The condition for dynamic stability in humans walking with feedback control.

The walking human body is mechanically unstable. Loss of stability and falling is more likely in certain groups of people, such as older adults or people with neuromotor impairments, as well as in certain situations, such as when experiencing conflicting or distracting sensory inputs. Stability during walking is often characterized biomechanically, by measures based on body dynamics and the base of support. Neural control of upright stability, on the other hand, does not factor into commonly used stability measures. Here we analyze stability of human walking accounting for both biomechanics and neural control, using a modeling approach. We define a walking system as a combination of biomechanics, using the well known inverted pendulum model, and neural control, using a proportional-derivative controller for foot placement based on the state of the center of mass at midstance. We analyze this system formally and show that for any choice of system parameters there is always one periodic orbit. We then determine when this periodic orbit is stable, i.e. how the neural control gain values have to be chosen for stable walking. Following the formal analysis, we use this model to make predictions about neural control gains and compare these predictions with the literature and existing experimental data. The model predicts that control gains should increase with decreasing cadence. This finding appears in agreement with literature showing stronger effects of visual or vestibular manipulations at different walking speeds.

Vertebral level specific modulation of paraspinal muscle activity based on vestibular signals during walking.

Evoking muscle responses by electrical vestibular stimulation (EVS) may help to understand the contribution of the vestibular system to postural control. Although paraspinal muscles play a role in postural stability, the vestibulo-muscular coupling of these muscles during walking has rarely been studied. This study aimed to investigate how vestibular signals affect paraspinal muscle activity at different vertebral levels during walking with preferred and narrow step width. Sixteen healthy participants were recruited. Participants walked on a treadmill for 8 min at 78 steps/min and 2.8 km/h, at two different step width, either with or without EVS. Bipolar electromyography was recorded bilaterally from the paraspinal muscles at eight vertebral levels from cervical to lumbar. Coherence, gain, and delay of EVS and EMG responses were determined. Significant EVS-EMG coupling (P < 0.01) was found at ipsilateral and/or contralateral heel strikes. This coupling was mirrored between left and right relative to the midline of the trunk and between the higher and lower vertebral levels, i.e. a peak occurred at ipsilateral heel strike at lower levels, whereas it occurred at contralateral heel strike at higher levels. EVS-EMG coupling only partially coincided with peak muscle activity. EVS-EMG coherence slightly, but not significantly, increased when walking with narrow steps. No significant differences were found in gain and phase between the vertebral levels or step width conditions. In summary, vertebral level specific modulation of paraspinal muscle activity based on vestibular signals might allow a fast, synchronized, and spatially co-ordinated response along the trunk during walking. KEY POINTS: Mediolateral stabilization of gait requires an estimate of the state of the body, which is affected by vestibular afference. During gait, the heavy trunk segment is controlled by phasic paraspinal muscle activity and in rodents the medial and lateral vestibulospinal tracts activate these muscles. To gain insight in vestibulospinal connections in humans and their role in gait, we recorded paraspinal surface EMG of cervical to lumbar paraspinal muscles, and characterized coherence, gain and delay between EMG and electrical vestibular stimulation, during slow walking. Vestibular stimulation caused phasic, vertebral level specific modulation of paraspinal muscle activity at delays of around 40 ms, which was mirrored between left, lower and right, upper vertebral levels. Our results indicate that vestibular afference causes fast, synchronized, and spatially co-ordinated responses of the paraspinal muscles along the trunk, that simultaneously contribute to stabilizing the centre of mass trajectory and to keeping the head upright.

Differences in the organization of the primary motor cortex in people with and without low back pain and associations with motor control and sensory tests.

