From Movements to Metrics: Evaluating Explainable AI Methods in Skeleton-Based Human Activity Recognition.

The advancement of deep learning in human activity recognition (HAR) using 3D skeleton data is critical for applications in healthcare, security, sports, and human-computer interaction. This paper tackles a well-known gap in the field, which is the lack of testing in the applicability and reliability of XAI evaluation metrics in the skeleton-based HAR domain. We have tested established XAI metrics, namely faithfulness and stability on Class Activation Mapping (CAM) and Gradient-weighted Class Activation Mapping (Grad-CAM) to address this problem. This study introduces a perturbation method that produces variations within the error tolerance of motion sensor tracking, ensuring the resultant skeletal data points remain within the plausible output range of human movement as captured by the tracking device. We used the NTU RGB+D 60 dataset and the EfficientGCN architecture for HAR model training and testing. The evaluation involved systematically perturbing the 3D skeleton data by applying controlled displacements at different magnitudes to assess the impact on XAI metric performance across multiple action classes. Our findings reveal that faithfulness may not consistently serve as a reliable metric across all classes for the EfficientGCN model, indicating its limited applicability in certain contexts. In contrast, stability proves to be a more robust metric, showing dependability across different perturbation magnitudes. Additionally, CAM and Grad-CAM yielded almost identical explanations, leading to closely similar metric outcomes. This suggests a need for the exploration of additional metrics and the application of more diverse XAI methods to broaden the understanding and effectiveness of XAI in skeleton-based HAR.

Development and Validation of a Deep Learning Method to Predict Cerebral Palsy From Spontaneous Movements in Infants at High Risk.

Importance: Early identification of cerebral palsy (CP) is important for early intervention, yet expert-based assessments do not permit widespread use, and conventional machine learning alternatives lack validity. Objective: To develop and assess the external validity of a novel deep learning-based method to predict CP based on videos of infants' spontaneous movements at 9 to 18 weeks' corrected age. Design, Setting, and Participants: This prognostic study of a deep learning-based method to predict CP at a corrected age of 12 to 89 months involved 557 infants with a high risk of perinatal brain injury who were enrolled in previous studies conducted at 13 hospitals in Belgium, India, Norway, and the US between September 10, 2001, and October 25, 2018. Analysis was performed between February 11, 2020, and September 23, 2021. Included infants had available video recorded during the fidgety movement period from 9 to 18 weeks' corrected age, available classifications of fidgety movements ascertained by the general movement assessment (GMA) tool, and available data on CP status at 12 months' corrected age or older. A total of 418 infants (75.0%) were randomly assigned to the model development (training and internal validation) sample, and 139 (25.0%) were randomly assigned to the external validation sample (1 test set). Exposure: Video recording of spontaneous movements. Main Outcomes and Measures: The primary outcome was prediction of CP. Deep learning-based prediction of CP was performed automatically from a single video. Secondary outcomes included prediction of associated functional level and CP subtype. Sensitivity, specificity, positive and negative predictive values, and accuracy were assessed. Results: Among 557 infants (310 [55.7%] male), the median (IQR) corrected age was 12 (11-13) weeks at assessment, and 84 infants (15.1%) were diagnosed with CP at a mean (SD) age of 3.4 (1.7) years. Data on race and ethnicity were not reported because previous studies (from which the infant samples were derived) used different study protocols with inconsistent collection of these data. On external validation, the deep learning-based CP prediction method had sensitivity of 71.4% (95% CI, 47.8%-88.7%), specificity of 94.1% (95% CI, 88.2%-97.6%), positive predictive value of 68.2% (95% CI, 45.1%-86.1%), and negative predictive value of 94.9% (95% CI, 89.2%-98.1%). In comparison, the GMA tool had sensitivity of 70.0% (95% CI, 45.7%-88.1%), specificity of 88.7% (95% CI, 81.5%-93.8%), positive predictive value of 51.9% (95% CI, 32.0%-71.3%), and negative predictive value of 94.4% (95% CI, 88.3%-97.9%). The deep learning method achieved higher accuracy than the conventional machine learning method (90.6% [95% CI, 84.5%-94.9%] vs 72.7% [95% CI, 64.5%-79.9%]; P < .001), but no significant improvement in accuracy was observed compared with the GMA tool (85.9%; 95% CI, 78.9%-91.3%; P = .11). The deep learning prediction model had higher sensitivity among infants with nonambulatory CP (100%; 95% CI, 63.1%-100%) vs ambulatory CP (58.3%; 95% CI, 27.7%-84.8%; P = .02) and spastic bilateral CP (92.3%; 95% CI, 64.0%-99.8%) vs spastic unilateral CP (42.9%; 95% CI, 9.9%-81.6%; P < .001). Conclusions and Relevance: In this prognostic study, a deep learning-based method for predicting CP at 9 to 18 weeks' corrected age had predictive accuracy on external validation, which suggests possible avenues for using deep learning-based software to provide objective early detection of CP in clinical settings.

