Exp Brain Res (2014) 232:3861–3872 DOI 10.1007/s00221-014-4069-8
RESEARCH ARTICLE
Quantification of postural stability in older adults using mobile technology Sarah J. Ozinga · Jay L. Alberts
Received: 3 March 2014 / Accepted: 5 August 2014 / Published online: 24 August 2014 © Springer-Verlag Berlin Heidelberg 2014
Abstract Traditional biomechanical systems used to capture kinematic data have shown that declines in postural stability are frequently present in older adults and neurological populations. Recent advances in processor speed and measuring capabilities of on-board electronics within mobile devices present an opportunity to gather kinematic data and apply biomechanical analyses to potentially quantify postural stability. The aim of this project was to determine if the kinematic data gathered using a mobile device were of sufficient quantity and quality to characterize postural stability in older adults. Twelve healthy older adults completed six different balance conditions under altered surface, stance and vision. Simultaneous kinematic measurements were gathered from a three-dimensional motion analysis system and iPad during balance conditions. Correlation between the two systems was significant across balance conditions and outcome measures: peak-to-peak (r = 0.70–0.99), normalized path length (r = 0.64–0.98),
root mean square (r = 0.73–0.99) of linear acceleration, 95 % volume (r = 0.96–0.99) of linear and angular acceleration and total power across different frequencies (r = 0.79–0.92). The mean absolute percentage error metric, used to evaluate time-series measurements point-bypoint, indicated that when measuring linear and angular acceleration, the iPad tracked the motion analysis system with average error between 6 and 10 % of motion analysis measurements across all balance conditions. Collectively, similar accuracy with the iPad compared to motion capture suggests the sensors provide sufficient accuracy and quality for the quantification of postural stability in older adults. The objectivity, portability, and ease of use of this device make it ideal for use in clinical environments, which often lack biomechanical systems.
S. J. Ozinga · J. L. Alberts (*) Department of Biomedical Engineering, Cleveland Clinic, Cleveland, OH, USA e-mail: [email protected]
Introduction
S. J. Ozinga e-mail: [email protected] S. J. Ozinga Department of Chemical and Biomedical Engineering, Cleveland State University, Cleveland, OH, USA J. L. Alberts Center for Neurological Restoration, Cleveland Clinic, Cleveland, OH, USA J. L. Alberts Cleveland FES Center, L. Stokes Cleveland VA Medical Center, Cleveland, OH, USA
Keywords Older adults · Postural stability · Consumer electronics device · Motion analysis systems
In the USA, the 2010 Census reported the greatest number and proportion of people age 65 and older in history: 40.3 million, or 13 % of the total population, with significant growth expected (Bourke and Lyons 2008). Approximately 30–60 % of the older adult population sustain a fall each year (Bronstein et al. 2011), leading to reduced functioning and premature nursing home admissions, morbidity, and mortality (Santos et al. 2011). Postural stability relies on sensory (i.e., visual, vestibular, and somatosensory) and motor components of the central nervous system (Shacham et al. 2007), and cognitive function (Kupsch et al. 2006; Jorgensen 2014). The process of aging is related to a decline in the integrity of these systems, increasing the risk
13
3862
of falling. Quantifying age-related changes in the postural control system may result in reduced fall risk, improved physiological and functional performance, and an improved quality of life (Volkmann et al. 2007). Interestingly, despite the high prevalence of significant consequences and the severity of the functional limitations resulting from balance deficits, there is little agreement among health professionals about the most appropriate tools with which to quantify this impairment. Therefore, the development and refinement of assessment and treatment approaches to improving balance control in older adults is a critical and unmet need. Functional performance-based tests of postural stability have proven helpful in documenting balance status and changes with intervention, while typically rating performance on a set of motor tasks on a three to five point scale or using a stop-watch to track duration that the subject can maintain balance in a particular posture (Horak 1997). Although these tests are practical clinically and have shown a moderate relationship to fall risk (Behrman et al. 2002; Berg and Norman 1996; Giorgetti et al. 1998), inter-rater reliability introduces bias and may mask important features of balance control that vary in each individual (Mannan et al. 2008). Of the many functional balance tests available, the Berg Balance Scale (BBS), Timed Up and Go (TUG), Balance Screening Tool (BST), and Fullerton Advanced Balance (FAB) scale have established reliability and validity with community dwelling older adults (Mannan et al. 2008). Unfortunately, capabilities of such clinical measures to predict falls are weak because of their relatively high false-positive and false-negative predictions, rudimentary scoring, and lack of sensitivity to small changes in balance control over time (Mannan et al. 2008; Mancini et al. 2011; McNames et al. 2004; Blum and KornerBitensky 2008). For example, Downs and colleagues recently found that the BBS has acceptable reliability; however, it may not detect mild, clinically important changes in balance in older adults (Bourke et al. 2007). Also, the BBS may not adequately challenge older adults who live independently, and for those who score highly initially, its use as an outcome measure is compromised (Rose et al. 2006). The TUG and BST do not have the depth of information to discriminate between the various sources of possible impairment (i.e., visual, vestibular, or somatosensory inputs), and meaningful scores cannot be recorded for the TUG where participants are physically unable to rise from a chair or walk independently (Rockwood et al. 2000). Finally, although the FAB includes more challenging tasks that test dynamic balance and challenge sensory inputs, use of test administration to assure reliability requires use of video-taped assessments, limiting its wide-spread use (Mannan et al. 2008). Consequently, there remains a need for a valid and