Differences in organization of the primary motor cortex and altered trunk motor control (sensing, processing and motor output) have been reported in people with low back pain (LBP). Little is known to what extent these differences are related. We investigated differences in 1) organization of the primary motor cortex and 2) motor and sensory tests between people with and without LBP, and 3) investigated associations between the organization of the primary motor cortex and motor and sensory tests. We conducted a case-control study in people with (N=25) and without (N=25) LBP. The organization of the primary motor cortex (Center of Gravity (CoG) and Area of the cortical representation of trunk muscles) was assessed using neuronavigated transcranial magnetic stimulation, based on individual MRIs. Sensory tests (quantitative sensory testing, graphaesthesia, two-point discrimination threshold) and a motor test (spiral-tracking test) were assessed. Participants with LBP had a more lateral and lower location of the CoG and a higher temporal summation of pain. For all participants combined, better vibration test scores were associated with a more anterior, lateral, and lower CoG and a better two-point discrimination threshold was associated with a lower CoG. A small subset of variables showed significance. Although this aligns with the concept of altered organization of the primary motor cortex in LBP, there is no strong evidence of the association between altered organization of the primary motor cortex and motor and sensory test performance in LBP. Focusing on subgroup analyses regarding pain duration can be a topic for future research.

The effect of head orientation on vestibular signal-based modulation of paraspinal muscle activity during walking.

BACKGROUND: Vestibulospinal reflexes play a role in maintaining the upright posture of the trunk. Head orientation has been shown to modify the vestibulospinal reflexes during standing. This study investigated how vestibular signals affect paraspinal muscle activity during walking, and whether head orientation changes these effects. METHODS: Sixteen participants were instructed to walk on a treadmill for 8 min at 78 steps/min and 2.8 km/h in four conditions defined by the presence of electrical vestibular stimulation (EVS) and by head orientation (facing forward and facing leftward), while bipolar electromyography (EMG) was recorded bilaterally from the paraspinal muscles from cervical to lumbar levels. RESULTS: In both head orientations, significant phasic EVS-EMG coherence in the paraspinal muscles was observed at ipsilateral and/or contralateral heel strikes. Compared to walking with the head forward, a significant decrease was found in EVS-evoked responses (i.e., EVS-EMG coherence and gain) when participants walked with the leftward head orientation, with which EVS induced disturbance in the sagittal plane. This overall decrease can be explained by less need of feedback control for walking stabilization in the sagittal plane compared to in the frontal plane. The decrease in coherence was only significant at the left lower vertebral levels and at the right upper vertebral levels around left heel strikes. CONCLUSION: These findings confirm the contribution of the vestibular afferent signals to the control of paraspinal muscle activity during walking and indicate that this control is changed in response to different head orientations.

Effects of vestibular stimulation on gait stability when walking at different step widths.

Vestibular information modulates muscle activity during gait, presumably to contribute to stability. If this is the case, stronger effects of perturbing vestibular information on local dynamic stability of gait, a measure of the locomotor system's response to small, naturally occurring perturbations, can be expected for narrow-base walking (which needs more control) than for normal walking and smaller effects for wide-base walking (which needs less control). An important mechanism to stabilize gait is to coordinate foot placement to center of mass (CoM) state. Vestibular information most likely contributes to sensing this CoM state. We, therefore, expected that stochastic electrical vestibular stimulation (EVS) would decrease the correlation between foot placement and CoM state during the preceding swing phase. In 14 healthy participants, we measured the kinematics of the trunk (as a proxy of the CoM), and feet, while they walked on a treadmill in six conditions: control (usual step width), narrow-base, and wide-base, each with and without stochastic EVS (peak amplitude of 5 mA; RMS of ~ 1.2 mA; frequency band from 0 to 25 Hz). Stochastic EVS decreased local dynamic stability irrespective of step width. Foot placement correlated stronger with trunk motion during walking with EVS than without in the control condition. However, residual variance in foot placement was increased when walking with EVS, indicating less precise foot placement. Thus, a vestibular error signal leads to a decrease in gait stability and precision of foot placement, but these effects are not consistently modulated by step width.

Trunk Velocity Changes in Response to Physical Perturbations Are Potential Indicators of Gait Stability.