Towards human-level performance on automatic pose estimation of infant spontaneous movements.

Assessment of spontaneous movements can predict the long-term developmental disorders in high-risk infants. In order to develop algorithms for automated prediction of later disorders, highly precise localization of segments and joints by infant pose estimation is required. Four types of convolutional neural networks were trained and evaluated on a novel infant pose dataset, covering the large variation in 1424 videos from a clinical international community. The localization performance of the networks was evaluated as the deviation between the estimated keypoint positions and human expert annotations. The computational efficiency was also assessed to determine the feasibility of the neural networks in clinical practice. The best performing neural network had a similar localization error to the inter-rater spread of human expert annotations, while still operating efficiently. Overall, the results of our study show that pose estimation of infant spontaneous movements has a great potential to support research initiatives on early detection of developmental disorders in children with perinatal brain injuries by quantifying infant movements from video recordings with human-level performance.

Performance Analysis in Ski Jumping with a Differential Global Navigation Satellite System and Video-Based Pose Estimation.

This study investigated the explanatory power of a sensor fusion of two complementary methods to explain performance and its underlying mechanisms in ski jumping. A differential Global Navigation Satellite System (dGNSS) and a markerless video-based pose estimation system (PosEst) were used to measure the kinematics and kinetics from the start of the in-run to the landing. The study had two aims; firstly, the agreement between the two methods was assessed using 16 jumps by athletes of national level from 5 m before the take-off to 20 m after, where the methods had spatial overlap. The comparison revealed a good agreement from 5 m after the take-off, within the uncertainty of the dGNSS (±0.05m). The second part of the study served as a proof of concept of the sensor fusion application, by showcasing the type of performance analysis the systems allows. Two ski jumps by the same ski jumper, with comparable external conditions, were chosen for the case study. The dGNSS was used to analyse the in-run and flight phase, while the PosEst system was used to analyse the take-off and the early flight phase. The proof-of-concept study showed that the methods are suitable to track the kinematic and kinetic characteristics that determine performance in ski jumping and their usability in both research and practice.

Classical Machine Learning Versus Deep Learning for the Older Adults Free-Living Activity Classification.

Physical activity has a strong influence on mental and physical health and is essential in healthy ageing and wellbeing for the ever-growing elderly population. Wearable sensors can provide a reliable and economical measure of activities of daily living (ADLs) by capturing movements through, e.g., accelerometers and gyroscopes. This study explores the potential of using classical machine learning and deep learning approaches to classify the most common ADLs: walking, sitting, standing, and lying. We validate the results on the ADAPT dataset, the most detailed dataset to date of inertial sensor data, synchronised with high frame-rate video labelled data recorded in a free-living environment from older adults living independently. The findings suggest that both approaches can accurately classify ADLs, showing high potential in profiling ADL patterns of the elderly population in free-living conditions. In particular, both long short-term memory (LSTM) networks and Support Vector Machines combined with ReliefF feature selection performed equally well, achieving around 97% F-score in profiling ADLs.

Predicting Advanced Balance Ability and Mobility with an Instrumented Timed Up and Go Test.

Extensive test batteries are often needed to obtain a comprehensive picture of a person's functional status. Many test batteries are not suitable for active and healthy adults due to ceiling effects, or require a lot of space, time, and training. The Community Balance and Mobility Scale (CBMS) is considered a gold standard for this population, but the test is complex, as well as time- and resource intensive. There is a strong need for a faster, yet sensitive and robust test of physical function in seniors. We sought to investigate whether an instrumented Timed Up and Go (iTUG) could predict the CBMS score in 60 outpatients and healthy community-dwelling seniors, where features of the iTUG were predictive, and how the prediction of CBMS with the iTUG compared to standard clinical tests. A partial least squares regression analysis was used to identify latent components explaining variation in CBMS total score. The model with iTUG features was able to predict the CBMS total score with an accuracy of 85.2% (84.9-85.5%), while standard clinical tests predicted 82.5% (82.2-82.8%) of the score. These findings suggest that a fast and easily administered iTUG could be used to predict CBMS score, providing a valuable tool for research and clinical care.