13
Exp Brain Res (2014) 232:3861–3872
reliable functional balance test that appropriately challenges balance of older adults, which is crucial for clinicians who want to monitor postural stability over time or for individuals with more subtle balance impairments that may be candidates for an intervention. In the last two decades, quantitative assessments have shown promising results in identifying postural instability earlier in the aging process (Laughton et al. 2003; Norris et al. 2005). To date, force platforms and optical motion analysis systems remain the most common methods to quantify balance in nonclinical settings. Force plate analysis quantifies center of pressure (COP) displacement, which represents a weighted average of all the pressures over the surface of the area in contact with the ground. Motion analysis quantifies movement of the whole-body center of mass (COM), which is the weighted average of the COM of each body segment in three-dimensional (3D) space. Movement of COM is an important measure due to the association of falls and decreased trunk mobility (Reilly et al. 2008) and sudden trunk movements (Bloem et al. 2001; Woollacott and Shumway-Cook 2002). Though force platforms and motion analysis systems are sensitive, objective, and have been well-accepted methods of measuring postural stability (Monsell et al. 1997; Simon 2004), clinical utilization is not widespread. Significant cost, space, and time requirements associated with traditional biomechanical approaches act as barriers to large-scale screening and clinical utilization to assess fall risk or evaluate the efficacy of fall prevention programs across large populations or sample sizes (Boers et al. 2001; Simon 2004). Despite these limitations, traditional biomechanical approaches provide the gold standard in terms of postural assessment and play an important role in understanding mechanisms underlying declines in postural stability. An alternative to traditional biomechanical methods is the use of body-worn accelerometers and gyroscopes to obtain kinematic values (van den Bogert et al. 1996; Dai et al. 1996; Luinge et al. 1999; Tong and Granat 1999; Veltink 1999; Veltink et al. 1996). Accelerometers are used to quantify trunk movements in medial–lateral (ML) and anterior–posterior (AP) directions during stance and gait tests (Gill et al. 2001; Whitney et al. 2011; O’Sullivan et al. 2009). Accelerometers have been tested for both precision and accuracy (Moe-Nilssen 1998) and have been shown to have the ability to significantly identify differences between test conditions, young and older adults, and fallers and nonfallers (Kamen et al. 1998; Cho and Kamen 1998; Moe-Nilssen and Helbostad 2002; Yack and Berger 1993; Mayagoitia et al. 2002a). Measurement of trunk rotation (TR) with the use of gyroscopes during stance and gait tasks can detect changes in postural stability with age (Gill et al. 2001), can identify pathological balance control in
Exp Brain Res (2014) 232:3861–3872
individuals with a vestibular deficit (Allum et al. 2001), and can track movements in balance control over time following recovery from a vestibular deficit (Allum and Adkin 2003). Combining accelerometers and gyroscopes in one device increases the accuracy in quantifying COM acceleration in clinical applications compared to devices composed of sole accelerometers or gyroscopes (Wu and Ladin 1996). With use of inertial sensors (i.e., accelerometers and gyroscopes), metrics comprising a subject’s movement of COM acceleration have been shown to have the ability to significantly identify increased postural sway in older adults (Rispens et al. 2014; Galan-Mercant and Cuesta-Vargas 2014), patients with Parkinson’s disease (PD) (Palmerini et al. 2011), and those at risk for developing PD (Maetzler et al. 2012). Despite the advantages of using accelerometer and gyroscope technology, they have not been embraced clinically. Possible reasons include increased setup time or offline data processing requirements, both of which disrupt the ever constricted clinical time and workflow. If proven accurate, the use of mobile devices that contain inertial sensors to quantify postural sway may provide a practical and low-cost alternative to clinical ratings, force plates, and optical motion analysis systems (Pohl et al. 2003; Pickering et al. 2007; Mayagoitia et al. 2002b) in the quantification of postural stability. To bridge the gap between subjective performance-based measures and expensive and sophisticated biomechanical assessment methods related to postural stability analysis, the aim of the present study was to determine if the kinematic data measured by the native accelerometer and gyroscope within the iPad (third generation) was of sufficient quantity and quality to accurately quantify postural stability in older adults. To verify the accuracy of the iPad measurements, a 3D model will be utilized to validate the use of the Cleveland Clinic Balance Assessment App (CC-BApp) within the iPad to assess postural stability across six balance conditions of varying difficulty against clinically valid and reliable measures of 3D motion analysis.