Response to challenging situations is important to avoid falls, especially after medial perturbations, which require active control. There is a lack of evidence on the relationship between the trunk's motion in response to perturbations and gait stability. Eighteen healthy adults walked on a treadmill at three speeds while receiving perturbations of three magnitudes. Medial perturbations were applied by translating the walking platform to the right at left heel contact. Trunk velocity changes in response to the perturbation were calculated and divided into the initial and the recovery phases. Gait stability after a perturbation was assessed using the margin of stability (MOS) at the first heel contact, MOS mean, and standard deviation for the first five strides after the perturbation onset. Faster speed and smaller perturbations led to a lower deviation of trunk velocity from the steady state, which can be interpreted as an improvement in response to the perturbation. Recovery was quicker after small perturbations. The MOS mean was associated with the trunk's motion in response to perturbations during the initial phase. Increasing walking speed may increase resistance to perturbations, while increasing the magnitude of perturbation leads to greater trunk motions. MOS is a useful marker of resistance to perturbations.

Reliability of IMU-Based Gait Assessment in Clinical Stroke Rehabilitation.

BACKGROUND: Gait is often impaired in people after stroke, restricting personal independence and affecting quality of life. During stroke rehabilitation, walking capacity is conventionally assessed by measuring walking distance and speed. Gait features, such as asymmetry and variability, are not routinely determined, but may provide more specific insights into the patient's walking capacity. Inertial measurement units offer a feasible and promising tool to determine these gait features. OBJECTIVE: We examined the test-retest reliability of inertial measurement units-based gait features measured in a two-minute walking assessment in people after stroke and while in clinical rehabilitation. METHOD: Thirty-one people after stroke performed two assessments with a test-retest interval of 24 h. Each assessment consisted of a two-minute walking test on a 14-m walking path. Participants were equipped with three inertial measurement units, placed at both feet and at the low back. In total, 166 gait features were calculated for each assessment, consisting of spatio-temporal (56), frequency (26), complexity (63), and asymmetry (14) features. The reliability was determined using the intraclass correlation coefficient. Additionally, the minimal detectable change and the relative minimal detectable change were computed. RESULTS: Overall, 107 gait features had good-excellent reliability, consisting of 50 spatio-temporal, 8 frequency, 36 complexity, and 13 symmetry features. The relative minimal detectable change of these features ranged between 0.5 and 1.5 standard deviations. CONCLUSION: Gait can reliably be assessed in people after stroke in clinical stroke rehabilitation using three inertial measurement units.

Modulation of soleus muscle H-reflexes and ankle muscle co-contraction with surface compliance during unipedal balancing in young and older adults.

This study aimed to assess modulation of lower leg muscle reflex excitability and co-contraction during unipedal balancing on compliant surfaces in young and older adults. Twenty healthy adults (ten aged 18-30 years and ten aged 65-80 years) were recruited. Soleus muscle H-reflexes were elicited by electrical stimulation of the tibial nerve, while participants stood unipedally on a robot-controlled balance platform, simulating different levels of surface compliance. In addition, electromyographic data (EMG) of soleus (SOL), tibialis anterior (TA), and peroneus longus (PL) and full-body 3D kinematic data were collected. The mean absolute center of mass velocity was determined as a measure of balance performance. Soleus H-reflex data were analyzed in terms of the amplitude related to the M wave and the background EMG activity 100 ms prior to the stimulation. The relative duration of co-contraction was calculated for soleus and tibialis anterior, as well as for peroneus longus and tibialis anterior. Center of mass velocity was significantly higher in older adults compared to young adults ([Formula: see text] and increased with increasing surface compliance in both groups ([Formula: see text]. The soleus H-reflex gain decreased with surface compliance in young adults [Formula: see text], while co-contraction increased [Formula: see text]. Older adults did not show such modulations, but showed overall lower H-reflex gains [Formula: see text] and higher co-contraction than young adults [Formula: see text]. These results suggest an overall shift in balance control from the spinal level to supraspinal levels in older adults, which also occurred in young adults when balancing at more compliant surfaces.

