"Using Wearable Sensors for Gait Analysis: Measurement Accuracy for Linear Measurements" by Evan Block

Using Wearable Sensors for Gait Analysis:

Measurement Accuracy for Linear Measurements

Evan Block, SUNY Albany

Abstract: Accurate measurement of human motion is critical for assessing mobility, diagnosing movement disorders, and predicting fall risk in clinical settings. Historically, motion data were typically collected using optical motion capture systems, which are expensive and can only be used in labs. Inexpensive, wearable sensors called Inertial Measurement Units (IMUs) can expand gait assessments beyond the laboratory, but their accuracy for certain measurements, including linear displacement, remains uncertain. In this study, we evaluate measurement accuracy for linear displacements obtained using Sparkfun OpenLog Artemis IMUs. We conducted a series of tests, moving the sensor along predetermined distances in different directions and evaluating its ability to capture displacement accurately. IMU data were processed using custom MATLAB algorithms to calibrate, filter, and double-integrate acceleration data to obtain recorded linear displacements in all three planes of motion (e.g., x, y, and z axes). Our results indicated that the IMUs could not accurately measure linear displacements, with average measurement errors ranging from 22.5% to 83.7%. We also found that the degree of measurement error differed based on the direction of measurement for this device. These results indicate that additional calibration steps should be pursued before using these devices for clinical distance measurement. In the future, we will incorporate an assessment of device accuracy to operating temperature and sensitivity settings, as well as inter-device error variation. This research is part of a larger series of IMU validation experiments that, if successful, will enable inexpensive IMU-based measurement of human motion with a broad range of applications.

---

Introduction

Across all ages, falls are one of the most common mechanisms of injury and have a high risk of morbidity and mortality across all ages. Globally, estimates show that falls ranked as the 18th leading cause of age-standardized rates of disability-adjusted life years (DALYs) in 2017, outranking conditions such as chronic kidney disease, Alzheimer's disease, and asthma. The mortality rate for elderly individuals aged 75 years and older due to falls has risen from 51.6 per 100,000 in 2000 to 122.2 per 100,000 in 2016, illustrating a troubling increase in fatality rates. Additionally, falls surpassed causes including drowning and interpersonal violence to emerge as the second most common cause of unintentional injury-related deaths in 2017, behind traffic injuries (James et al., 2020). Given the extent of this burden internationally, proactive intervention methods must be developed to address this issue in an increasingly aging population.

Falls among elderly individuals in long-term care facilities represent a critical public health concern, which is the leading cause of morbidity and mortality in this demographic. Functions that maintain normal gait, such as effective coordination of the basal ganglia brainstem, regulated muscle tone, and functional proprioception, tend to decline with age. It is also common for older individuals to accumulate medical issues, as well as medications that can alter balance. Polypharmacy, an umbrella term used by healthcare professionals to describe a patient who is using multiple medications (at least 4) for different conditions, is the 6th most influential risk factor for falls. Antiplatelet or anticoagulant medications (e.g., Heparin, Eliquis) can cause dizziness and drowsiness; antiarrhythmics, benzodiazepines, and diuretics can further contribute to altered states of consciousness, dehydration, and confusion (Appendu, 2023).

Beyond the immediate physical consequences of falls, such as hip fractures, traumatic brain injuries, and spinal damage, the psychological impact is equally severe. Many elderly individuals develop a fear of falling, leading to reduced mobility, decreased social participation, and an overall decline in quality of life. This fear-induced restriction contributes to muscle deconditioning, increasing the likelihood of subsequent falls. The highest risk factor was a history of falls; 30% of individuals over 65 fall yearly. In approximately half of the cases, the falls are recurrent (Appendu, 2023). Typically, gait is a good benchmark of health and longevity in older adults and indicates overall well-being. Gait also serves as an essential tool for various clinical applications, including, but not limited to, the evaluation of disease progression, predicting falls in older adults, assessing life satisfaction, and predicting patients' survival.