Materials and methods Participants A single-center, validation study was completed in order to validate the accuracy of an inertial sensor system within a mobile application designed to assess postural stability. Twelve healthy community-living older adult subjects (5 male and 7 female) aged between 60 and 85 (mean age ± sd, 68.3 ± 6.9) completed all study conditions (see Table 1 for participant demographics). All data were collected during one visit to Cleveland Clinic Biomechanics Laboratory. Inclusion criteria included: age (60 and older),
3863 Table 1 Summary of subject characteristics Subject
Gender
Age (years)
BMI (kg/m2)
1 2 3 4
M F F F
69 66 70 60
31.7 23.7 28.8 22.9
5 6 7 8 9 10 11 12
M M F M F F F M Mean
69 81 69 60 79 68 60 67 68
14.3 24.8 27.3 32.7 29.3 22.1 45.7 31.1 27.9
7
7.6
SD
ability to demonstrate an understanding of the study, ability to follow two-step commands, ability to walk comfortably unaided, and ability to stand comfortably for at least 5 min unaided. Primary exclusion criteria included: the presence of diagnosed dementia of any etiology, history of musculoskeletal or neurological injury that currently impairs postural stability, and a premorbid condition that affects balance. All participants were informed of the experimental procedures, which had received ethical committee approval from the institutional review board and provided witnessed written consent prior to testing. Data collection Biomechanical data were collected using a motion analysis system (Motion Analysis Corporation Eagle System; Santa Rosa, CA) with eight infrared Eagle digital cameras. For collection of iPad data, an in-house application was written to acquire and store accelerometer and gyroscope data during balance testing. In order to compare postural stability parameters obtained using inertial sensors within the iPad and the motion analysis system, data were simultaneously collected from both systems for all subjects during data collection. To synchronize the data collection between the two systems, an Arduino Pro Mini 3.3v and LED light were used to align the balance data temporally. Upon data collection, an output signal from the iPad audio headphone jack triggered the LED light to illuminate. The light provided the signal to the motion analysis digital cameras indicating that the iPad had begun collecting data. Each trial consisted of 60 s of data collection; the motion analysis system recorded 90-s trials to ensure capturing of movement during iPad collection time.
13
3864
Exp Brain Res (2014) 232:3861–3872
Fig. 1 Illustration of experimental paradigm and measurement setup
Procedure Participants completed a series of postural stability tests while the iPad was securely attached via a clip to a belt around their waist. The iPad was placed as close as possible to the second sacral vertebra to approximate body COM during data collection with CC-BApp. Reflective markers, using the Helen Hayes marker-set (Kadaba et al. 1990; Davis et al. 1991), were attached to 24 anatomical landmarks for data collection with the motion analysis system (Fig. 1). Static calibration trials were conducted to determine the center of rotation of body segments for the camera-based system. The balance test included two 60-s trials under six different conditions similar to the Balance Error Scoring System (BESS) (Bell et al. 2011), but adapted and targeted for use in the older population as well as in patients with a neurological disorder. The balance conditions include: (1) double-leg stance, eyes open, firm surface; (2) double-leg stance, eyes closed, firm surface; (3) tandem stance, eyes open, firm surface; (4) double-leg stance, eyes open, foam surface; (5) double-leg stance, eyes closed, foam surface; and (6) tandem stance, eyes open, foam surface. During the double-leg stance, participants were asked to stand with feet together. During the tandem stance, participants were asked to stand with their dominant leg in front of the other. The first trial of each condition was used in the analysis unless the subject had to be prevented from falling by a spotter. If this occurred, the second trial was used. During all trials, participants were barefoot, with hands resting on
13
their iliac crests, and they were instructed to visually fixate on a target on the wall (approximately 3 m from the participant) and stand as still as possible for 60 s. Total study time, including setup, was approximately 60 min. Specifications of accelerometer and gyroscope The iPad is equipped with an accelerometer (ST Micro LIS3DH) and gyroscope (ST Micro L3G4200D), along with a Wi-Fi transreceiver. The accelerometer measured raw linear acceleration (units: mg) with sensitivity of 0.9– 1.1 milli-gravitational units at 12-bit representation with the range of ±2 g. The device’s gyroscope measured raw rotation-rate (units: mdps) with sensitivity of 8.75 m° per second at 12-bit representation with the range of ±250 dps. Acceleration and rotation-rate data were sampled at 100 Hz. Measurements were made in three dimensions, with linear acceleration measured along each of the iPad’s three major axes and rotation-rate measured about each of those axes. Using the Core Motion framework of iOS 6, the CCBApp collected data from the iPad’s built-in accelerometer and gyroscope. The application measured two acceleration vectors: gravity and user acceleration. User acceleration was the acceleration recorded independent of gravitational effects and reflected only accelerations imparted on the device by the user’s movement. Because the Core Motion framework was able to track the iPad’s altitude, or its given orientation relative to a given frame of reference, it was able to differentiate between gravity and user acceleration.
