Background

Shoulder kinematics analysis is a booming research field due to the emergent need to improve diagnosis and rehabilitation procedures [1]. The shoulder complex is the human joint characterized by the greatest range of motion (ROM) in the different planes of space.

Commonly, several scales and tests are used to evaluate shoulder function, e.g., the Constant-Murley score (CMS), the Simple Shoulder test (SST), the Visual Analogue Scale (VAS) and the Disability of the Arm, Shoulder, and Hand (DASH) score [2,3,4]. However, despite their easy-to-use and wide application in clinical settings, these scores conceal an intrinsic subjectivity [2,3,4], inaccuracy in approaching diagnosis, follow-up and treatment of the pathologies. Quantitative and objective analyses are rapidly developing as a valid alternative to evaluate shoulder activity level, to gauge its functioning and to provide information about movement quality, e.g., velocity, amplitude and frequency [5, 6]. This interest in the use of measuring systems is growing in many medical fields to record information of clinical relevance. For example, electromyography (EMG), force sensors, inertial measurement units (IMU), accelerometers, fiber optic sensors and strain sensors are employed for human motion analysis, posture and physiological parameters monitoring [7,8,9,10]. From a technological viewpoint, the monitoring of shoulder motion is challenging due to the complexity of joint kinematic which require the development of protocols exploiting sensing technology as much as possible reliable and unobtrusive. In the last years, a great number of human motion analysis systems have been largely employed for objective monitoring. These systems can be classified into two main categories: wearable and non-wearable [11]. The last one includes electromagnetic tracking systems (e.g., Fastrak) [12], ultrasound-based motion analysis systems (e.g., Zebris) [13], stereo-photogrammetric and optoelectronic systems (e.g., VICON, Optotrak, BTS SMART-D) often used as gold standard [14,15,16,17]. These systems based on magnetic field, ultrasound and cameras are effectively suitable for 3D motion tracking and analysis due to their accuracy, precision and reliability [18]. On the other hand, such systems require expensive equipment, frequent calibration and, overall, they restrict measurements in structured environment [19]. Wearable systems overcome these shortcomings and they are a promising solution for continuous and long-term monitoring of human motion in daily living activities. Gathering data in unstructured environment continuously (e.g., home environment) provide additional information compared to those obtainable inside a laboratory [20].

Wearable sensor-based systems, intended for kinematics data extraction and analyses, are acquiring more and more influence in diagnostic applications, rehabilitation follow-up, and treatments of neurological and musculoskeletal disorders [21, 22]. Such systems comprise accelerometers, gyroscopes, IMU, among others [23]. Patients’ acceptance of monitoring systems that should be worn for long-time relies on sensors’ features whose must be lightweight, unobtrusive and user-friendly [24]. The increasing trend to adopt such wearable systems has been promoted by the innovative technology of micro-electro-mechanical systems (MEMS). MEMS technology has fostered sensors’ miniaturization, paving the way for a revolutionary technology suited to a wide range of applications, including extraction of clinical-relevant kinematics parameters. In recent years, there has been growth in the use of smart textile-based systems which integrate sensing units directly into garments [11, 25, 26]. Moreover, in the era of big data, machine learning technical analysis can improve home rehabilitation thanks to the recognition of the quality level of performed physical exercises and the possibility to prevent disorders in patients’ movement [27].

The aim of this systematic literature review is to describe the wearable systems for monitoring shoulder kinematics. The authors want to summarize the main features of the current wearable systems and their applicability in clinical settings and rehabilitation for shoulder kinematics assessment.

Methods

Literature search strategy and study selection process

A systematic review was executed applying the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [28]. Full-text articles and conference proceedings were selected from a comprehensive search of PubMed, Medline, Google Scholar and IEEE Xplore databases. The search strategy included free text terms and Mesh (Medical Subject Headings) terms, where suited. These terms were combined using logical Boolean operators. Keywords and their synonyms were combined in each database as follows: (“shoulder biomechanics” OR “upper extremit*” OR “shoulder joint” OR “scapular-humeral” OR “shoulder kinematics” OR “upper limb”) AND (“wearable system*” OR “wearable device*” OR “wearable technolog*” OR “wearable electronic device*” OR “wireless sensor*” OR “sensor system” OR “textile” OR “electronic skin” OR “inertial sensor”). No filter was applied on the publication date of the articles, and all results of each database were included up to July 2019. After removal of duplicates, all articles were evaluated through a screening of title and abstract by three independent reviewers. The same three reviewers performed an accurate reading of all full-text articles assessed for eligibility to this study and they performed a collection of data to minimize the risk of bias. In case of disagreement among investigators regarding the inclusion and exclusion criteria, the senior investigator made the final decision.