In the United States, the average cost per fall-related hospitalization among individuals aged 65 and older is $17,483, with annual national healthcare expenditures on nonfatal falls exceeding $50 billion and an additional $754 million on fatal falls (Hadad et al, 2024). According to a 2014 study conducted by Haddad and colleagues, 29% of older adults reported falling, which resulted in over 3 million emergency department (ED) visits and 28,000 deaths. The study estimated the national cost of fall-related injuries at approximately $49.5 billion. Medicare accounted for a significant share, spending $31.3 billion on fall-related medical treatment in 2015. State-level data highlighted significant disparities, with California, Florida, and New York incurring the highest costs—California alone spent $4.4 billion.

Elderly patients admitted for fall-related injuries experience more extended hospital stays (median of 5 days) and higher rehospitalization rates (14% within 30 days). Furthermore, 69% of elderly fall patients are discharged to rehabilitation or long-term care facilities rather than returning home. Siracuse and colleagues reported that fall-related hospitalizations for patients aged 75+ cost a median of $11,000, with cardiac conditions, orthopedic procedures, and pneumonia as significant cost drivers. More extended hospital stays are also typically associated with a higher incidence of contracting lethal bacterial or viral infections.

Current Methods of Fall Risk Assessment

Given the widespread consequences of falls among older adults, the healthcare industry needed to develop proactive intervention methods and understand how certain conditions impact the gait cycle. Since the 1990s, the Berg Balance Scale (BBS) has been the gold standard of fall risk assessment in older adults and individuals with neurological conditions. The test consists of 14 simple balance-related tasks, ranging from standing up to sitting to balancing on one foot. One parameter that the BBS utilizes is the Timed Up-and-Go (TUG) test, which measures how long it takes a person to stand up from a chair, walk 10 feet, turn around, walk back, and sit back down. According to a 2006 meta-analysis conducted by Bohannon, the mean TUG time for individuals at least 60 years of age was 9.4 seconds. While there are some discrepancies regarding the exact benchmarks of a standard TUG test, individuals who take more than 20 seconds to complete the test are classified as "prone to falling."

While TUG performance shows some ability to differentiate fallers from non-fallers, its discriminative power is weak in high-functioning individuals, where the time difference between groups is minimal and clinically insignificant. The test is more effective in lower-functioning and institutionalized adults, where fallers took up to 3.59 seconds longer to complete. The study highlights poor to moderate diagnostic accuracy overall, with widely inconsistent cutoff values across studies (ranging from 8.1 to 32.6 seconds), making it unreliable for universal application. Additionally, TUG performance was not retained as an independent predictor of falls in most multivariate models, indicating that other factors, such as cognitive function, use of walking aids, and fall location, play a crucial role in fall risk. The authors discourage reliance on TUG as a standalone screening tool, advocating for multifactorial assessments that consider gait, balance, cognitive impairments, and environmental hazards to provide a more accurate fall risk evaluation in older adults.

Oliver et al. (2008) critically evaluate the STRATIFY tool, a widely used fall risk assessment system for hospitalized patients, revealing significant limitations in its predictive validity. While STRATIFY demonstrates moderate sensitivity (67.2%) and high negative predictive value (86.5%), indicating it can effectively rule out patients unlikely to fall, its low specificity (51.2%) and poor positive predictive value (23.1%) suggest it frequently misclassifies patients as high risk. The study also highlights concerns about STRATIFY's static nature, as it assigns risk scores at admission without accounting for daily changes in patient condition, such as gait instability or agitation. These findings underscore the need for more adaptive fall risk protocols beyond simple risk scoring to enhance hospital fall prevention strategies.