Exp Brain Res (2014) 232:3861–3872
The raw data were stored locally on the device in JavaScript Object Notation (JSON), and following test completion for each subject, the administrator extracted the data from the iPad using a Mac OS X application called CC-BApp. Data analysis Postural stability (movement of COM) was compared between two systems: reflective markers of the motion analysis system and inertial sensors within the iPad. The similarities between the two methods validate the accuracy of the inertial sensor system with respect to the currently accepted camera-based method using time- and frequencydomain outcome measures. Such an inertial sensor system may provide an alternative that is suitable for use in a clinical setting. In regards to the motion analysis system, the XYZ location of the COM for each of the 24 reflective markers on specific body segments was computed for kinematic calculations. Position of whole-body COM was determined using the mass of each body segment and the location of the COM for each segment. The second-order derivative of the positional data with respect to time was then calculated to obtain linear and angular acceleration for all balance trials. In regards to the inertial sensors, angular velocities and linear accelerations from the gyroscope and accelerometer, respectively, were processed to allow direct comparison with the parameters available with the camera-based system. The movement of COM was calculated from both the iPad and motion analysis systems recorded at a 100-Hz sampling frequency and after applying a 3.5-Hz cutoff, fourth-order, low-pass Butterworth filter. A custom MATLAB (MathWorks, Natick, MA) program was written to analyze the accuracy and experimental concurrent validity of iPad-based measures compared to motion analysis measures of postural stability. For each trial, five variables were computed from the ML and AP linear acceleration, TR angular acceleration, and resultant linear and angular acceleration to characterize postural steadiness. In the time domain, four measures that characterized the iPad’s trajectory were computed: (1) peak-to-peak (P2P), which quantifies displacement amplitudes, (2) normalized path length (NPL), (3) root-mean-square distance (RMS) which quantifies the magnitude of COM displacements, and (4) ellipsoid volume (95 % volume) that, with 95 % of probability, contained the center of the points of sway in 3D. In the frequency domain, a spectral property was assessed by quantifying the average total power of the iPad’s signal (total power). Total power was calculated using the magnitude of the ML, AP, and TR resultant acceleration. Additionally, due to having approximately 6,000 data points per balance trial, with stretches of steady readings followed by rapid changes, it is possible for unsynchronized
3865
signals with similar distributions to appear statistically related. Thus, each pair of signals was compared on a point-by-point basis. For comparison between the iPad and motion analysis system, mean absolute percentage error (MAPE) values (with larger MAPE values indicating lower reliability) (Hamilton 1994) were calculated to capture the degree of disparity between the ML, AP, and TR COM time-series derived from the iPad and motion analysis systems. Motion analysis values were used as the reference measure and the iPad-based measures as the experimental measure. For each acceleration data point recorded during each balance trial, the absolute difference or error between the iPad and motion analysis sway metric is divided by the measured value (motion analysis) and multiplied by 100. MAPE values were collapsed across sway directions within each condition and averaged across subjects. The Pearson product-moment correlation with a 95 % confidence interval for each correlation coefficient was used to assess the relationship between the iPad and motion analysis systems. Differences between conditions were determined using a one-way ANOVA and Tukey’s post hoc test to determine which pairs of conditions were significantly different. Differences were assumed significant when p
RESEARCH ARTICLE
Quantification of postural stability in older adults using mobile technology Sarah J. Ozinga · Jay L. Alberts
Received: 3 March 2014 / Accepted: 5 August 2014 / Published online: 24 August 2014 © Springer-Verlag Berlin Heidelberg 2014
Abstract Traditional biomechanical systems used to capture kinematic data have shown that declines in postural stability are frequently present in older adults and neurological populations. Recent advances in processor speed and measuring capabilities of on-board electronics within mobile devices present an opportunity to gather kinematic data and apply biomechanical analyses to potentially quantify postural stability. The aim of this project was to determine if the kinematic data gathered using a mobile device were of sufficient quantity and quality to characterize postural stability in older adults. Twelve healthy older adults completed six different balance conditions under altered surface, stance and vision. Simultaneous kinematic measurements were gathered from a three-dimensional motion analysis system and iPad during balance conditions. Correlation between the two systems was significant across balance conditions and outcome measures: peak-to-peak (r = 0.70–0.99), normalized path length (r = 0.64–0.98),