Machine Learning of Infant Spontaneous Movements for the Early Prediction of Cerebral Palsy: A Multi-Site Cohort Study.

BACKGROUND: Early identification of cerebral palsy (CP) during infancy will provide opportunities for early therapies and treatments. The aim of the present study was to present a novel machine-learning model, the Computer-based Infant Movement Assessment (CIMA) model, for clinically feasible early CP prediction based on infant video recordings. METHODS: The CIMA model was designed to assess the proportion (%) of CP risk-related movements using a time-frequency decomposition of the movement trajectories of the infant's body parts. The CIMA model was developed and tested on video recordings from a cohort of 377 high-risk infants at 9-15 weeks corrected age to predict CP status and motor function (ambulatory vs. non-ambulatory) at mean 3.7 years age. The performance of the model was compared with results of the general movement assessment (GMA) and neonatal imaging. RESULTS: The CIMA model had sensitivity (92.7%) and specificity (81.6%), which was comparable to observational GMA or neonatal cerebral imaging for the prediction of CP. Infants later found to have non-ambulatory CP had significantly more CP risk-related movements (median: 92.8%, p = 0.02) compared with those with ambulatory CP (median: 72.7%). CONCLUSION: The CIMA model may be a clinically feasible alternative to observational GMA.

Improved Prediction of Falls in Community-Dwelling Older Adults Through Phase-Dependent Entropy of Daily-Life Walking.

Age and age-related diseases have been suggested to decrease entropy of human gait kinematics, which is thought to make older adults more susceptible to falls. In this study we introduce a new entropy measure, called phase-dependent generalized multiscale entropy (PGME), and test whether this measure improves fall-risk prediction in community-dwelling older adults. PGME can assess phase-dependent changes in the stability of gait dynamics that result from kinematic changes in events such as heel strike and toe-off. PGME was assessed for trunk acceleration of 30 s walking epochs in a re-analysis of 1 week of daily-life activity data from the FARAO study, originally described by van Schooten et al. (2016). The re-analyzed data set contained inertial sensor data from 52 single- and 46 multiple-time prospective fallers in a 6 months follow-up period, and an equal number of non-falling controls matched by age, weight, height, gender, and the use of walking aids. The predictive ability of PGME for falls was assessed using a partial least squares regression. PGME had a superior predictive ability of falls among single-time prospective fallers when compared to the other gait features. The single-time fallers had a higher PGME (p < 0.0001) of their trunk acceleration at 60% of their step cycle when compared with non-fallers. No significant differences were found between PGME of multiple-time fallers and non-fallers, but PGME was found to improve the prediction model of multiple-time fallers when combined with other gait features. These findings suggest that taking into account phase-dependent changes in the stability of the gait dynamics has additional value for predicting falls in older people, especially for single-time prospective fallers.

Fractional Stability of Trunk Acceleration Dynamics of Daily-Life Walking: Toward a Unified Concept of Gait Stability.

Over the last decades, various measures have been introduced to assess stability during walking. All of these measures assume that gait stability may be equated with exponential stability, where dynamic stability is quantified by a Floquet multiplier or Lyapunov exponent. These specific constructs of dynamic stability assume that the gait dynamics are time independent and without phase transitions. In this case the temporal change in distance, d(t), between neighboring trajectories in state space is assumed to be an exponential function of time. However, results from walking models and empirical studies show that the assumptions of exponential stability break down in the vicinity of phase transitions that are present in each step cycle. Here we apply a general non-exponential construct of gait stability, called fractional stability, which can define dynamic stability in the presence of phase transitions. Fractional stability employs the fractional indices, α and ß, of differential operator which allow modeling of singularities in d(t) that cannot be captured by exponential stability. The fractional stability provided an improved fit of d(t) compared to exponential stability when applied to trunk accelerations during daily-life walking in community-dwelling older adults. Moreover, using multivariate empirical mode decomposition surrogates, we found that the singularities in d(t), which were well modeled by fractional stability, are created by phase-dependent modulation of gait. The new construct of fractional stability may represent a physiologically more valid concept of stability in vicinity of phase transitions and may thus pave the way for a more unified concept of gait stability.

Performance Evaluation of State of the Art Systems for Physical Activity Classification of Older Subjects Using Inertial Sensors in a Real Life Scenario: A Benchmark Study.