Scott et al. (2007) systematically reviewed fall-risk assessment tools across community, home-support, long-term care, and acute care settings. The researchers analyzed 34 studies testing 38 tools, categorizing them into Multifactorial Assessment Tools, which assess a broad range of risk factors, and Functional Mobility Assessments, which focus on gait, balance, and strength. Findings revealed that no single tool demonstrated consistent predictive validity across all settings. Tools like the Elderly Fall Screening Test (83% sensitivity, 69% specificity) and the Fall-Risk Assessment Tool (93% sensitivity, 78% specificity in acute care) performed well in specific populations but lacked universal applicability. The STRATIFY tool, commonly used in hospitals, showed high sensitivity (93%) but variable specificity (60–88%), limiting its reliability. Additionally, many tools lacked standardization, varied in inter-rater reliability, and showed inconsistent performance across studies, making direct comparisons difficult. Researchers emphasize the need for multifactorial assessments tailored to specific patient populations rather than relying on single-score tools and call for future research to standardize validation methods and improve the clinical applicability of fall-risk assessments.

Exploring the Role of IMUs in Fall Risk Assessment

The traditional assessment techniques mentioned in the previous section rely on episodic in-lab visits and observational scales, which are increasingly scrutinized for reliability, validity, and objectivity. Their sporadic administration, clinical constraints such as limited space and time, and the inherent subjectivity of questionnaires raise concerns about their accuracy in capturing real-world movement patterns. Digital health technologies (DHTs) offer a promising alternative by enabling gait measurement in free-living conditions within individuals' natural environments, which reduces the need for travel, enhances patient engagement, and is particularly beneficial for populations requiring assistance, including the elderly, individuals with physical or cognitive impairments, and those in remote areas. By facilitating decentralized clinical trials, DHTs expand access to research participation. Moreover, in-lab assessments are susceptible to the Hawthorne effect, wherein individuals alter their behavior due to awareness of being observed (Camerlingo et al, 2024). In contrast, continuous gait monitoring in free-living conditions provides a more representative measure of an individual's physiological state, especially when data is collected over extended periods to mitigate confounding factors such as seasonality, temporary illnesses, socioeconomic status, and weekend variability. By improving diversity and inclusion in clinical trials and expanding access to care, DHTs represent a significant advancement in patient-centered assessment methodologies.

The integration of wearable sensor technology, particularly Inertial Measurement Units (IMUs), has emerged as a promising tool for objective, continuous gait monitoring in elderly populations. IMUs containing accelerometers, gyroscopes, and magnetometers can track spatiotemporal gait parameters such as stride length, cadence, walking speed, and postural stability. Unlike traditional optical motion capture systems or sensor-embedded floors, which are expensive and limited to laboratory environments, IMU-based assessments allow for real-time monitoring in natural settings.

While several studies have demonstrated the feasibility of IMUs in gait analysis, concerns regarding measurement accuracy and clinical reliability remain. Álvarez et al. (2023) demonstrated that G-STRIDE IMUs improved fall risk identification by 11% over traditional clinical tests, though their study did not assess long-term predictive effectiveness. In another study, addressing sensor placement and data fusion challenges, Baghdadi et al. (2018) found that Extended Kalman Filters (EKF) achieved lower error rates (6.5%) than Unscented Kalman Filters (UKF, 13.02%), emphasizing the importance of advanced signal processing techniques.

In this study, we evaluate measurement accuracy for linear displacements obtained using Sparkfun OpenLog Artemis IMUs. We conducted a series of tests, moving the sensor along predetermined distances in different directions and evaluating its ability to capture displacement accurately. IMU data were processed using custom MATLAB algorithms to calibrate, filter, and double-integrate acceleration data to obtain recorded linear displacements in all three planes of motion (e.g., x, y, and z axes). 

Methods

Materials

This study aims to determine if the Sparkfun OpenLog Artemis IMU can accurately reconstruct linear displacement in controlled conditions. The SparkFun OpenLog Artemis is an open-source data logger preprogrammed to automatically log IMU, GPS, serial data, and various pressure, humidity, and distance sensors (Figure 1). The device has an IMU for built-in logging of the triaxial accelerometer, rate gyro, and magnetometer. Data from each trial was stored on a microSD card, which was transferred to a desktop computer for analysis.