root mean square (r = 0.73–0.99) of linear acceleration, 95 % volume (r = 0.96–0.99) of linear and angular acceleration and total power across different frequencies (r = 0.79–0.92). The mean absolute percentage error metric, used to evaluate time-series measurements point-bypoint, indicated that when measuring linear and angular acceleration, the iPad tracked the motion analysis system with average error between 6 and 10 % of motion analysis measurements across all balance conditions. Collectively, similar accuracy with the iPad compared to motion capture suggests the sensors provide sufficient accuracy and quality for the quantification of postural stability in older adults. The objectivity, portability, and ease of use of this device make it ideal for use in clinical environments, which often lack biomechanical systems.
S. J. Ozinga · J. L. Alberts (*) Department of Biomedical Engineering, Cleveland Clinic, Cleveland, OH, USA e-mail: [email protected]
Introduction
S. J. Ozinga e-mail: [email protected] S. J. Ozinga Department of Chemical and Biomedical Engineering, Cleveland State University, Cleveland, OH, USA J. L. Alberts Center for Neurological Restoration, Cleveland Clinic, Cleveland, OH, USA J. L. Alberts Cleveland FES Center, L. Stokes Cleveland VA Medical Center, Cleveland, OH, USA
Keywords Older adults · Postural stability · Consumer electronics device · Motion analysis systems
In the USA, the 2010 Census reported the greatest number and proportion of people age 65 and older in history: 40.3 million, or 13 % of the total population, with significant growth expected (Bourke and Lyons 2008). Approximately 30–60 % of the older adult population sustain a fall each year (Bronstein et al. 2011), leading to reduced functioning and premature nursing home admissions, morbidity, and mortality (Santos et al. 2011). Postural stability relies on sensory (i.e., visual, vestibular, and somatosensory) and motor components of the central nervous system (Shacham et al. 2007), and cognitive function (Kupsch et al. 2006; Jorgensen 2014). The process of aging is related to a decline in the integrity of these systems, increasing the risk
13
3862
of falling. Quantifying age-related changes in the postural control system may result in reduced fall risk, improved physiological and functional performance, and an improved quality of life (Volkmann et al. 2007). Interestingly, despite the high prevalence of significant consequences and the severity of the functional limitations resulting from balance deficits, there is little agreement among health professionals about the most appropriate tools with which to quantify this impairment. Therefore, the development and refinement of assessment and treatment approaches to improving balance control in older adults is a critical and unmet need. Functional performance-based tests of postural stability have proven helpful in documenting balance status and changes with intervention, while typically rating performance on a set of motor tasks on a three to five point scale or using a stop-watch to track duration that the subject can maintain balance in a particular posture (Horak 1997). Although these tests are practical clinically and have shown a moderate relationship to fall risk (Behrman et al. 2002; Berg and Norman 1996; Giorgetti et al. 1998), inter-rater reliability introduces bias and may mask important features of balance control that vary in each individual (Mannan et al. 2008). Of the many functional balance tests available, the Berg Balance Scale (BBS), Timed Up and Go (TUG), Balance Screening Tool (BST), and Fullerton Advanced Balance (FAB) scale have established reliability and validity with community dwelling older adults (Mannan et al. 2008). Unfortunately, capabilities of such clinical measures to predict falls are weak because of their relatively high false-positive and false-negative predictions, rudimentary scoring, and lack of sensitivity to small changes in balance control over time (Mannan et al. 2008; Mancini et al. 2011; McNames et al. 2004; Blum and KornerBitensky 2008). For example, Downs and colleagues recently found that the BBS has acceptable reliability; however, it may not detect mild, clinically important changes in balance in older adults (Bourke et al. 2007). Also, the BBS may not adequately challenge older adults who live independently, and for those who score highly initially, its use as an outcome measure is compromised (Rose et al. 2006). The TUG and BST do not have the depth of information to discriminate between the various sources of possible impairment (i.e., visual, vestibular, or somatosensory inputs), and meaningful scores cannot be recorded for the TUG where participants are physically unable to rise from a chair or walk independently (Rockwood et al. 2000). Finally, although the FAB includes more challenging tasks that test dynamic balance and challenge sensory inputs, use of test administration to assure reliability requires use of video-taped assessments, limiting its wide-spread use (Mannan et al. 2008). Consequently, there remains a need for a valid and