The popularity of using wearable inertial sensors for physical activity classification has dramatically increased in the last decade due to their versatility, low form factor, and low power requirements. Consequently, various systems have been developed to automatically classify daily life activities. However, the scope and implementation of such systems is limited to laboratory-based investigations. Furthermore, these systems are not directly comparable, due to the large diversity in their design (e.g., number of sensors, placement of sensors, data collection environments, data processing techniques, features set, classifiers, cross-validation methods). Hence, the aim of this study is to propose a fair and unbiased benchmark for the field-based validation of three existing systems, highlighting the gap between laboratory and real-life conditions. For this purpose, three representative state-of-the-art systems are chosen and implemented to classify the physical activities of twenty older subjects (76.4 ± 5.6 years). The performance in classifying four basic activities of daily life (sitting, standing, walking, and lying) is analyzed in controlled and free living conditions. To observe the performance of laboratory-based systems in field-based conditions, we trained the activity classification systems using data recorded in a laboratory environment and tested them in real-life conditions in the field. The findings show that the performance of all systems trained with data in the laboratory setting highly deteriorates when tested in real-life conditions, thus highlighting the need to train and test the classification systems in the real-life setting. Moreover, we tested the sensitivity of chosen systems to window size (from 1 s to 10 s) suggesting that overall accuracy decreases with increasing window size. Finally, to evaluate the impact of the number of sensors on the performance, chosen systems are modified considering only the sensing unit worn at the lower back. The results, similarly to the multi-sensor setup, indicate substantial degradation of the performance when laboratory-trained systems are tested in the real-life setting. This degradation is higher than in the multi-sensor setup. Still, the performance provided by the single-sensor approach, when trained and tested with real data, can be acceptable (with an accuracy above 80%).

The FARSEEING real-world fall repository: a large-scale collaborative database to collect and share sensor signals from real-world falls.

BACKGROUND: Real-world fall events objectively measured by body-worn sensors can improve the understanding of fall events in older people. However, these events are rare and hence challenging to capture. Therefore, the FARSEEING (FAll Repository for the design of Smart and sElf-adaptive Environments prolonging Independent livinG) consortium and associated partners started to build up a meta-database of real-world falls. RESULTS: Between January 2012 and December 2015 more than 300 real-world fall events have been recorded. This is currently the largest collection of real-world fall data recorded with inertial sensors. A signal processing and fall verification procedure has been developed and applied to the data. Since the end of 2015, 208 verified real-world fall events are available for analyses. The fall events have been recorded within several studies, with different methods, and in different populations. All sensor signals include at least accelerometer measurements and 58 % additionally include gyroscope and magnetometer measurements. The collection of data is ongoing and open to further partners contributing with fall signals. The FARSEEING consortium also aims to share the collected real-world falls data with other researchers on request. CONCLUSIONS: The FARSEEING meta-database will help to improve the understanding of falls and enable new approaches in fall risk assessment, fall prevention, and fall detection in both aging and disease.

Exergaming in Older Adults: Movement Characteristics While Playing Stepping Games.

Despite frequent use of exergames in intervention studies to improve physical function in older adults, we lack knowledge about the movements performed during exergaming. This causes difficulties for interpreting results of intervention studies and drawing conclusions about the efficacy of exergames to exercise specific functions important for the elderly population. The aim of the current study was to investigate whether game and game level affect older adults' stepping and upper body movements while playing stepping exergames. A 3D-motion capture experiment was performed with 20 elderly (12 women and 8 men; age range 65-90 years), playing two exergames, The Mole from SilverFit and LightRace in YourShape: Fitness Evolved, on two difficulty levels, with five 1-min trials for each game and level. Reflective markers were placed on bases of first toe, heels, and lower back. Movement characteristics were analyzed with a linear mixed model. Results indicated that both game and game level affected movement characteristics. Participants took shorter steps and had lower step velocity when playing The Mole compared to LightRace, while The Mole prompted more variation in step length and step velocity. Compared to LightRace, The Mole elicited larger upper body movements in both ML- and AP-directions and participants' feet and upper body covered a larger area. Increasing difficulty level from Easy to Medium resulted in overall decrease of movement, except for number of steps and step speed when playing LightRace. Even with only two games, two levels, and five trials at each, this study indicates that the choice of exergame is not indifferent when aiming to exercise specific functions in older adults and that exergames need to be chosen and designed carefully based on the goals of the intervention.