Inclusion criteria were:

  1. i)

    The studies concern wearable systems as a tool to assess upper limb kinematics;

  2. ii)

    The studies used sensors directly stuck on the human skin by means of adhesive, embedded within pockets, straps or integrated into fabrics;

  3. iii)

    Systems intended for motion recognition and rehabilitation;

  4. iv)

    Articles are written in English language;

  5. v)

    Papers are published in a peer-reviewed journal or presented in a conference;

Exclusion criteria were:

  1. i)

    Use of prosthetics, exoskeleton or robotic systems;

  2. ii)

    Wearable system not directly worn or tested on human;

  3. iii)

    The study concerns wearable systems for full-body motion tracking;

  4. iv)

    Shoulder joint is not included;

  5. v)

    Reviews, books.

Data extraction process

Data extraction was executed on 73 articles. Data was extracted on the base of the following checklist: authors, year and type of publication (i.e., conference or full-text); typology, number, brand and placement of the sensors used to measure or track the kinematic of the interested joint, wearability of the system, target parameters with regard to the shoulder; system used as gold standard to assess the wearable systems’ performance; tasks executed in the assessment protocol; characteristics of the participants involved in the study and aim of the study.

Results

The literature search returned 1811 results and additional 14 studies were identified through other sources. A total of 73 studies fulfilled the inclusion criteria (Fig. 1), of which 27% were published on conference proceedings and the remaining 73% on peer-reviewed journal.

Fig. 1
figure 1

PRISMA 2009 flow diagram

Full size image

Three levels of analysis have been emphasized in this survey: A. application field and main aspects covered, B. the typology of sensors exploited to measure kinematic parameters, C. the placement of the single measurement units on the body segment of interest and how sensing modules are integrated into the wearable system from a wearability viewpoint.

Application field

Fifteen out of the 73 studies focused on evaluating upper limbs motion in case of musculoskeletal diseases (e.g., osteoarthritis, rotator cuff tear, frozen shoulder), 26 on neurological diseases and ap-plication in neurorehabilitation (e.g., stroke, multiple sclerosis), 15 on general rehabilitation aspects (e.g., home rehabilitation, physiotherapy monitoring) and 17 focusing on validation and development of systems and algorithm for monitoring shoulder kinematics. Tables 1, 2, 3 and 4 include, for each of the identified application fields, data listed in the previous data extraction process section.

Table 1 Shoulder motion monitoring for application in patients with musculoskeletal disorders
Full size table
Table 2 Shoulder motion monitoring for application in patients with neurological disorders
Full size table
Table 3 Shoulder motion monitoring for application in patients undergoing rehabilitation
Full size table
Table 4 Studies focused on validation/development of systems/algorithms for monitoring shoulder motion
Full size table

Sensing technology

Some studies combined different sensors in their measurements system. The most used sensors are accelerometers, gyroscopes and magnetometers, a combination of them (n = 55) or with other sensors (n = 8), or built-in into other devices (e.g., smartphones, smartwatch) (n = 6); additional studies (n = 4) utilized strain sensors for motion analysis.