Does a Perturbation-Based Gait Intervention Enhance Gait Stability in Fall-Prone Stroke Survivors? A Pilot Study.

A recent review indicated that perturbation-based training (PBT) interventions are effective in reducing falls in older adults and patients with Parkinson's disease. It is unknown whether this type of intervention is effective in stroke survivors. We determined whether PBT can enhance gait stability in stroke survivors. A total of 10 chronic stroke survivors who experienced falls in the past 6 months participated in the PBT. Participants performed 10 training sessions over a 6-week period. The gait training protocol was progressive, and each training contained unexpected gait perturbations and expected gait perturbations. Evaluation of gait stability was performed by determining steady-state gait characteristics and daily-life gait characteristics. We previously developed fall prediction models for both gait assessment methods. We evaluated whether predicted fall risk was reduced after PBT according to both models. Steady-state gait characteristics significantly improved, and consequently, predicted fall risk was reduced after the PBT. However, daily-life gait characteristics did not change, and thus, predicted fall risk based on daily-life gait remained unchanged after the PBT. A PBT resulted in more stable gait on a treadmill and thus lower predicted fall risk. However, the more stable gait on the treadmill did not transfer to a more stable gait in daily life.

Evidence of splinting in low back pain? A systematic review of perturbation studies.

PURPOSE: The purpose of this systematic review was to assess whether LBP patients demonstrate signs of splinting by evaluating the reactions to unexpected mechanical perturbations in terms of (1) trunk muscle activity, (2) kinetic and (3) kinematic trunk responses and (4) estimated mechanical properties of the trunk. METHODS: The literature was systematically reviewed to identify studies that compared responses to mechanical trunk perturbations between LBP patients and healthy controls in terms of muscle activation, kinematics, kinetics, and/or mechanical properties. If more than four studies reported an outcome, the results of these studies were pooled. RESULTS: Nineteen studies were included, of which sixteen reported muscle activation, five kinematic responses, two kinetic responses, and two estimated mechanical trunk properties. We found evidence of a longer response time of muscle activation, which would be in line with splinting behaviour in LBP. No signs of splinting behaviour were found in any of the other outcome measures. CONCLUSIONS: We conclude that there is currently no convincing evidence for the presence of splinting behaviour in LBP patients, because we found no indications for splinting in terms of kinetic and kinematic responses to perturbation and derived mechanical properties of the trunk. Consistent evidence on delayed onsets of muscle activation in response to perturbations was found, but this may have other causes than splinting behaviour.

Slow maturation of planning in obstacle avoidance in humans.

Complex gait (e.g., obstacle avoidance) requires a higher cognitive load than simple steady-state gait, which is a more automated movement. The higher levels of the central nervous system, responsible for adjusting motor plans to complex gait, develop throughout childhood into adulthood. Therefore, we hypothesize that gait strategies in complex gait are likely to mature until adulthood as well. However, little is known about the maturation of complex gait from childhood into adolescence and adulthood. To address this issue, we investigated obstacle avoidance in forty-four 8- to 18-yr-old participants who walked at preferred speed along a 6-m walkway on which a planar obstacle (150% of step length, 1 m wide) was projected. Participants avoided the obstacle by stepping over this projection, while lower body kinematics were recorded. Results showed that step length and speed adjustments during successful obstacle avoidance were similar across all ages, even though younger children modified step width to a greater extent. Additionally, the younger children used larger maximal toe elevations and take-off distances than older children. Moreover, during unsuccessful trials, younger children deployed exaggerated take-off distances, which resulted in obstacle contact upon the consecutive heel strike. These results indicate that obstacle avoidance is not fully matured in younger children, and that the inability to plan precise foot placements is an important factor contributing to failures in obstacle avoidance.

