Exoskeletons are frequently explored for their efficacy in rehabilitation and in assisting in daily activities in people with motor disorders, yet relatively few have convincing evidence for use, nor do many aim to personalize assistance for maximum benefit of each user. Here, we describe a cable-driven ankle exoskeleton that assists persons with dropfoot and excessive inversion after stroke. We propose a multi-objective human-in-the-loop optimization that simultaneously adjusts exoskeleton assistive profiles to minimize two independent gait quality indices: foot kinematics deviation and step length asymmetry. This optimization proposes, rather than a single solution, a group of solutions that minimize both indices. We present experimental findings from one participant with chronic stroke, specifically the feasibility of the device in assisting ankle motion in two anatomical planes, the resulting gait pattern and individualized optimal assistive profiles. Following the design goals, the exoskeleton improved the dropfoot and excessive inversion gait pattern; with the exoskeleton, the participant's foot inclination angle at initial contact increased from -2 degrees to 10 degrees, ankle inversion angle decreased from 2 degrees to 0 degrees, foot clearance during swing increased from 31 mm to 44 mm, and step length asymmetry decreased from 12% to 6%. The optimal solutions are an inventory of profiles from which the user can choose based on priorities. We believe the device and approach have a large potential in achieving individualized assistance.

Wearable robotic exoskeletons are frequently explored for their efficacy in rehabilitation and in assistance in daily activities in people with motor disorders, yet relatively few have convincing evidence for use. Here we describe a cable-driven ankle exoskeleton that provides assistance to the ankle in sagittal and frontal planes simultaneously, aimed for persons with dropfoot and excessive inversion after e.g. stroke. In this study, we propose a multi-objective human-in-the-loop optimization that adjusts exoskeleton control parameters to improve two independent gait quality measures, specifically foot segment kinematics and step length symmetry. We illustrate how the identified solutions represent a balance between the two objectives.

Introduction: Musculoskeletal simulations can guide the search for optimal strategies to assist motion and reveal causal relationships between assistive moments and muscle dynamics. Assistive devices such as exoskeletons can complement muscle forces based on various aims, such as minimum muscle effort or maximal force distribution. In this study, we present a simulation framework to systematically identify optimal assistance, formulated as a bilevel optimization in a single inverse simulation scheme that seeks optimal assistive moments that fulfill different assistive device aims. Methods: Bilevel optimization of assistive moment was structured as an inner optimization problem to solve the muscle redundancy problem nested within an outer optimization problem that executes the inner problem iteratively, seeking an assistive moment that best satisfies the assistive aim. We used this framework to predict optimal ankle plantarflexion, hip extension, hip flexion, and hip abduction assistance, for three different aims: minimal muscle activations, minimal metabolic rates, and minimal muscle moments. Experimental data from twelve participants walking at preferred speed were used in this study. Results: We found that the optimal moment trajectory is unique for a given assistive aim; i.e., the assistive aim matters. Differences in the assistive trajectories are explained at the muscle level, and as active and passive force contributions to the net muscle moments and muscle mechanical work. Interestingly, the assistive moments for minimal metabolic rates predicted an assistance period and peak timing similar to those reported from experimental studies. Conclusions: Our findings suggest that explicit assistive aim formulation is required to investigate human-device interaction under optimal assistance.

Machine learning (ML) has emerged as a powerful tool to analyze gait data, yet the “black-box” nature of many ML models hinders their clinical application. Explainable artificial intelligence (XAI) promises to enhance the interpretability and transparency of ML models, making them more suitable for clinical decision-making. This systematic review, registered on PROSPERO (CRD42024622752), assessed the application of XAI in gait analysis by examining its methods, performance, and potential for clinical utility. A comprehensive search across four electronic databases yielded 3676 unique records, of which 31 studies met inclusion criteria. These studies were categorized into model-agnostic (n = 16), model-specific (n = 12), and hybrid (n = 3) interpretability approaches. Most applied local interpretation methods such as SHAP and LIME, while others used Grad-CAM, attention mechanisms, and Layer-wise Relevance Propagation. Clinical populations studied included Parkinson’s disease, stroke, sarcopenia, cerebral palsy, and musculoskeletal disorders. Reported outcomes highlighted biomechanically relevant features such as stride length and joint angles as key discriminators of pathological gait. Overall, the findings demonstrate that XAI can bridge the gap between predictive performance and interpretability, but significant challenges remain in standardization, validation, and balancing accuracy with transparency. Future research should refine XAI frameworks and assess their real-world clinical applicability across diverse gait disorders.

Background: Individuals who have experienced spinal cord injury (SCI) may exhibit various muscle-related neurophysiological adaptations, including alterations in motor unit (MU) size and firing behavior. However, due to the technical challenges of in vivo measurement, our understanding of the alterations in the electrophysiological parameters of these MUs remains limited. This study proposed an integrated approach using high-density electromyography (HD-EMG) decomposition and motor neuron (MN) modelling to estimate the intrinsic properties of MUs in vivo and investigated alterations of these properties in persons with SCI.

Methods: HD-EMG signals were recorded during submaximal isometric dorsiflexion and plantar flexion tasks on tibialis anterior (TA), soleus, and gastrocnemius medialis muscles from twenty-six participants with SCI and eighteen non-disabled controls. The HD-EMG signals were subsequently decomposed into MN spike trains and the common synaptic input to the MN pool was estimated. A simplified leaky integrate-and-fire neuron model was then used to simulate MN spiking trains, with soma size and inert period as tunning parameters, which are crucial for MU recruitment and firing patterns, respectively. These parameters were estimated by fitting the instantaneous discharge frequencies of decomposed and simulated spike trains via a genetic algorithm.