B.1 wearable systems based on inertial sensors and magnetometers

An IMU allows estimating both translational and rotational movements. Such sensors comprise gyroscopes that measure angular velocity and accelerometers that measure proper acceleration, i.e. gravitational force (static) and force due to movements (dynamic) [92]. The main limitation of the gyroscopes is the issue bias due to drift. Gyroscopes do not have an external reference, as opposed to accelerometers that use gravity vector as reference; in the orientation estimation, gyroscopes suffer of drift during the integration procedures. To compensate such issue, these sensors are combined with magnetometers that measure magnetic field and use the Earth’s magnetic field as reference. The main limitation of magnetometers is the interference due to the presence of ferromagnetic materials in the surrounding environment [92]. We refer to these hybrid sensors as M-IMU (magnetic and inertial measurement unit). By integrating the information derived from each sensor (i.e., acceleration, angular velocity and magnetic field) through sensor-fusion algorithms, M-IMUs provide an accurate estimation of the 3D-position and 3D-orientation of a rigid body. The upper limb can be modelled as a kinematic chain constituted by a series of rigid segments, i.e., thorax, upper arm, forearm and hand, linked to each other by joints that allow relative motion among consecutive links [17]. In the kinematic chain, the shoulder joint consists of three degrees of freedom (DOFs) correspondent to abduction-adduction (AB-AD), internal-external rotation (IER), and flexion-extension (FLX-EXT) [15, 54, 57, 71, 79]. Shoulder rotations can be described using Euler angles that identify the anatomical DOFs with the roll-pitch-yaw angles [17, 33, 37, 88]. Sensor-fusion algorithms can exploit two main approaches, deterministic or stochastic. The deterministic approach includes the complementary filter that merges a high pass filter for gyroscope data (to avoid drift) and a low pass filter for accelerometer and magnetometer data [64, 82, 90, 92]. The stochastic approach includes the Kalman Filter and its more sophisticated versions [7, 55, 66, 67, 78,79,80, 91, 92]. The Kalman filter (KF) is the most used algorithm to process M-IMU and IMU data due to its accuracy and reliability [15, 38, 54, 75, 83, 93].

Wearable systems based on IMU or M-IMU include a variable number of sensor nodes that, properly distributed on each body segment of interest, provide kinematic parameters such as joint ROM, position, orientation, and velocity. Fifty-one out of the included studies used exclusively IMUs (n = 15) or M-IMUs (n = 36). Systems performances were analyzed in terms of the agreement between results obtained from the M-IMU or IMU-based systems and those collected by a gold standard system. Several types of systems were used as gold standard, such as ultrasound-based system (e.g., Zebris CMS-HS [29]), diagnostic imaging (e.g., Magnetic Resonance [86]), optical-based systems (e.g., VICON [37, 53, 54, 80, 85, 90, 91], BTS Bioengineering [15, 17, 56, 67, 79], Eagle Analogue System [78], Optotrack [16], Optitrack [61], CODA [45, 55]), goniometer [53, 54]. Results from an inertial system were benchmarked against an ultrasound-based reference system, showing a root mean square error (RMSE) of 5.81° and a mean error of 1.80° in the estimation of shoulder angles of FLX-EXT, AB-AD and IER evaluated in the sagittal, frontal and transversal planes, respectively [29]. Accuracy of a protocol based on commercial inertial sensors (MT9B, Xsens) was tested and compared to a VICON system to measure humerothoracic, scapulothoracic joint angles and elbow kinematics [37]. Results demonstrated high accuracy in the estimation of upper limb kinematics with an RMSE lower than 3.2° for 97% of data pairs. A BTS reference system was used to validate accuracy of a wearable system comprised of commercial sensors (Xsens) and results showed a mean error difference of 13.82° for FLX-EXT, 7.44° for AB-AD, 28.88° for IR [15]. In a protocol-validation study, commercial Opal sensors were compared to a BTS system to assess upper limb joint kinematics during simulated swimming movements. Data showed a median RMSE always better than 10° considering movements of AB-AD, IER and FLX-EXT in front-crawl and breaststroke [17]. Opal wearable sensors were compared to optical motion capture systems to estimate shoulder and elbow angles [78, 80]. Planar shoulder FLX-EXT and AB-AD were performed showing an RMSE of 5.5° and 4.4°, respectively [80]; a good correlation between the measurements performed on shoulder motion with the two systems was also found in [78] (no data regarding measurements error were proposed).

Some studies (n = 11) compared data obtained from wearable sensors, custom or commercial, with a gold standard to validate their own sensors data fusion algorithm (for more details see Table 4). Two different algorithms were compared to a customized KF [79]. Comparing the results derived from the BTS system and the inertial-based system (Xsens), the proposed algorithm showed a smaller error than the other two methods for computing shoulder FLX-EXT (RMSE = 2.4°), AB-AD (RMSE = 0.9°), IER (RMSE = 2.9°) [79]. The addition of the magnetometer-based heading correction in the sensor data fusion algorithm was investigated to test the accuracy of an inertial-based motion tracking system using the Optotrak Certus (Northern Digital Inc., Waterloo, ON, Canada) as reference. Results showed a RMSE of 4.9°, 1.2° and 2.9° for shoulder azimuth, elevation and internal rotation, respectively [16].