Figure 1: The SparkFun OpenLog Artemis IMU captures linear acceleration along three orthogonal axes: X (vertical), Y (horizontal), and Z (depth).

The device was initialized using the Arduino IDE application. The terminal log rate was 100 Hertz, and the serial terminal baud rate was 115200 bits per second. Additionally, the device's internal accelerometer low pass was disabled to enable the device to provide the "rawest" acceleration signal possible so that the code for calibration and filtering algorithms can be applied consistently. Because the acceleration data is being double-integrated, any internal automatic filtering mechanisms could introduce phase shifts in the plotted data and/or extraneous data points.

Procedure

Prior to any data recording, the device needed to be calibrated to ensure that the sensors were responding correctly to changes in acceleration. The calibration protocol begins after initializing the device and plugging it into a power source. The "reset" button on the bottom of the device was clicked prior to each trial to name a new code file that would record data. The sensor was placed on a flat surface and tapped downward in the Z-direction 3 times, with each subsequent tap occurring 3 seconds apart. After a 5-second waiting period, the sensor was held facing upward in each of the three planes of direction to ensure it was correctly measuring acceleration in all three planes.  The sensor would remain resting on a flat surface for an additional 5 seconds before it was moved in a linear motion across the top of the surface for a predetermined distance of 80 cm 10 times. To obtain the actual known distance, the length, width, and height of the case were subtracted from the original predetermined distance. 75.5 cm, 76.8 cm, and 70.05 cm were the actual known distances in the X, Y, and Z directions, respectively. The device was moved in all three axes of measurement back and forth, resting 3-5 seconds between movements.  In this study, we compared the efficacy of the ±2g vs. ±4g settings in the Accelerometer Full Scale to measure linear displacement accurately. The "Accelerometer Full Scale" setting tells the IMU the maximum range of acceleration it should measure (for example, ±2g, ±4g, ±8g, or ±16g). In other words, when configuring a more extensive full-scale range, the sensor is ready to measure a larger range of acceleration, but does so at a lower resolution. Conversely, a smaller full-scale range gives a higher resolution for smaller accelerations but will "clip" or saturate if the sensor undergoes accelerations beyond that smaller range. Each trial took about 90 seconds, and the data obtained from the device was transferred to a local computer for processing.

Data Processing and Analysis

All data was processed in MATLAB_R2023b. To address sensor drift and offsets in the data, MATLAB code was developed to apply a fourth-order low-pass Butterworth filter with a cutoff of 5 Hz and a sampling frequency of 85 Hz. This filter is used to accelerate data to smooth out any high-frequency noise and scale the data correctly. Proper scale factors are established by calculating the mean values (offsets) for each axis, subtracting those offsets, and then calculating scale factors (so that a known 1 g orientation registers as 1 g instead of some arbitrary value). If the sensor is supposed to measure 1 g (e.g., with one axis pointing straight up under gravity) but only reads 0.95 g or 1.05 g, those errors can become significant when you integrate acceleration to determine velocity and displacement.

Once the data has been appropriately scaled, a separate MATLAB script is applied to use the calibrated accelerations to integrate and compute the velocity and displacement over the course of each trial. The script plots the already-filtered acceleration signals (Ax, Ay, Az), and a start and end point are selected to isolate the interval of interest (e.g., the portion of the trial where the sensor is physically moved). A trapezoidal integration is applied to filtered signals over the chosen time window, producing velocity estimates (Vx, Vy, and Vz). A high-pass Butterworth filter is used to remove drift in the velocity signals. The cutoff frequency is set at 0.5 Hz, and then the velocity vectors are filtered. Another trapezoidal integration is performed on the high-pass-filtered velocity to compute displacement in each axis (Dx1, Dy1, Dz1). The script calculates the total displacement in X, Y, and Z by subtracting the start from the end of each axis's filtered displacement signal. The filtered data points for each direction were entered into an Excel spreadsheet and analyzed for percentage error, average error, and standard deviation.