13
Exp Brain Res (2014) 232:3861–3872
reliable functional balance test that appropriately challenges balance of older adults, which is crucial for clinicians who want to monitor postural stability over time or for individuals with more subtle balance impairments that may be candidates for an intervention. In the last two decades, quantitative assessments have shown promising results in identifying postural instability earlier in the aging process (Laughton et al. 2003; Norris et al. 2005). To date, force platforms and optical motion analysis systems remain the most common methods to quantify balance in nonclinical settings. Force plate analysis quantifies center of pressure (COP) displacement, which represents a weighted average of all the pressures over the surface of the area in contact with the ground. Motion analysis quantifies movement of the whole-body center of mass (COM), which is the weighted average of the COM of each body segment in three-dimensional (3D) space. Movement of COM is an important measure due to the association of falls and decreased trunk mobility (Reilly et al. 2008) and sudden trunk movements (Bloem et al. 2001; Woollacott and Shumway-Cook 2002). Though force platforms and motion analysis systems are sensitive, objective, and have been well-accepted methods of measuring postural stability (Monsell et al. 1997; Simon 2004), clinical utilization is not widespread. Significant cost, space, and time requirements associated with traditional biomechanical approaches act as barriers to large-scale screening and clinical utilization to assess fall risk or evaluate the efficacy of fall prevention programs across large populations or sample sizes (Boers et al. 2001; Simon 2004). Despite these limitations, traditional biomechanical approaches provide the gold standard in terms of postural assessment and play an important role in understanding mechanisms underlying declines in postural stability. An alternative to traditional biomechanical methods is the use of body-worn accelerometers and gyroscopes to obtain kinematic values (van den Bogert et al. 1996; Dai et al. 1996; Luinge et al. 1999; Tong and Granat 1999; Veltink 1999; Veltink et al. 1996). Accelerometers are used to quantify trunk movements in medial–lateral (ML) and anterior–posterior (AP) directions during stance and gait tests (Gill et al. 2001; Whitney et al. 2011; O’Sullivan et al. 2009). Accelerometers have been tested for both precision and accuracy (Moe-Nilssen 1998) and have been shown to have the ability to significantly identify differences between test conditions, young and older adults, and fallers and nonfallers (Kamen et al. 1998; Cho and Kamen 1998; Moe-Nilssen and Helbostad 2002; Yack and Berger 1993; Mayagoitia et al. 2002a). Measurement of trunk rotation (TR) with the use of gyroscopes during stance and gait tasks can detect changes in postural stability with age (Gill et al. 2001), can identify pathological balance control in
Exp Brain Res (2014) 232:3861–3872
individuals with a vestibular deficit (Allum et al. 2001), and can track movements in balance control over time following recovery from a vestibular deficit (Allum and Adkin 2003). Combining accelerometers and gyroscopes in one device increases the accuracy in quantifying COM acceleration in clinical applications compared to devices composed of sole accelerometers or gyroscopes (Wu and Ladin 1996). With use of inertial sensors (i.e., accelerometers and gyroscopes), metrics comprising a subject’s movement of COM acceleration have been shown to have the ability to significantly identify increased postural sway in older adults (Rispens et al. 2014; Galan-Mercant and Cuesta-Vargas 2014), patients with Parkinson’s disease (PD) (Palmerini et al. 2011), and those at risk for developing PD (Maetzler et al. 2012). Despite the advantages of using accelerometer and gyroscope technology, they have not been embraced clinically. Possible reasons include increased setup time or offline data processing requirements, both of which disrupt the ever constricted clinical time and workflow. If proven accurate, the use of mobile devices that contain inertial sensors to quantify postural sway may provide a practical and low-cost alternative to clinical ratings, force plates, and optical motion analysis systems (Pohl et al. 2003; Pickering et al. 2007; Mayagoitia et al. 2002b) in the quantification of postural stability. To bridge the gap between subjective performance-based measures and expensive and sophisticated biomechanical assessment methods related to postural stability analysis, the aim of the present study was to determine if the kinematic data measured by the native accelerometer and gyroscope within the iPad (third generation) was of sufficient quantity and quality to accurately quantify postural stability in older adults. To verify the accuracy of the iPad measurements, a 3D model will be utilized to validate the use of the Cleveland Clinic Balance Assessment App (CC-BApp) within the iPad to assess postural stability across six balance conditions of varying difficulty against clinically valid and reliable measures of 3D motion analysis.