Characteristics of daily life gait in fall and non fall-prone stroke survivors and controls.

BACKGROUND: Falls in stroke survivors can lead to serious injuries and medical costs. Fall risk in older adults can be predicted based on gait characteristics measured in daily life. Given the different gait patterns that stroke survivors exhibit it is unclear whether a similar fall-prediction model could be used in this group. Therefore the main purpose of this study was to examine whether fall-prediction models that have been used in older adults can also be used in a population of stroke survivors, or if modifications are needed, either in the cut-off values of such models, or in the gait characteristics of interest. METHODS: This study investigated gait characteristics by assessing accelerations of the lower back measured during seven consecutive days in 31 non fall-prone stroke survivors, 25 fall-prone stroke survivors, 20 neurologically intact fall-prone older adults and 30 non fall-prone older adults. We created a binary logistic regression model to assess the ability of predicting falls for each gait characteristic. We included health status and the interaction between health status (stroke survivors versus older adults) and gait characteristic in the model. RESULTS: We found four significant interactions between gait characteristics and health status. Furthermore we found another four gait characteristics that had similar predictive capacity in both stroke survivors and older adults. CONCLUSION: The interactions between gait characteristics and health status indicate that gait characteristics are differently associated with fall history between stroke survivors and older adults. Thus specific models are needed to predict fall risk in stroke survivors.

Gait asymmetry during early split-belt walking is related to perception of belt speed difference.

Gait adaptation is essential for humans to walk according to the different demands of the environment. Although locomotor adaptation has been studied in different contexts and in various patient populations, the mechanisms behind locomotor adaptation are still not fully understood. The aim of the present study was to test two opposing hypotheses about the control of split-belt walking, one based on avoidance of limping and the other on avoiding limb excursion asymmetry. We assessed how well cerebellar patients with focal lesions and healthy control participants could sense differences between belt speeds during split-belt treadmill walking and correlated this to split-belt adaptation parameters. The ability to perceive differences between belt speeds was similar between the cerebellar patients and the healthy controls. After combining all participants, we observed a significant inverse correlation between stance time symmetry and limb excursion symmetry during the early phase of split-belt walking. Participants who were better able to perceive belt speed differences (e.g., they had a lower threshold and hence were able to detect a smaller speed difference) walked with the smallest asymmetry in stance time and the largest asymmetry in limb excursion. Our data support the hypothesis that humans aim to minimize (temporal) limping rather than (spatial) limb excursion asymmetry when using their perception of belt speed differences in the early phase of adaptation to split-belt walking.

Adaptation and aftereffects of split-belt walking in cerebellar lesion patients.

To walk efficiently and stably on different surfaces under various constrained conditions, humans need to adapt their gait pattern substantially. Although the mechanisms behind locomotor adaptation are still not fully understood, the cerebellum is thought to play an important role. In this study we aimed to address the specific localization of cerebellar involvement in split-belt adaptation by comparing performance in patients with stable focal lesions after cerebellar tumor resection and in healthy controls. We observed that changes in symmetry of those parameters that were most closely related to interlimb coordination (such as step length and relative double stance time) were similar between healthy controls and cerebellar patients during and after split-belt walking. In contrast, relative stance times (proportions of stance in the gait cycle) were more asymmetric for the patient group than for the control group during the early phase of the post-split-belt condition. Patients who walked with more asymmetric relative stance times were more likely to demonstrate lesions in vermal lobules VI and Crus II. These results confirm that deficits in gait adaptation vary with ataxia severity and between patients with different types of cerebellar damage.

Gait parameters affecting the perception threshold of locomotor symmetry: comment on Lauzière, et al. (2014).