Results

Our results indicated that the IMUs were not able to measure linear displacements accurately, with average measurement errors ranging from 22.5% to 61.4%. We also found that the degree of measurement error differed based on the direction of measurement for this device. Under both conditions, with the Accelerometer Full Scale set at ± 2g and at ± 4g, the known distances were 75.5 cm, 76.8 cm, and 70.05 cm in the X, Y, and Z directions, respectively. With the Accelerometer Full Scale set at ±2g, the average percentage error of the 10 measurements was 20.1%, 83.7%, and 63.1% with respect to each axis of measurement (Figure 2). When the Accelerometer Full Scale was set at ±4g, the average percentage error remained statistically insignificant, with 22.5%, 58.1%, and 61.4%, with respect to each axis of measurement (Figure 3).

Figure 2: Graphical representation of known vs. measured displacement data at ± 2g.

Figure 3: Graphical representation of known vs. measured displacement data at ± 4g.

Discussion

The goal of this study was to determine whether the SparkFun OpenLog Artemis IMU could accurately reconstruct linear displacement measurements in controlled, single-plane movements. We conducted a series of trials using both the ±2g and ±4g Accelerometer Full Scale settings and compared measured displacements to known distances across the X, Y, and Z axes. The findings showed significant measurement inaccuracies across all axes, with error margins ranging from approximately 22.5% to 83.7%. 

It was hypothesized that the IMU could measure linear displacements with minimal error. However, the data did not support this hypothesis. The IMU's double integration of the acceleration data and the sensor's internal error matrix introduced substantial drift and noise that rendered the displacement measurements unreliable. Interestingly, the Accelerometer Full Scale setting (±2g vs. ±4g) had a relatively small influence on measurement accuracy. Although there were some reductions in error on the Y-axis when switching from ±2g to ±4g, these differences were not statistically meaningful, and overall accuracy remained poor. This suggests that factors other than dynamic range likely contribute to the observed inaccuracies.

The literature highlighted that while IMUs can accurately track velocity and acceleration over short intervals, calibration errors and noise accumulation over time can compound when integrated twice. This aligns with our findings and further emphasizes the challenges of using consumer-grade IMUs for precise displacement measurement. The study succeeded in creating a structured MATLAB pipeline for calibration, filtering, integration, and analysis—an asset that can be adapted for future sensor-based work. In a 2018 study Al-Amiri and colleagues, Al-Amiri and colleagues evaluated whether the Xsens MVN BIOMECH sensor system, a commercially available set of IMUs, could effectively capture clinically relevant functional activities, such as squat depth and gait pattern. To compare reliability and concurrent validity, optical motion cameras (VICON Motion Systems) were also used during each activity. Additionally, different practitioners took turns running each trial, so as to rule out human error when evaluating the results.

Among the 26 healthy participants, the researchers found fair-to-excellent reliability (ICC = 0.6-0.95) in the IMUs compared to optical motion capture. During walking and stair climbing, the mean differences in lower-limb joint angles between the two systems were between 1.4 and 6.7, and the coefficient of multiple correlation (CMC) was between 0.39 and 0.99. While discrepancies between raters and body planes highlighted the need for standardized protocols, this study presents reasonable data to suggest that the MVN BIOMECH System can accurately quantify joint kinematics during complex functional movements. 

Both studies, mine and Al-Amiri et al., employed IMUs with comparable operating systems that combine accelerometers, gyroscopes, and magnetometers to capture motion data. Each experimental approach incorporated a MATLAB-based analysis pipeline to process and interpret sensor outputs. Despite differences in focus—joint kinematics versus linear displacement—both works demonstrate the growing potential of wearable sensors for clinically relevant movement assessment. With continued refinement in calibration, filtering, and modeling, these devices may ultimately be used to reliably predict fall risk in vulnerable populations.