Materials and methods Participants A single-center, validation study was completed in order to validate the accuracy of an inertial sensor system within a mobile application designed to assess postural stability. Twelve healthy community-living older adult subjects (5 male and 7 female) aged between 60 and 85 (mean age ± sd, 68.3 ± 6.9) completed all study conditions (see Table 1 for participant demographics). All data were collected during one visit to Cleveland Clinic Biomechanics Laboratory. Inclusion criteria included: age (60 and older),
3863 Table 1 Summary of subject characteristics Subject
Gender
Age (years)
BMI (kg/m2)
1 2 3 4
M F F F
69 66 70 60
31.7 23.7 28.8 22.9
5 6 7 8 9 10 11 12
M M F M F F F M Mean
69 81 69 60 79 68 60 67 68
14.3 24.8 27.3 32.7 29.3 22.1 45.7 31.1 27.9
7
7.6
SD
ability to demonstrate an understanding of the study, ability to follow two-step commands, ability to walk comfortably unaided, and ability to stand comfortably for at least 5 min unaided. Primary exclusion criteria included: the presence of diagnosed dementia of any etiology, history of musculoskeletal or neurological injury that currently impairs postural stability, and a premorbid condition that affects balance. All participants were informed of the experimental procedures, which had received ethical committee approval from the institutional review board and provided witnessed written consent prior to testing. Data collection Biomechanical data were collected using a motion analysis system (Motion Analysis Corporation Eagle System; Santa Rosa, CA) with eight infrared Eagle digital cameras. For collection of iPad data, an in-house application was written to acquire and store accelerometer and gyroscope data during balance testing. In order to compare postural stability parameters obtained using inertial sensors within the iPad and the motion analysis system, data were simultaneously collected from both systems for all subjects during data collection. To synchronize the data collection between the two systems, an Arduino Pro Mini 3.3v and LED light were used to align the balance data temporally. Upon data collection, an output signal from the iPad audio headphone jack triggered the LED light to illuminate. The light provided the signal to the motion analysis digital cameras indicating that the iPad had begun collecting data. Each trial consisted of 60 s of data collection; the motion analysis system recorded 90-s trials to ensure capturing of movement during iPad collection time.
13
3864
Exp Brain Res (2014) 232:3861–3872
Fig. 1 Illustration of experimental paradigm and measurement setup
Procedure Participants completed a series of postural stability tests while the iPad was securely attached via a clip to a belt around their waist. The iPad was placed as close as possible to the second sacral vertebra to approximate body COM during data collection with CC-BApp. Reflective markers, using the Helen Hayes marker-set (Kadaba et al. 1990; Davis et al. 1991), were attached to 24 anatomical landmarks for data collection with the motion analysis system (Fig. 1). Static calibration trials were conducted to determine the center of rotation of body segments for the camera-based system. The balance test included two 60-s trials under six different conditions similar to the Balance Error Scoring System (BESS) (Bell et al. 2011), but adapted and targeted for use in the older population as well as in patients with a neurological disorder. The balance conditions include: (1) double-leg stance, eyes open, firm surface; (2) double-leg stance, eyes closed, firm surface; (3) tandem stance, eyes open, firm surface; (4) double-leg stance, eyes open, foam surface; (5) double-leg stance, eyes closed, foam surface; and (6) tandem stance, eyes open, foam surface. During the double-leg stance, participants were asked to stand with feet together. During the tandem stance, participants were asked to stand with their dominant leg in front of the other. The first trial of each condition was used in the analysis unless the subject had to be prevented from falling by a spotter. If this occurred, the second trial was used. During all trials, participants were barefoot, with hands resting on
13
their iliac crests, and they were instructed to visually fixate on a target on the wall (approximately 3 m from the participant) and stand as still as possible for 60 s. Total study time, including setup, was approximately 60 min. Specifications of accelerometer and gyroscope The iPad is equipped with an accelerometer (ST Micro LIS3DH) and gyroscope (ST Micro L3G4200D), along with a Wi-Fi transreceiver. The accelerometer measured raw linear acceleration (units: mg) with sensitivity of 0.9– 1.1 milli-gravitational units at 12-bit representation with the range of ±2 g. The device’s gyroscope measured raw rotation-rate (units: mdps) with sensitivity of 8.75 m° per second at 12-bit representation with the range of ±250 dps. Acceleration and rotation-rate data were sampled at 100 Hz. Measurements were made in three dimensions, with linear acceleration measured along each of the iPad’s three major axes and rotation-rate measured about each of those axes. Using the Core Motion framework of iOS 6, the CCBApp collected data from the iPad’s built-in accelerometer and gyroscope. The application measured two acceleration vectors: gravity and user acceleration. User acceleration was the acceleration recorded independent of gravitational effects and reflected only accelerations imparted on the device by the user’s movement. Because the Core Motion framework was able to track the iPad’s altitude, or its given orientation relative to a given frame of reference, it was able to differentiate between gravity and user acceleration.