The complexity of daily life walking in older adult community-dwelling fallers and non-fallers.

Complexity of human physiology and physical behavior has been suggested to decrease with aging and disease and make older adults more susceptible to falls. The present study investigates complexity in daily life walking in community-dwelling older adult fallers and non-fallers measured by a 3D inertial accelerometer sensor fixed to the lower back. Complexity was expressed using new metrics of entropy: refined composite multiscale entropy (RCME) and refined multiscale permutation entropy (RMPE). The study re-analyses data of 3 days daily-life activity originally described by Weiss et al. (2013). The data set contains inertial sensor data from 39 older persons reporting less than 2 falls and 32 older persons reporting two or more falls during the previous year. The RCME and the RMPE were derived for trunk acceleration and velocity signals from walking epochs of 50s using mean and variance coarse graining of the signals. Discriminant abilities of the entropy metrics were assessed using a partial least square discriminant analysis. Both RCME and RMPE successfully distinguished between the daily-life walking of the fallers and non-fallers (AUC>0.8) and performed better than the 35 conventional gait features investigated by Weiss et al. (2013). Higher complexity was found in the vertical and mediolateral directions in the non-fallers for both entropy metrics. These findings suggest that RCME and RMPE can be used to improve the assessment of fall risk in older people.

A comparison study of local dynamic stability measures of daily life walking in older adult community-dwelling fallers and non-fallers.

In the present study we compared the performance of three different estimations of local dynamic stability λ to distinguish between the dynamics of the daily-life walking of elderly fallers and non-fallers. The study re-analyses inertial sensor data of 3-days daily-life activity originally described by Weiss et al. (2013). The data set contains inertial sensor data from 39 older persons who reported less than 2 falls and 31 older persons who reported two or more falls the previous year. 3D-acceleration and 3D-velocity signals from walking epochs of 50s were used to reconstruct a state space using three different methods. Local dynamic stability was estimated with the algorithms proposed by Rosenstein et al. (1993), Kantz (1994), and Ihlen et al. (2012a). Median λs assessed by Ihlen׳s and Kantz׳ algorithms discriminated better between elderly fallers and non-fallers (highest AUC=0.75 and 0.73) than Rosenstein׳s algorithm (highest AUC=0.59). The present results suggest that the ability of λ to distinguish between fallers and non-fallers is dependent on the parameter setting of the chosen algorithm. Further replication in larger samples of community-dwelling older persons and different patient groups is necessary before including the suggested parameter settings in fall risk assessment and prediction models.

Video analysis validation of a real-time physical activity detection algorithm based on a single waist mounted tri-axial accelerometer sensor.

We have validated a real-time activity classification algorithm based on monitoring by a body worn system which is potentially suitable for low-power applications on a relatively computationally lightweight processing unit. The algorithm output was validated using annotation data generated from video recordings of 20 elderly volunteers performing both a semi-structured protocol and a free-living protocol. The algorithm correctly identified sitting 75.1% of the time, standing 68.8% of the time, lying 50.9% of the time, and walking and other upright locomotion 82.7% of the time. This is one of the most detailed validations of a body worn sensor algorithm to date and offers an insight into the challenges of developing a real-time physical activity classification algorithm for a tri-axial accelerometer based sensor worn at the waist.

Fall detection algorithms for real-world falls harvested from lumbar sensors in the elderly population: a machine learning approach.

Automatic fall detection will promote independent living and reduce the consequences of falls in the elderly by ensuring people can confidently live safely at home for linger. In laboratory studies inertial sensor technology has been shown capable of distinguishing falls from normal activities. However less than 7% of fall-detection algorithm studies have used fall data recorded from elderly people in real life. The FARSEEING project has compiled a database of real life falls from elderly people, to gain new knowledge about fall events and to develop fall detection algorithms to combat the problems associated with falls. We have extracted 12 different kinematic, temporal and kinetic related features from a data-set of 89 real-world falls and 368 activities of daily living. Using the extracted features we applied machine learning techniques and produced a selection of algorithms based on different feature combinations. The best algorithm employs 10 different features and produced a sensitivity of 0.88 and a specificity of 0.87 in classifying falls correctly. This algorithm can be used distinguish real-world falls from normal activities of daily living in a sensor consisting of a tri-axial accelerometer and tri-axial gyroscope located at L5.

The Discriminant Value of Phase-Dependent Local Dynamic Stability of Daily Life Walking in Older Adult Community-Dwelling Fallers and Nonfallers.