In a recent work on locomotor symmetry while walking on a split-belt treadmill, Lauzière and co-workers determined the perception threshold of gait symmetry in a sample of healthy elderly. In addition, they aimed to determine which particular gait parameters affect the symmetry of the perception threshold. Although only temporal and kinetic gait parameters were measured (and no kinematics), it was suggested that stance time symmetry is an important criterion that participants use to identify the threshold. Here it is argued that several other gait parameters could qualify equally well as main criteria used to identify the threshold and that these parameters should be taken into account in future studies.

Machine learning approaches to predict whether MEPs can be elicited via TMS.

BACKGROUND: Transcranial magnetic stimulation (TMS) is a valuable technique for assessing the function of the motor cortex and cortico-muscular pathways. TMS activates the motoneurons in the cortex, which after transmission along cortico-muscular pathways can be measured as motor-evoked potentials (MEPs). The position and orientation of the TMS coil and the intensity used to deliver a TMS pulse are considered central TMS setup parameters influencing the presence/absence of MEPs. NEW METHOD: We sought to predict the presence of MEPs from TMS setup parameters using machine learning. We trained different machine learners using either within-subject or between-subject designs. RESULTS: We obtained prediction accuracies of on average 77 % and 65 % with maxima up to up to 90 % and 72 % within and between subjects, respectively. Across the board, a bagging ensemble appeared to be the most suitable approach to predict the presence of MEPs. CONCLUSIONS: Although within a subject the prediction of MEPs via TMS setup parameter-based machine learning might be feasible, the limited accuracy between subjects suggests that the transfer of this approach to experimental or clinical research comes with significant challenges.

Assessment of stabilizing feedback control of walking: A tutorial.

Walking without falling requires stabilization of the trajectory of the body center of mass relative to the base of support. Model studies suggest that this requires active, feedback control, i.e., the nervous system must process sensory information on the state of the body to generate descending motor commands to the muscles to stabilize walking, especially in the mediolateral direction. Stabilization of bipedal gait is challenging and can be impaired in older and diseased individuals. In this tutorial, we illustrate how gait analysis can be used to assess the stabilizing feedback control of gait. We present methods ranging from those that require limited input data (e.g. position data of markers placed on the feet and pelvis only) to those that require full-body kinematics and electromyography. Analyses range from simple kinematics analyses to inverse dynamics. These methods assess stabilizing feedback control of human walking at three levels: 1) the level of center of mass movement and horizontal ground reaction forces, 2) the level of center of mass movement and foot placement and 3) the level of center of mass movement and the joint moments or muscle activity. We show how these can be calculated and provide a GitHub repository (https://github.com/VU-HMS/Tutorial-stabilizing-walking) which contains open access Matlab and Python code to calculate these. Finally, we discuss what information on feedback control can be learned from each of these.

The significance of frontal plane static alignment in anticipating dynamic knee moment among transtibial prosthesis users: A cross-sectional study.

BACKGROUND: In the process of transtibial prosthetic fitting, alignment is the process of positioning the prosthetic foot relative to the residual limb. Changes in frontal plane alignment can impact knee moments during walking, which can either cause or, when aligned properly, prevent injuries. However, clinical evaluation of dynamic knee moments is challenging, limiting prosthetists' insights into dynamic joint loading. Typically, knee joint loading is assessed in static stance using the knee moment arm as a proxy for subsequent dynamic alignment. It remains uncertain if static alignment accurately represents actual dynamics during walking. RESEARCH QUESTION: Is the frontal knee moment arm in stance predictive for the knee moment arm and external knee adduction moment during gait in transtibial bone-anchored prosthesis users? METHODS: In this cross-sectional study, twenty-seven unilateral transtibial bone-anchored prosthesis users underwent data acquisition on the M-Gait instrumented treadmill. Static and dynamic measurements were conducted, and knee moment arm and external knee adduction moment were calculated. Pearson's correlation and linear regression analyses were performed to examine relationships between static and dynamic knee moment arms and external knee adduction moments. RESULTS: The static knee moment arm showed significant associations with dynamic knee moment arm at the ground reaction force peaks (First: r=0.60, r2=35%, p<0.001; Second: r=0.62, r2=38%, p=0.001) and knee adduction moment (First: r=0.42, r2=17%, p=0.030; Second: r=0.59, r2=35%, p=0.001). A 1 mm between-subject difference in static knee moment arm corresponded, on average, with a 0.9% difference in knee adduction moment at the first peak and a 1.5% difference at the second peak of the ground reaction force. SIGNIFICANCE: While static alignment is important to optimize adduction moments during stance it may only partly mitigate excessive moments during gait. The fair correlation and limited percentage of explained variance underscores the importance of dynamic alignment in optimizing the body's dynamic load during walking.