There are several significant limitations to note. First, the experiment was conducted under idealized, controlled conditions that do not reflect the complexity of natural human gait or clinical environments. The poor performance under such ideal circumstances suggests that performance may degrade even further in more dynamic, real-life settings. Second, the relatively small number of trials (n = 10 per axis) and reliance on only one hardware device mean that results may not be generalizable across other sensor configurations or populations.

Future Directions

Reconstructing linear displacement measurements using double integration proved to be unreliable. However, the literature suggests that a double integration method is possible when using multiple devices – something we might consider for future experiments. A 2007 case study by Alvarez et al. evaluated the feasibility of estimating specific stride length parameters, Heel Off (HO) and Footfall (FF), by combining data from two IMUs placed on the front of each foot of a subject. A direct double integration of the horizontal acceleration data was gathered from the IMUs and compared against the camcorder data. When identifying HO and FF, the overall percent error (compared to the arbitrary known value of 0%) was 10.069% +/- 6.167, with estimations ranging from –14.87% to +9.12%.

Additionally, we might consider introducing AI-based correction models to reduce inaccuracies in building the code and while coding the data. The scaling and filtration of raw data were done manually, which presents the possibility of human error negatively affecting our results. We would also consider examining the sensor's internal error matrix more closely, which might enable us to build a more effective code to analyze the data.

Conclusion

Despite the limitations, this study contributes meaningfully to the broader conversation on improving fall-risk diagnostics in elderly populations. It reinforces the need for caution in relying solely on IMU-based displacement data for clinical decision-making. The study also validates the idea that IMUs can still be applied to other aspects of fall risk assessments, such as velocity tracking, postural sway analysis, or activity recognition, where single integrations or pattern-based models (rather than raw displacement metrics) are sufficient.

References

Al-Amri, M., Nicholas, K., Button, K., Sparkes, V., Sheeran, L., & Davies, J. (2018). Inertial measurement units for clinical movement analysis: reliability and concurrent validity. Sensors, 18(3), 719. https://doi.org/10.3390/s18030719

Álvarez, M. N., Ruiz, A. R., Neira, G. G., Cerda, M. T., Delgado, L. P., Robles, E. R., & del-Ama, A.J. (2023). Assessing falls in the elderly population using G-STRIDE foot-mounted inertial sensor. Scientific Reports, 13(1), 1-12. https://doi.org/10.1038/s41598-023-36241-x

Alvarez, J. C., Gonzalez, R. C., Alvarez, D., Lopez, A. M., & Rodriguez-Uria, J. (2007). Multisensor approach to walking distance estimation with foot inertial sensing. 2007 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 5719–5722. https://doi.org/10.1109/iembs.2007.4353645

Appeadu, M. K. (2023, June 4). Falls and fall prevention in older adults. StatPearls. https://www.ncbi.nlm.nih.gov/books/NBK560761/

Bancroft, J. B., & Lachapelle, G. (2011). Data fusion algorithms for multiple inertial measurement units. Sensors, 11(7), 6771–6798. https://doi.org/10.3390/s110706771

Baghdadi, A., Cavuoto, L. A., & Crassidis, J. L. (2018). Hip and trunk kinematics estimation in gait through Kalman filter using IMU data at the ankle. IEEE Sensors Journal, 18(10), 4253-4260. https://doi.org/10.1109/jsen.2018.2817228

Boutaayamou, M., Schwartz, C., Stamatakis, J., Denoël, V., Maquet, D., Forthomme, B., Croisier, J., Macq, B., Verly, J. G., Garraux, G., & Brüls, O. (2015). Development and validation of an accelerometer-based method for quantifying gait events. Medical Engineering & Physics, 37(2), 226-232. https://doi.org/10.1016/j.medengphy.2015.01.001