Results: The results showed a prolonged inert period in the TA of the persons with SCI. This finding suggested that the MUs in the TA have a slower recovery period before becoming excitable again, which may result in a lower firing rate of MUs in the TA muscle. No significant differences were observed in the soleus and gastrocnemius medialis muscles between the SCI and control groups for either the soma size or inert period parameters.

Conclusions: The simplified leaky integrate-and-fire model exhibited robustness in estimating MN parameters in vivo, offering valuable insights into personalized MU behavior monitoring. To the best knowledge of authors, this is the first study to combine HD-EMG and MU modeling to investigate MU electrophysiological changes in persons with SCI in vivo. This novel approach offers a comprehensive understanding of MU properties adaptations following neurological disorders and informs the development of novel rehabilitation strategies.

Common optimization approaches for solving the muscle redundancy problem in musculoskeletal simulations can predict shoulder contact forces that either violate or barely satisfy joint stability requirements, with force directions falling outside or near the perimeter of the glenoid cavity. In this study, several glenohumeral stability formulations were tested against in vivo measurements of glenohumeral contact forces from the Orthoload dataset on one participant data in lateral, posterior, and anterior dumbbell raises. The investigated formulations either constrained the contact force direction to remain within different shapes of a stability perimeter, or added a penalty term that discouraged contact force directions from deviating from the glenoid cavity center. All stability formulations predicted contact force magnitudes that agreed relatively well to the in vivo measured forces except for the strictest formulation that constrained the joint contact force directly to the glenoid cavity center. Constraint and conditional penalty models estimated force vectors that largely lay along the perimeters. Continuous penalty models estimated relatively more accurate contact force directions within the glenoid cavity than constraint models. Our findings support the proposed penalty formulations as more reasonable and accurate than other investigated existing glenohumeral stability formulations.

Wearable robotic exoskeletons are frequently explored for their efficacy in physical rehabilitation and for assistance in daily activities in people with motor disorders, yet relatively few have convincing evidence for use. The concept of human-in-the-loop optimization has been used to identify ideal exoskeleton assistive torques based on measured individual performance metrics, whereas few studies report optimizing several performance metrics simultaneously. In this study, we propose a multi-objective human-in-the-loop optimization that adjusts exoskeleton assistive profiles to improve several gait quality measures, specifically foot segment kinematics and step length symmetry. A preliminary proof-of-concept evaluation was conducted with five non-disabled participants, where a weighted shoe was used to simulate the gait deviations associated with dropfoot and excessive inversion. The gait quality metrics improved more for each new generation. With optimal assistive solutions, foot segment kinematics improved (10 mm higher foot clearance height and a 7∘ increase in inclination angle) and step length became more symmetric (asymmetry reduced from 9% to 2%), compared to simulated impairment. Within this set of solutions, each represents a unique balance between the two objectives, wherein a solution that prioritized normalizing step length symmetry could slightly sacrifice normalizing foot segment kinematics, and vice versa. By taking advantage of the trade-off relationship between the two objectives, more flexible, individualized, and situation-dependent assistance can be obtained. These results demonstrate the method's usefulness in determining subject-specific exoskeleton control parameters that improve several measures of gait. Future applications include refining this protocol for actual dropfoot, offering personalized assistance to restore their functional mobility and reduce compensatory movements, and validating its long-term effects.

Background: Spinal cord injury (SCI) can lead to various neurophysiological changes, altering the neural motor control strategies. The lower limb muscles are of high importance for locomotion; however, there exists a significant knowledge gap in neurophysiological changes following SCI in these muscles. This study aims to explore the neuromuscular adaptations in the soleus and tibialis anterior muscles in persons with incomplete SCI. Methods: Ankle joint torque, high-density electromyography (HD-EMG) and motor unit parameters of tibialis anterior and soleus were analyzed during repeated sub-maximal voluntary isometric contractions (20% and 50% of the maximal torque) and compared to those from a control cohort. Results: We observed muscle-dependent alterations in motor control between the SCI and control groups. Namely, SCI group required significantly higher normalized EMG amplitudes than the control group to achieve the same contraction levels. At 50% contraction level, compared to control group, the SCI group motor units were recruited at lower thresholds in both muscles and fired at lower rates in the tibialis anterior muscle. We observed no significant differences in intramuscular motor unit coherence or muscle co-contraction between the two groups. Conclusions: The observed combination of between-group differences in motor unit behavior may indicate that even in muscles of high-functioning individuals with incomplete SCI, is a shift towards larger motor units in both tibialis anterior and soleus muscles. These results contribute to knowledge of neurophysiological modifications in major ankle muscles following a SCI and provide deeper insights into neurophysiological changes that can be used complementary to clinical SCI evaluation.

Breakthroughs in assistive exoskeletons have occurred in the recent decade; both active and passive devices that provide partial joint moments in the lower limbs have reduced metabolic costs during walking by assisting muscle action. Musculoskeletal simulation is highly useful in describing the interaction between assistive moments, muscle-tendon mechanics, and walking energetics. In this study, we computed optimal assistive moments in ankle plantarflexion and hip flexion that produce minimal muscle activations during walking, described the muscle energetics, and estimated the potential reduction in metabolic cost. We described with analyses of muscle-tendon mechanics and motor control how reductions in muscle activation do not always result in metabolic cost savings.