Four studies used only accelerometers [42, 47, 49, 81]. Systems performance analysis in measurement of arm motion, showed a RMSE lower than 3.5° and 3.68° for shoulder ROM when results from the accelerometers-based systems were benchmarked against a goniometer and commercial M-IMUs, respectively [47, 81]. Evaluation of upper limbs’ physical activity was performed recording data of accelerometers built-in wearable device as ActiGraph (Pensacola, Florida, Model GT3XP-BTLE) to obtain objective outcomes in patients after reverse shoulder arthroplasty [41].

Shoulder ROM has been also estimated by means of a single sensor node which integrated an accelerometer and a magnetometer [69]. Sensor fusion algorithms of accelerometers and magnetometers data provide accurate orientation estimation in static or semi static condition, e.g., in a rehabilitation session in which patients perform slow movements [81]. M-IMUs comprised of a 3D accelerometer, 3D gyroscope and 3D magnetometer are the most appropriate choice for motion tracking either in static that in dynamic condition.

Two accelerometer-based sensors were combined with those built-in a smartphone to realize a smart rehabilitation platform for shoulder home-rehabilitation [68]. Mobile phone or a smartwatch, with their built-in inertial sensor units, were used as mobile monitoring devices [27, 76, 84]. These results give proof of the growing trend in the application of commercial devices in clinical setting for rehabilitation purposes. Data has been processed using machine learning algorithms to extract salient features and for gesture recognition related to shoulder motion. In these techniques, the main steps are the data collection, followed by segmentation process, feature extraction and classification [27, 49]. For instance, the identification of different types of RC physiotherapy exercises has been performed processing data from inertial sensors built-in a wrist-worn smartwatch [27]. Data from inertial sensors built-in a smartphone were benchmarked against a manual goniometer. Angular differences between a machine learning-based application and goniometer measurements resulted less than 5° for all shoulder ROM (i.e., AD, forward FLX, IR, ER) [76].

Two studies combined accelerometer(s) with Optical Linear Encoder (OLE) [68, 84]. An OLE-based system acts as a goniometer providing measures of joint angles. Despite of the simplicity and low cost of the proposed systems, differences in shoulder ROM estimation resulted not negligible when data collected by the wearable systems were compared against an inertial-based motion capture (i.e., IGS-190 [54]) and a fiber optics-based system (i.e., ShapeWrap [71]).

Three studies included EMG sensors in their assessment tool in combination with accelerometers [58], IMUs [59] and M-IMU [39]. EMG sensors placed on the biceps, triceps [59] and deltoid muscles [39] provide additional information about upper limb motor function and shoulder assessment, evaluating muscles activity. Quantification of upper limb motion was executed through a wearable device, MYO armband by Thalamic labs, that combines EMG sensors to record electrical impulses of the muscles [7, 87].

B.2 wearable systems based on strain sensors

Four studies used smart-textiles instrumented by strain sensors with piezoresistive properties to estimate kinematic parameters and to perform motion analysis [8, 11, 46, 56]. Such sensing elements are stretched or compressed during movements of the examined body segments, with consequent variation of their electrical resistance [94, 95]. Using a M-IMU system as reference, accuracy evaluation of a smart-textile with printed strain sensors showed a mean error of 9.6° in planar motions measurements of shoulder joint [11]. Shoulder kinematics was assessed combining a strain sensor for scapular sliding detection with two M-IMUs for HT orientation measure [56].

Piezoresistive strain sensors directly adhered to the skin were used to estimate shoulder ROM; the comparison between reference data from an optical-based system (i.e., Optitrack) and strain sensors showed a RMSE less than 10° in shoulder FLX-EXT and AB-AD estimation [72].

Sensors placement and wearability

Placement of the sensing technology on the body landmarks has shown a heterogeneous distribution linked to the different nature of the employed technology and to the purpose for which monitoring system was designed. With respect to the monitored upper limb, 53 out of the 73 studies included in this review showed a unilateral distribution of the sensing elements while the remaining studies utilized a bilateral placement. Several configurations using different number of sensors and placements have been investigated as reported in detail in each table and Fig. 2.