Camerlingo, N., Cai, X., Adamowicz, L., Welbourn, M., Psaltos, D. J., Zhang, H., Messere, A., Selig, J., Lin, W., Sheriff, P., Demanuele, C., Santamaria, M., & Karahanoglu, F. I. (2024). Measuring gait parameters from a single chest-worn accelerometer in healthy individuals: a validation study. Scientific Reports, 14(1), 1-13. https://doi-org.libproxy.albany.edu/10.1038/s41598-024-62330-6

Cho, Y.-S., Jang, S.-H., Cho, J.-S., Kim, M.-J., Lee, H. D., Lee, S. Y., & Moon, S.-B. (2018). Evaluation of validity and reliability of inertial measurement unit-based Gait Analysis Systems. Annals of Rehabilitation Medicine, 42(6), 872-883. https://doi.org/10.5535/arm.2018.42.6.872

Di Raimondo, G., Willems, M., Killen, B. A., Havashinezhadian, S., Turcot, K., Vanwanseele, B., & Jonkers, I. (2023). Peak tibiofemoral contact forces estimated using IMU-based approaches are not significantly different from motion capture-based estimations in patients with knee osteoarthritis. Sensors, 23(9), 4484. https://doi.org/10.3390/s23094484

Haddad, Y. K., Miller, G. F., Kakara, R., Florence, C., Bergen, G., Burns, E. R., & Atherly, A. (2024). Healthcare spending for non-fatal falls among older adults, USA. Injury Prevention: Journal of the International Society for Child and Adolescent Injury Prevention, 30(4), 272-276. https://doi.org/10.1136/ip-2023-045023

James, S. L., Lucchesi, L. R., Bisignano, C., Castle, C. D., Dingels, Z. V., Fox, J. T., Hamilton, E. B., Henry, N. J., Krohn, K. J., Liu, Z., McCracken, D., Nixon, M. R., Roberts, N. L., Sylte, D. O., Adsuar, J. C., Arora, A., Briggs, A. M., Collado-Mateo, D., Cooper, C., … Murray, C. J. (2020). The global burden of falls: Global, regional and national estimates of morbidity and mortality from the global burden of disease study 2017. Injury Prevention, 26(Suppl 2), i3-i11.

Kaufmann, M., Nüesch, C., Clauss, M., Pagenstert, G., Eckardt, A., Ilchmann, T., Stoffel, K., Mündermann, A., & Ismailidis, P. (2023). Functional assessment of total hip arthroplasty using inertial measurement units: Improvement in gait kinematics and association with patient-reported outcome measures. Journal of Orthopaedic Research, 41(4), 759-770. https://doi.org/10.1002/jor.25421

Kocuvan, P., Hrastič, A., Kareska, A., & Gams, M. (2023). Predicting a fall based on gait anomaly detection: a comparative study of wrist-worn three-axis and mobile phone-based accelerometer sensors. Sensors, 23(19), 8294. https://doi-org.libproxy.albany.edu/10.3390/s23198294

Kim, Yeon-Wook, et al. “Wearable IMU-based human activity recognition algorithm for clinical balance assessment using 1D-CNN and GRU Ensemble model.” Sensors, vol. 21, no. 22, 17 Nov. 2021, p. 7628, https://doi.org/10.3390/s21227628.

Lin, A., Lin, T., Tan, Y., Pan, W., Shih, C., Lee, C., Chen, S., & Wang, F. (2022). Superior gait symmetry and postural stability among yoga instructors—inertial measurement unit-based evaluation. Sensors, 22(24), 9683. https://doi.org/10.3390/s22249683.

Marimon, X., Mengual, I., López-de-Celis, C., Portela, A., Rodríguez-Sanz, J., Herráez, I. A., & Pérez-Bellmunt, A. (2024). Kinematic analysis of human gait in healthy young adults using IMU sensors: Exploring relevant machine learning features for clinical applications. Bioengineering, 11(2), 105. https://doi.org/10.3390/bioengineering11020105.

