camera was found to be commercially available. The Echo5D is described by the manufacturer Atlas5D (Lincoln, MA) as an ambient measurement system (AMS) comprising a single depth camera and bespoke software for use in the home. Although it was suitable for use in a defined environment, validated use was for a single metric - walking speed - specifically in an MS population (Bethoux et al. 2018), therefore, the use for an OA or other MSK clinical populations may be limited. Although all individual depth camera devices found in the results are commercially available, none, other than the Echo5D were identified as being available as a standalone system specifically for human movement measures. However, they offer adaptive potential for research and data extraction purposes, offering significant and growing potential for the OA researcher.
Both the Kinect and the Nintendo Wii systems were developed primarily as gaming technologies for the entertainment market and were subsequently recognised by researchers for their potential. The original Kinect system (V1 and V2) has now been retired, therefore, products incorporating it are not commercially available. Kinect V1 and V2 were superseded by the launch of the next generation AI Microsoft Azure Kinect sensor released in 2020. Azure Kinect has a suite of applications including the Bodytracking SDK pose estimation model of human movement focussed on non-gaming industries including healthcare, MSK diagnosis and exercise evaluation (Mangal and Tiwari 2021). The Azure Kinect has been reported as demonstrating improved results for spatial measures compared to the original Kinect. Good comparison validity measures were found for finger and thumb joint angles when compared to optical systems (Albert et al. 2020; Zhu et al. 2021) as well as full body tracking for joint angles during treadmill walking (Yeung et al. 2021). However, caution is required with camera viewing angles when using a range of depth sensors for kinematic gait measurements. Considering the limitations, depth cameras are useful as a portable motion capture tool but may still require a small defined environment.
4.2 Location / Application of Technology Use
Freedom to use the technologies in any environment or location and their ability to be applied for a variety of uses without specialist knowledge or support are fundamental to classifying the technologies as suitable for ‘remote’ use or ‘mobile research’. Many of the technologies in the review results lacked a methodology or reference for real-life, real-time assessment of remote or non-laboratory use. Therefore, most were only described as hypothetically suitable for remote use and in some cases no method for remote use was suggested. Equally many studies lacked detail on how data would be recovered and analysed, e.g., in real time or via additional processing. Additional factors such as battery life, range of use, method of data recovery and analysis would also impact usability and availability of the data.
Very few (9%) of the technologies could be determined as fully remote with two-thirds (68%) classified as part remote and the remainder (23%) portable only (Fig. 4). ‘Portable’ technology offers OA researchers additional tools to use in community, clinic, or other settings outside of the traditional laboratory and may still offer new and more cost-effective ways of gathering kinetic and kinematic metrics than those currently available. Therefore, we can conclude that use of technology outside of the laboratory for OA research is both feasible and possible.
Most technologies identified as commercially available (Table 4) were identified as ‘part remote’ and measuring SPT parameters. This highlights that trained users (patients/researchers) have an increasing number of opportunities to collect real world data in a variety of settings and is likely to continue growing and developing. Although small, the identification of ‘fully remote’ technologies could offer researchers the potential to gain new insights into the lives of those with OA through the ability to collect data in an unrestricted and unobserved way, and potentially collection of data over a longer period, enabling emerging patterns to be analysed as opposed to one off laboratory visits.
4.3 Experimental Technologies
The results demonstrated a wide range of technologies under experimental investigation for the gathering of useful OA research data. Whilst some of these (commercially available products) were similar to the IMU or insole platforms, others suggested alternative remote approaches, e.g., a reliable self-measurement hand ROM tool using the Apple iPhone (Alford 2020) as well as a proposed ultrasonic sensor network system for convenient at home gait assessment (Ashhar et al. 2017). These systems are complimented by findings in other work advocating the use of non-contact, low impact sensing such as smartphone apps for the measurement of ankle ROM (Wang 2019) and pulsed Doppler radar (Impulse radio ultra-wide band) to understand human walking patterns (Rana et al. 2019).
It is likely that further rapid developments of smart wearable technologies, AI and other technologies will gain greater focus for gait research resulting in a paradigm shift to acquire complex data employing predictive analytics (Mohan et al. 2021). It is also highly likely that further advances in gaming technology (such as VR) will be better deployed for biomedical use (Bonnechere et al. 2016) leading to further advances in marker-less data capture.
4.4 Limitations of study
A narrative overview of identified technologies was the primary objective of the research, however, it would be beneficial to perform in-depth comparative analysis within technology type/metrics measured. Other technologies that did not fall within the inclusion criteria, due to their size or operating requirements, may still be suitable for remote or community use. Most studies did not include an OA population, an aging population, or a population mixed across the socio-economic divide. Translation to an OA population may be essential for evaluation depending on the research requirements. Most studies did not evaluate intra-operator reliability which contributes to the feasibility of translation of remote technology for use with OA patients. This also affects technology usability, a critical element for successful use of remote technology in research (Lilien et al. 2019).
Quality scoring of technology could have considered advantages and disadvantages based on economic factors, research skills and usability, environmental feasibility technical specifications or cost, (and thus practical elements that may impact usability such as weight, size, battery life, operation range and user interface complexity). Equally no consideration was given to the nature of the data recorded and how this data could be accessed or harvested from the device, or the ease of analysis or interpretation of this data.