Fig. 2
figure 2

Placement of sensing units (NOTE One study [90] is not included because the specific position of each sensor nodes is not so clear. Legend: N = number of studies, U = Unilateral, B = Bilateral)

Full size image

Regarding the wearability, we classified the systems in terms of how the sensors were fixed to the human body: i) by adhesive patch, ii) by means of straps or embedded within pocket, iii) the sensing element is physically integrated into the fabric. Four studies did not specify the method of attachment, 12 studies have stuck sensors directly on human skin by means of adhesive patch, 52 studies have attached sensors through straps or embedding them in modular clothing, and 5 studies have integrated sensors directly into garments. For more details refers to Tables 1, 2, 3 and 4.

Discussion

This paper summarizes the main features of wearable systems that have been employed in clinical setting and research field to evaluate upper limb functional performance and particularly for shoulder ROM assessment. Shoulder complex is characterized by the greatest mobility among all human joints and, due to its complexity, reviewed articles evidenced heterogeneity on the more suitable protocol for capturing joint ROM [96].

Wearable technology

Although 73% of the reviewed papers use commercial products for tracking joint angles, many of these personalize the positioning of the sensors, the calibration methodology and the algorithms used to process the recorded data. This customization makes strenuous a direct comparison among protocols, especially if sensing units of different nature (e.g., M-IMU vs. strain sensors) are used to measure the same kinematic parameters, leaving still open the issue of the protocols’ definition with general validity.

About studies using inertial-based motion tracking systems, most in this summary (88%), calibration procedures before data acquisition and data processing represent a relevant issue about accuracy and reliability of the system. Typically, the M-IMUs are attached on the segment of interest to estimate its orientation, so the calibration is necessary to relate sensors’ measurements to movements of the tracked body segment. Sometimes the manufacturer suggests how to perform calibration, e.g., positioning sensors on a flat surface [15, 35] to align coordinate system or assuming static anatomical position [65], as N-pose [79], to compute orientation differences between segments and sensors coordinates in order to obtain sensor-to-segment alignment [56]. Dynamic or functional anatomical calibration has also been performed in some studies, but the sequence of movements executed varied among these [17, 33, 55, 83]. One interesting improvement that may be done to have a positive impact on the accuracy of inertial-based motion tracking systems, is to define a standard set of movements for the initial calibration and a standard method of data processing by which extrapolate kinematic parameters of high clinical relevance.

Some works have reported remarkable results in human motion tracking using e-textile sensors [8, 46]. Technological improvements in the development of conductive elastomers allowed to integrate such strain sensors directly into garments making them comfortable and unobtrusive [11, 56]. Although conductive elastomers ensure flexibility and performances comparable with those of the M-IMU sensors, the main limitations are the hysteresis, uniaxial measurements and non-negligible transient time [56]. Wearable systems based on strain sensors are a promising technology for kinematics analysis that may overcome the main M-IMUs drawbacks, as interferences due to surrounding ferromagnetic materials, gyroscopes’ error drift and long-term use. On the other hand, errors may occur with strain sensors-based systems in the estimation of shoulder kinematics for their inherent hysteresis behaviour.

Among wearable systems reviewed in this summary, differences resulted in terms of sensors typology, number and size, placement, and wearability features. Sensors placement and method of attachment must be carefully investigated as they could influence the outcomes reliability (e.g., effects of soft tissues’ artefacts). Human skeleton is covered by skin tissue and muscles. The combination of skin’s elasticity and muscle activity may cause negative effects in the measurement of the bones’ movement. In studies where M-IMU sensors were used to track shoulder kinematics, soft tissue properties were opportunely included in mathematical models to reduce soft tissue artifacts [79, 85]. The body fat percentage was found the main influencing factor that negatively affects the inertial sensors’ orientation [85]. To reduce such source of error, either when sensors are directly adherent to the skin that embedded in a textile, sensing units should be placed as near as possible to the bone segment to reduce soft tissue artifacts [97, 98]. Wearability is a key factor to consider because it can influence the level of patients’ acceptance [26].There are several relevant requirements that wearable systems must meet to encourage their applications in continuous monitoring of patient status. Indeed, execution of movements, either in home environments or in clinical settings, should not be hindered by measurements systems so they must be non-invasive, modular, lightweight, unobtrusive and include a minimal number of sensors [33, 40, 51, 56, 66, 67, 91]. Most studies have employed magneto and inertial-based tracking systems in which sensors were attached to the upper limb through Velcro straps or including them in modular brace and garments [26, 45, 67, 82, 88].

