The use of inertial sensors system for human motion analysis

1Kinematic measurements are used widely by clinicians and researchers alike. Such measures have been used to quantify both normal and pathological movements, quantify the degree of impairment, plan rehabilitation strategies and assess the effect of various interventions.

2Clinical systems of motion analysis are often quick and simple to use; however, such systems often lack valuable kinematic data. Tape measures and goniometers provide information in single planes and only for static positions. Electrogoniometers and inclinometers may offer solutions for more than one plane, as well as provide dynamic data; however, the physical design of such sensors can restrict motion. Therefore, it remains difficult for the clinician to gain information about dynamic three-dimensional movements.

3In contrast, laboratory systems are complex and expensive but are capable of resolving three-dimensional movements. Two laboratory systems commonly found within the literature are electromagnetic systems and video-based optoelectronic systems.

4Electromagnetic tracking devices consist of a source that emits an electromagnetic field, which is used to determine the location and orientation of sensors. Such a system has been shown to be highly reliable and accurate.– A limitation of electromagnetic systems is that they can be affected adversely by the presence of metals, the correction of which is lengthy and complicated, and accuracy can be compromised if the subject moves towards the edge of the defined operating field. This constrains the task that can be analyzed.

5Video based optoelectronic systems are often thought of as the laboratory gold standard. This system utilizes retro-reflective markers visualized by multiple video cameras; such a set-up offers great flexibility, enabling the visualization of multiple body regions. It is possible to track motions in three-dimensions; however, inherent limitations include its complexity and time-consuming operation. Such a system is constrained by the operating environment due to line-of-sight difficulties, which can result in missed data.

6It is therefore clear that a large discrepancy exists between the currently available clinical systems of motion analysis and those used in the laboratory. Recently, new technology borrowed from aerospace, industrial and robotic engineering, appears a promising development in the field of motion analysis. Small, low-powered electromechanical sensors using technologies such as accelerometers, magnetometers and gyroscopes may be able to bridge the gap between large laboratory systems and clinical systems, providing the potential for dynamic three-dimensional motion analysis without the constraints outlined above. Numerous studies have reported using systems based on different types of inertial sensors, including (but not limited to) those based on accelerometers– or gyroscopes;,– however, commonly, these two types of sensors (accelerometers and gyroscopes) are combined for the study of human motions, resulting in increased accuracy.–,–,,–, Due to their small size and portability, these sensors could be an attractive option for ‘in the field’ motion analysis. However, before such technology can be used routinely, reliability and validity needs to be reviewed to compare its performance against a gold standard. Previous reviews have either focused on discussing the advantages and disadvantages of a variety of motion analysis systems or have provided a discussion of possible clinical applications.

7It is therefore the aim of this article to review systematically and appraise critically the literature surrounding the research comparing inertial sensors with accepted laboratory gold standards for measuring human movement (electrogoniometry, optoelectronic systems, electromagnetic systems, etc.). Such information would enable clinicians and researchers to determine whether this technology could be applied to a particular application.

8The literature search began with the retrieval of published reports indexed on health-, biomechanics- and engineering-related electronic databases from MEDLINE, EMBASE, CINAHL, PEDro and SCOPUS. The search was performed to identify all possible studies pertinent to the research question. Search terms included validation, kinematic, inertial tracking devices and inertial sensor. The search was limited to the English language, humans and the last 10 years. We identified papers and relevant conference proceedings, which were hand searched. To be included in this review, studies had to meet the criteria outlined below (based on the PICO model).

9PopulationIncluding reports of the validation for a variety of body sensing regions.

10Including reports of the validation for a variety of body sensing regions.

11InterventionValidation and suitability of inertial sensors for human motion analysis.

12Validation and suitability of inertial sensors for human motion analysis.

13Control/comparisonAccepted methods of human movement analysis (e.g. electrogoniometry, optoelectronic systems, electromagnetic systems, etc.).

14Accepted methods of human movement analysis (e.g. electrogoniometry, optoelectronic systems, electromagnetic systems, etc.).

15OutcomesReliability coefficients and/or a measurement of error.

16Reliability coefficients and/or a measurement of error.

17Studies were excluded if they did not meet the above criteria or if they dealt with drugs or surgery.

18The review was conducted to examine its primary relevance to the measurement of human posture and movement. The title and abstracts identified by the initial search strategy were screened by the first named author to identify potentially eligible reports and retrieve full-text articles. When the title or abstract did not clearly indicate whether an article should be included, the complete article was obtained and reviewed.

19Studies directly comparing inertial sensors with accepted methods of human motion analysis (e.g. electrogoniometry, optoelectronic systems, electromagnetic systems, etc.) formed the basis of this review.