External Validation and Further Exploration of Fall Prediction Models Based on Questionnaires and Daily-Life Trunk Accelerometry.

Ambulatory measurements of trunk accelerations can provide valuable insight into the amount and quality of daily life activities. Such information has been used to create models to identify individuals at high risk of falls. However, external validation of such prediction models is lacking, yet crucial for clinical implementation. We externally validated 3 previously described fall prediction models. Complete questionnaires and 1-week trunk acceleration data were obtained from 263 community-dwelling people (mean age 71.8 years, 68.1% female). To validate models, we first used the coefficients and optimal cutoffs from the original cohort, then recalibrated the original models, as well as optimized parameters based on our new cohort. Among all participants, 39.9% experienced falls during a 6-month follow-up. All models showed poor precision (0.20-0.49), poor sensitivity (0.32-0.58), and good specificity (0.45-0.89). Calibration of the original models had limited effect on model performance. Using coefficients and cutoffs optimized on the external cohort also had limited benefits. Lastly, the odds ratios in our cohort were different from those in the original cohort, which indicated that gait characteristics, except for the index of harmonicity ML (medial-lateral direction), were not statistically associated with falls. Fall risk prediction in our cohort was not as effective as in the original cohort. Recalibration as well as optimized model parameters resulted in a limited increase in accuracy. Fall prediction models are highly specific to the cohort studied. This highlights the need for large representative cohorts, preferably with an external validation cohort.

Exploring unsupervised feature extraction of IMU-based gait data in stroke rehabilitation using a variational autoencoder.

BACKGROUND: Variational AutoEncoders (VAE) might be utilized to extract relevant information from an IMU-based gait measurement by reducing the sensor data to a low-dimensional representation. The present study explored whether VAEs can reduce IMU-based gait data of people after stroke into a few latent features with minimal reconstruction error. Additionally, we evaluated the psychometric properties of the latent features in comparison to gait speed, by assessing 1) their reliability; 2) the difference in scores between people after stroke and healthy controls; and 3) their responsiveness during rehabilitation. METHODS: We collected test-retest and longitudinal two-minute walk-test data using an IMU from people after stroke in clinical rehabilitation, as well as from a healthy control group. IMU data were segmented into 5-second epochs, which were reduced to 12 latent-feature scores using a VAE. The between-day test-retest reliability of the latent features was assessed using ICC-scores. The differences between the healthy and the stroke group were examined using an independent t-test. Lastly, the responsiveness was determined as the number of individuals who significantly changed during rehabilitation. RESULTS: In total, 15,381 epochs from 107 people after stroke and 37 healthy controls were collected. The VAE achieved data reconstruction with minimal errors. Five latent features demonstrated good-to-excellent test-retest reliability. Seven latent features were significantly different between groups. We observed changes during rehabilitation for 21 and 20 individuals in latent-feature scores and gait speed, respectively. However, the direction of the change scores of the latent features was ambiguous. Only eleven individuals exhibited changes in both latent-feature scores and gait speed. CONCLUSION: VAEs can be used to effectively reduce IMU-based gait assessment to a concise set of latent features. Some latent features had a high test-retest reliability and differed significantly between healthy controls and people after stroke. Further research is needed to determine their clinical applicability.