Upper limb includes the shoulder, elbow and wrist joints (Fig. 3a). Humerus, scapula, clavicle, and thorax constitute the shoulder complex: humeral head articulates in the glenoid fossa of the scapula to form GH joint, the AC joint is the articulation between the lateral end of the clavicle and the acromion process, the SC joint articulates the medial end of the clavicle and the sternum and the functional ST joint allows rotational and translational movements of the scapula with respect to the thorax [96] (Fig. 3b). The ST joint and GH joint act togheter in arm elevation according to scapular-humeral rhythm described in [99]. From a biomechanical point of view, shoulder complexity is justified by the high degree of coupling and coordination between shoulder joints (i.e., shoulder rhythm) and the action of more than one muscles over more than one joints in the execution of a movement. Data extraction of shoulder kinematics is frequently based on movements pattern in the sagittal, frontal and transversal planes, so monitoring of complex movements (e.g., daily activities) in multiple planes, performed through wearable sensors, requires a more stringent evaluation and accurate interpretation. As resulted in the review, the shoulder is generally approximated as a ball-and-socket joint [56]. This assumption provides an approximate representation of the whole shoulder girdle (e.g., it neglects the contribution of scapular movements). A standardized protocol has been proposed (i.e., The ISEO®, INAIL Shoulder and Elbow Outpatient protocol) to improve the performance of M-IMUs in the estimation of scapular kinematics, by locating inertial sensors on the back in correspondence of scapula [33, 35,36,37,38, 40, 65]. An adequate investigation of scapular motions may be beneficial to assess shoulder disorders [100].

Fig. 3
figure 3

a Anatomy of the Upper limb; b Anatomy of the Shoulder complex

Full size image

For long-term monitoring of shoulder kinematics considering also scapular motions, the combination of M-IMUs and smart-textile with embedded strain sensors is a perfect balancing of accuracy, flexibility and wearability (i.e., strain sensors positioned on the scapula could increase the portability and acceptance of the wearable system for long-term monitoring of ADLs) [86].

Applicability in clinical setting and rehabilitation

Alterations in the complex shoulder kinematics can derive by both neurological or musculoskeletal disorders and result in pain and limited movements [68]. Compensatory movements in patients with shoulder disorders are the most common consequential responses to pain or to difficulty in performing free-pain movements. In such situations, information retrieved by posture monitoring may be beneficial in clinical application and rehabilitation [26]. In the last years, the application of wearable devices for gathering motion data outside the laboratory settings is growing. Avoiding complex laboratory set-up, wearable systems employed to assess upper limb kinematics have proven to be a well-founded alternative to obtain quantitative motions parameters. Quantitative outcomes about shoulder motions recorded by wearable sensors are beneficial in clinical practice in terms of time-saving and they are becoming a promising alternative to improve assessment accuracy overcoming the subjectivity of clinical scales. The automatic assessment of motor abilities can also provide therapists a tangible and, therefore, measurable awareness of the effectiveness of the treatment and the recovery path chosen.