20Data extraction was independently completed by two reviewers, with a consensus opinion adopted to resolve disagreement. A standardized data extraction tool was constructed to identify and detail key features of each study. The reviewers independently piloted the form with a small subset of representative studies to confirm its content. The extracted study details focused on participants’ characteristics, the anatomical region involved, study procedures, type of sensor/portability, the gold standard, biomechanical models and statistical analysis. Primary outcomes of accuracy and/or reliability were also extracted.

21Two independent reviewers completed the quality appraisal, with disagreements resolved by consensus. The studies were critically appraised using the Critical Appraisal Skills Programme Español (CASPe) tool. Appraisal criteria were not applied to the conference proceedings or abstract-only reports because their brevity limited the provision of methodological detail.

22The initial search strategy retrieved 242 articles, which were reduced to 24 relevant to this review. These 24 articles were reviewed in full-text and 10 were excluded for not achieving the necessary criteria (Fig. 1).

23Methodological quality as scored on the CASPe can be found in Table 1. There were no irresolvable disagreements between authors. All 14 studies scored greater than five. This CASPe tool has not been an elimination criterion. The studies included in this review share common threats to validity as most studies score negatively in the same areas. Frequently, a detailed description of the sample was absent and all studies failed to score for blinding or for the calculation of likelihood ratios.

24A summary of the studies comparing motion analysis systems can be found in Table 2. Seven studies report correlation coefficients, six studies reported a coefficient of multiple correlation (CMC) values,,,,, and three of these were focused on the measurement of the trunk (e.g. pelvis, lumbar and/or thoracic).,, One study focused on the lower limb (e.g. hip, knee and ankle). The remaining two studies reporting CMC values measured the upper limb, including the shoulder, wrist and elbow., In the measurement of the trunk, CMC values ranged from 0.829 for rotation of the pelvis to 0.998 for global pelvis angles rotation. Importantly, high CMC values were maintained across a wide range of tasks suggesting a good level of consistency for trunk motion measurement regardless of the gold standard used for comparison. CMC values for upper limb movements were excellent (especially in the elbow), as they were for the majority of lower limb kinematics investigated.

25Thirteen studies reported an error measurement in degrees, with four of these studies reporting the root mean square error. Eight studies reported error measurements for analysis of trunk motion, including the cervical region. Greatest errors were reported by Wong and Wong, which ranged (in absolute values) from 22.8 to 24.7°, for thoracic and lumbar regions, respectively. In contrast, Jasiewicz et al. reported trunk monitoring errors of less than 0.7° for the coronal plane spine, errors of less than 1.2° for the saggital plane measurement and errors of less than 0.9° for rotation of cervical spine. Greater errors were reported by Martin-Schepers et al. for thoracic motion (ranging from 4.3 to 4.5°). Roetenber et al. found error values in the thoracic trunk to range from 2.4 to 2.6°. Plamondon et al. and Goodvin et al. gave angular error results in average values. In this case, the values in the thoracic trunk were in the range of 0.03–0.7° for the lateral bendings. The average error values were always less than 2.2°.

26It appears that errors associated with upper limb movement are more consistent, with a range reported from 2.3° (Ref. 7) to 4.83°. The study of Zhou and Hu confirmed the effectiveness of the Kalman filter, providing results before and after the filter of 14.62–2.13°. However, the lower limb appears relatively inconsistent with errors ranging from 0.49° (Ref. 36) to 8.3°.

27The sensor size ranged from 64×64×25 mm for the larger sensor to 12×12×5 mm for the smaller sensor. All sensors are portable, either wireless or with single wire attachment, however the sensor size is an important consideration depending on the anatomical region to be investigated.

28In this review, 14 studies were identified, which compared directly inertial sensors to any kind of gold standard for human motion analysis (e.g. electrogoniometry, optoelectronic systems, electromagnetic systems, etc.). This review provides the first synthesis of the studies relating to the validity, reliability and accuracy of inertial sensors compared to accepted technology; this gold standard has to be a tool for measuring human movement. It appears that inertial sensors can be applied to many body regions accurately and reliably. The degree of accuracy and reliability displayed suggests that it can be used to measure repeatedly specific motions in varying contexts. The actual degree of reliability is site specific but it is evident that inertial sensors provide a viable option for motion analysis.

29The diversity in the reported studies precludes a simple synthesis of results. A systematic comprehensive analysis of the results was not considered to be appropriate given the diversity among a fairly small number of studies, the varied sample and the heterogeneity of movements studied, the marked variability in the quality of the data, differing methods and statistical analysis and the heterogeneity of results. Most studies were of poor methodological quality with studies related to the development and calibration of the sensors less important to the authors. Under these circumstances, the review comprised a pseudo quantitative analysis of the research available.