The present study compares phase-dependent measures of local dynamic stability of daily life walking with 35 conventional gait features in their ability to discriminate between community-dwelling older fallers and nonfallers. The study reanalyzes 3D-acceleration data of 3-day daily life activity from 39 older people who reported less than 2 falls during one year and 31 who reported two or more falls. Phase-dependent local dynamic stability was defined for initial perturbation at 0%, 20%, 40%, 60%, and 80% of the step cycle. A partial least square discriminant analysis (PLS-DA) was used to compare the discriminant abilities of phase-dependent local dynamic stability with the discriminant abilities of 35 conventional gait features. The phase-dependent local dynamic stability λ at 0% and 60% of the step cycle discriminated well between fallers and nonfallers (AUC = 0.83) and was significantly larger (p < 0.01) for the nonfallers. Furthermore, phase-dependent λ discriminated as well between fallers and nonfallers as all other gait features combined. The present result suggests that phase-dependent measures of local dynamic stability of daily life walking might be of importance for further development in early fall risk screening tools.

Association between vestibulo-ocular reflex suppression, balance, gait, and fall risk in ageing and neurodegenerative disease: protocol of a one-year prospective follow-up study.

BACKGROUND: Falls frequency increases with age and particularly in neurogeriatric cohorts. The interplay between eye movements and locomotion may contribute substantially to the occurrence of falls, but is hardly investigated. This paper provides an overview of current approaches to simultaneously measure eye and body movements, particularly for analyzing the association of vestibulo-ocular reflex (VOR) suppression, postural deficits and falls in neurogeriatric risk cohorts. Moreover, VOR suppression is measured during head-fixed target presentation and during gaze shifting while postural control is challenged. Using these approaches, we aim at identifying quantitative parameters of eye-head-coordination during postural balance and gait, as indicators of fall risk. METHODS/DESIGN: Patients with Progressive Supranuclear Palsy (PSP) or Parkinson's disease (PD), age- and sex-matched healthy older adults, and a cohort of young healthy adults will be recruited. Baseline assessment will include a detailed clinical assessment, covering medical history, neurological examination, disease specific clinical rating scales, falls-related self-efficacy, activities of daily living, neuro-psychological screening, assessment of mobility function and a questionnaire for retrospective falls. Moreover, participants will simultaneously perform eye and head movements (fixating a head-fixed target vs. shifting gaze to light emitting diodes in order to quantify vestibulo-ocular reflex suppression ability) under different conditions (sitting, standing, or walking). An eye/head tracker synchronized with a 3-D motion analysis system will be used to quantify parameters related to eye-head-coordination, postural balance, and gait. Established outcome parameters related to VOR suppression ability (e.g., gain, saccadic reaction time, frequency of saccades) and motor related fall risk (e.g., step-time variability, postural sway) will be calculated. Falls will be assessed prospectively over 12 months via protocols and monthly telephone interviews. DISCUSSION: This study protocol describes an experimental setup allowing the analysis of simultaneously assessed eye, head and body movements. Results will improve our understanding of the influence of the interplay between eye, head and body movements on falls in geriatric high-risk cohorts.

Age-related changes in inter-joint coordination during walking.

Existing methods to assess inter-joint coordination in human walking have several important weaknesses. These methods are unable to define 1) the instantaneous changes in coordination within the stride cycle, 2) coordination between multiple joints, or 3) the coupling strength of joint rotations rather than their phase relationships. The present paper introduces a new method called generalized wavelet coherence analysis (GWCA) that solves these three fundamental limitations of previous methods. GWCA combines wavelet coherence analysis with a matrix correlation method to define instantaneous correlation coefficients as the coupling strength for an arbitrary number of joint rotations. The main purpose of the present study is to develop GWCA to quantify inter-joint coordination and thereby assess age-related differences in the coordination of human gaits. Nine young and 19 healthy older persons walked 5 min on a treadmill at three different gait speeds. Joint rotations of the lower extremities were assessed by a Vicon three-dimensional motion capture system. The main results indicated that the older group had significant weaker correlations (t-tests: P < 0.0001) in the preswing phase compared with the younger group for all gait speeds. The age-related differences in inter-joint coordination were more pronounced than the age-related differences in rotations of the individual joints. The intra-stride changes in inter-joint coordination were in agreement with recent findings of intra-stride modulations in neural activity in the sensorimotor cortex. Thus change in the inter-joint coordination assessed by GWCA might be an early indicator of functional decline.