In clinical practice, the severity level of patients’ condition with musculoskeletal disease is usually assessed through questionnaire-based scores [36, 42]. Algorithms for kinematic scores computing were developed to evaluate shoulder functional performance after surgery in subjects with GH osteoarthritis and RC diseases, elaborating data obtained from IMU sensors [29, 31]. High correlation (0.61–0.8) between shoulder kinematic scores (i.e., power score, range of angular velocity score and moment score) and clinical scales (e.g., DASH, SST, VAS) was found [31]. Unlike clinical scores, kinematic scores showed greater sensitivity in detecting significant functional changes in shoulder activity at each post-operative follow-up with respect to the baseline status [29, 31]. In a five-year follow-up study, asymmetry in shoulder movements was evaluated in patients with subacromial impingements syndrome. Asymmetry scores, derived from an IMU-based system, showed post-treatment improvements with greater sensitivity than clinical scores and only a weak correlation was found with DASH (r = 0.39) and SST (r = 0.32) [32]. Quantitative evaluation of arm usage and quality of movements in every kind of shoulder impairment contributes to outline a clinical picture about the functional recovery and the effectiveness of the treatment [30, 49]. Using the same number of IMU (n = 3) and the same placement on both humeri and sternum, the shoulder function was evaluated before and after treatment, in patients underwent surgery for RC tear [5, 34]. Results showed significative differences in movements frequency between patients and control group during activities of daily life [5], with limited use of arm at 3 months after surgery [34]. With a bilateral configuration based on 5 IMU, shoulder motion was assessed to extrapolate relevant clinical outcomes about Total Shoulder Arthroplasty (TSA) and Reverse Total Shoulder Arthroplasty (RTSA) [6]. Patients underwent either TSA or RTSA showed shoulder ROM below 80° of elevation, indiscriminately; but, on average, patients treated with RTSA performed movements above 100° less frequently [6]. Objective measurements (i.e., mean activity value and activity frequency) of limb function after RTSA did not show significant improvements 1 year after surgery, despite DASH scores and pain perception have improved compared to preoperative outcomes [41].

In patient with neurological impairments (e.g., stroke), assessments of motor abilities performed through wearable sensors showed a time saving compared to clinical scores (e.g., Fugl-Meyer Assessment Test) measured by the clinician [50, 53]. Data from accelerometers-based systems demonstrated accurate capability in the estimation of clinical scores for quality of movement (e.g., FAS score) and in prediction of shoulder features about shoulder portion of Fugl-Meyer scale with errors near 10% [42, 49]. Generally, the main evaluated features comprise coordination, smoothness, presence of compensatory movements, speed, amplitude of ROM. Quantitative measurements, such as movement time and smoothness, showed a strong correlation with Action research arm test scores in patients after stroke [7]. Spatiotemporal parameters (e.g., ROM, movement time) extracted from inertial sensors’ data provided an accurate evaluation of patients with multiple sclerosis and they distinguished affected and unaffected upper limbs in children with hemiparesis significantly [60, 62].

Digital simulations and virtual reality implementation in upper limb rehabilitation context aim to reproduce accurately limb movements processing data from wearable sensors and give a direct feedback about the adequacy or not of the executed movements [40]. The long-term monitoring, associated with suitable feedback strategy (e.g., visive, auditory, vibrational), can foster the correction of wrong postures [40, 52]. In addition, wearable systems allows a more supervised home-rehabilitation giving substantial improvements to patient healing: total patient involvement in rehabilitation programs can advantage the motor learning process and, at the same time, providing a direct feedback (e.g., visual, auditory) about performance level can increase patient interest and motivation [44, 48]. A new trend is the use of smartphone as monitoring systems or user-interface [53, 76, 84]. Implementation of suitable application (i.e., App) can provide a direct feedback to the patients and therapists about the progress in motor performance [26]. Gathered data could be remotely evaluated by the therapists [64]. Remote monitoring can provide useful information about patients’ status at every stage of rehabilitation pathway and, at the same time, it implies a greater centralization of patients role in the management of their own health associated to a more direct clinician control [101]. A typical architecture of remote monitoring systems includes: i) wearable sensing unit to gather movements data; ii) data storage and management in cloud computing; iii) software to analyse data and extract relevant clinical parameters [58, 66]. This approach implies collection of big amounts of data regarding personal information that requires ethical considerations and the definition of legal responsibility [102].

Most of the reviewed articles limited the application of wearable systems in short-time session for shoulder motion evaluation; only few studies performed longer monitoring periods of ADLs until 7 or 11 monitoring hours of 1 day [5, 6, 34].

Conclusion

This review reveals that wearable systems are becoming an efficient and promising tool to evaluate shoulder health after neurological trauma or musculoskeletal injuries. Wearable systems have the potential to provide quantitative and meaningful clinical information about movement quality and progress in a rehabilitation pathway. The magneto-inertial measurements systems resulted the most used in clinical and research settings, followed by the growing application of smart-textiles for joint angles assessment. Despite of the accuracy of the current wearable systems in shoulder kinematics assessment, additional investigation needs to be executed to ensure long-term applicability in clinical settings and rehabilitation.