Exp Brain Res (2014) 232:3861–3872
The raw data were stored locally on the device in JavaScript Object Notation (JSON), and following test completion for each subject, the administrator extracted the data from the iPad using a Mac OS X application called CC-BApp. Data analysis Postural stability (movement of COM) was compared between two systems: reflective markers of the motion analysis system and inertial sensors within the iPad. The similarities between the two methods validate the accuracy of the inertial sensor system with respect to the currently accepted camera-based method using time- and frequencydomain outcome measures. Such an inertial sensor system may provide an alternative that is suitable for use in a clinical setting. In regards to the motion analysis system, the XYZ location of the COM for each of the 24 reflective markers on specific body segments was computed for kinematic calculations. Position of whole-body COM was determined using the mass of each body segment and the location of the COM for each segment. The second-order derivative of the positional data with respect to time was then calculated to obtain linear and angular acceleration for all balance trials. In regards to the inertial sensors, angular velocities and linear accelerations from the gyroscope and accelerometer, respectively, were processed to allow direct comparison with the parameters available with the camera-based system. The movement of COM was calculated from both the iPad and motion analysis systems recorded at a 100-Hz sampling frequency and after applying a 3.5-Hz cutoff, fourth-order, low-pass Butterworth filter. A custom MATLAB (MathWorks, Natick, MA) program was written to analyze the accuracy and experimental concurrent validity of iPad-based measures compared to motion analysis measures of postural stability. For each trial, five variables were computed from the ML and AP linear acceleration, TR angular acceleration, and resultant linear and angular acceleration to characterize postural steadiness. In the time domain, four measures that characterized the iPad’s trajectory were computed: (1) peak-to-peak (P2P), which quantifies displacement amplitudes, (2) normalized path length (NPL), (3) root-mean-square distance (RMS) which quantifies the magnitude of COM displacements, and (4) ellipsoid volume (95 % volume) that, with 95 % of probability, contained the center of the points of sway in 3D. In the frequency domain, a spectral property was assessed by quantifying the average total power of the iPad’s signal (total power). Total power was calculated using the magnitude of the ML, AP, and TR resultant acceleration. Additionally, due to having approximately 6,000 data points per balance trial, with stretches of steady readings followed by rapid changes, it is possible for unsynchronized
3865
signals with similar distributions to appear statistically related. Thus, each pair of signals was compared on a point-by-point basis. For comparison between the iPad and motion analysis system, mean absolute percentage error (MAPE) values (with larger MAPE values indicating lower reliability) (Hamilton 1994) were calculated to capture the degree of disparity between the ML, AP, and TR COM time-series derived from the iPad and motion analysis systems. Motion analysis values were used as the reference measure and the iPad-based measures as the experimental measure. For each acceleration data point recorded during each balance trial, the absolute difference or error between the iPad and motion analysis sway metric is divided by the measured value (motion analysis) and multiplied by 100. MAPE values were collapsed across sway directions within each condition and averaged across subjects. The Pearson product-moment correlation with a 95 % confidence interval for each correlation coefficient was used to assess the relationship between the iPad and motion analysis systems. Differences between conditions were determined using a one-way ANOVA and Tukey’s post hoc test to determine which pairs of conditions were significantly different. Differences were assumed significant when p