McGinnis, R. S., Cain, S. M., Davidson, S. P., Vitali, R. V., McLean, S. G., & Perkins, N. C. (2017). Inertial sensor and cluster analysis for discriminating agility run technique and quantifying changes across load. Biomedical Signal Processing and Control, 32, 150-156. https://doi.org/10.1016/j.bspc.2016.10.013.

Oliver, D., Papaioannou, A., Giangregorio, L., Thabane, L., Reizgys, K., & Foster, G. (2008). A systematic review and meta-analysis of studies using the STRATIFY tool for prediction of falls in hospital patients: how well does it work? Age and Ageing, 37(6), 621-627. https://doi.org/10.1093/ageing/afn203.

Rampp, A., Barth, J., Schuelein, S., Gassmann, K.-G., Klucken, J., & Eskofier, B. M. (2014). Inertial sensor-based stride parameter calculation from gait sequences in geriatric patients. IEEE Transactions on Biomedical Engineering, 62(4), 1089-1097. https://doi.org/10.1109/tbme.2014.2368211.

Rekant, J., Rothenberger, S., & Chambers, A. (2024). Obesity-specific considerations for assessing gait using inertial measurement unit-based versus optokinetic motion capture. Sensors, 24(4), 1232. https://doi.org/10.3390/s24041232

Rojas-Valverde, D., Gómez-Carmona, C. D., Gutiérrez-Vargas, R., & Pino-Ortega, J. (2019). From big data mining to technical sport reports: The case of inertial measurement units. BMJ Open Sport & Exercise Medicine, 5(1), e000565.

Sakaguchi, T., Sake, N., Tanaka, M., Fujiwara, Y., Arataki, S., Taoka, T., Kodama, Y., Takamatsu, K., Yasuda, Y., Nakagawa, M., Utsunomiya, K., & Tomiyama, H. (2024). Use of a triaxial accelerometer to measure changes in gait sway and related motor function after corrective spinal fusion surgery for adult spinal deformity. Journal of Clinical Medicine, 13(7), 1923. https://doi-org.libproxy.albany.edu/10.3390/jcm13071923.

Schoene, D., Wu, S. M., Mikolaizak, A. S., Menant, J. C., Smith, S. T., Delbaere, K., & Lord, S. R. (2013). Discriminative ability and predictive validity of the timed up and go test in identifying older people who fall: A systematic review and meta-analysis. Journal of the American Geriatrics Society, 61: 202-208.

Scott, V., Votova, K., Scanlan, A., & Close, J. (2007). Multifactorial and functional mobility assessment tools for fall risk among older adults in community, home-support, long-term and acute care settings. Age and Ageing, 36(2), 130-139. https://doi.org/10.1093/ageing/afl165.

Thong, Y. K., Woolfson, M. S., Crowe, J. A., Hayes-Gill, B. R., & Jones, D. A. (2004). Numerical double integration of acceleration measurements in noise. Measurement, 36(1), 73-92. https://doi.org/10.1016/j.measurement.2004.04.005.

Vayalapra, S., Wang, X., Qureshi, A., Vepa, A., Rahman, U., Palit, A., Williams, M. A., King, R., & Elliott, M. T. (2023). Repeatability of inertial measurement units for measuring pelvic mobility in patients undergoing total hip arthroplasty. Sensors, 23(1), 377. https://doi.org/10.3390/s23010377.

Wang, F., Liang, W., Afzal, H. M., Fan, A., Li, W., Dai, X., Liu, S., Hu, Y., Li, Z., & Yang, P. (2023). Estimation of lower limb joint angles and joint moments during different locomotive activities using the inertial measurement units and a hybrid deep learning model. Sensors, 23(22), 9039. https://doi.org/10.3390/s23229039.