30Whether inertial sensor data are reliable enough remains a question that can be answered in the context of the proposed use, with the degree of acceptable measurement variation relating directly to the intended application. Clearly, it is beyond the scope of this review to specify the acceptable limits of reliability for all possible clinical applications of inertial sensors. Following McGinley et al.’s research, we accepted that in most common clinical situations, an error of 2° or less is considered acceptable, as such errors are probably too small to require explicit consideration during data interpretation. Errors of between 2 and 5° are also likely to be regarded as reasonable but may require consideration in data interpretation. We suggest that errors in excess of 5° should raise concern and may be large enough to mislead clinical interpretation. Data from the studies reporting errors revealed that the majority of studies show errors that fall between 2 and 5°. Thoracic and lumbar trunk clearly showed the highest error, although it is noteworthy that some studies reported lower error of 2° for the same variable, suggesting that the lower error is currently achievable.,,,

31The benefits of such a system for the clinician and researcher lie in its inherent portability, accuracy and reliability in the context of proposed use. The sensors are either connected to a personal computer, data logger or may be operated wirelessly providing a wide variety of applications. This freedom enables the system to be used in any environment. Furthermore, these systems can be operated over a range of sampling frequencies enabling tasks of long duration to be studied, such as sitting at a desk at work; very rapid tasks, such as the golf swing can also be studied. Algorithms can be created to provide real-time feedback to the user providing an instant tool to observe and correct motion.

32Common threats to validity are evident throughout the studies and some important aspects should be considered. Appropriate sample composition and inclusion/exclusion criteria should ensure that the range of characteristics of interest in a clinical target population is most likely to be present in a sample, and that the findings can be generalized. It is evident that the studies reviewed failed to describe adequately the baseline characteristics, limiting the reader’s understanding of the threat this poses to external validity. The potential influence of the assessor characteristics on the reliability of inertial tracking devices data received limited focus within the studies in this review. Due to the complexity of understanding three-dimensional kinematics and the necessity for the development of automated algorithms, a level of expertise and experience may be important in identifying and removing sources of error. Furthermore, models often require the use of landmark-specific markers, the placement of which may influence accuracy and reliability.

33Although the majority of studies described the use of standardized protocols, wide variation was apparent in the duration between measurement sessions. The selection of an optimal interval in repeated measures requires consideration of both practical and theoretical issues. In principle, intervals should be designed to minimize fatigue and biological variation associated with repetitive human motion. Artificially-short intervals are often most feasible, yet the presence of visible marker residue could ‘unblind’ a repeat assessment and may influence results.

34Blinding of assessors to prior measurements is typical practice within repeatability studies. Although the potential for assessor bias is less apparent with instrumented measures, it remains a potential factor in some of the studies reviewed. It is particularly important to be blinded prior to measurements in comparative studies.

35A fundamental question in the reliability of inertial tracking devices is whether the measures are reliable enough for clinical decision-making. Although indices, such as the CMCs and other correlation coefficients were commonly reported, it is now well-recognized that, in isolation, correlation indices do not tell us whether the measures are reliable. It provides a measure of similarity between the systems but is not a measure of the difference. To make a proper assessment, reliability measures of both are required. This enables the degree to which a system can resolve measurements of interest to be determined, and should be presented in the data. This is a significant limitation of much of the existing literature, with about half of the papers only reporting errors in absolute terms. It is recommended that future studies investigating reliability of inertial sensors include measures of absolute error.

36The prevalence of reports using the CMC warrants particular attention as the calculation method of the CMC is influenced markedly by the joint range of motion.,,, In those movements which require the contribution of a large number of joints, such as walking, the CMC value is lower. By contrast, in studies utilizing a more specific joint, the CMC is much higher.,,, This is complicated when studying a number of joints and even more complex when studying a global movement, such as gait. Therefore, it is acknowledged that the greater the number of joints involved in the study, the lower the ensuing reliability.

37This review is limited to those articles identified by the search strategies, and study quality was reviewed only by the criterion tool, CASPe. Future studies of the reliability of inertial tracking devices require careful consideration of optimal design to enhance the generalizability of the findings. If the intention is to apply the reliability estimates to clinical populations, then careful attention is necessary to recruit and describe samples, which are representative of the clinical populations of interest. Protocols should consider carefully what standardized measurement interval is most appropriate and minimize predictable sources of assessor bias. Appropriate statistical strategies should include reliability estimates and errors in units of degrees to enhance interpretation. The refinement and optimization of test protocols will help enable the minimization of errors.

38This review concludes that inertial sensors can offer an accurate and reliable method to study human motion, but the degree of accuracy and reliability is site and task specific. They are able to measure differing body regions and overcome the problem of line-of-sight or metallic disturbance associated with other methods. They offer a tool, which has the potential to span many applications in many environments outside of a laboratory and therefore, they warrant further development to continue to improve their systems and their application for human motion analysis.