PUBLICATIONS
The evidence behind the work.
Selected publications connecting human movement with sensing, interfaces and rehabilitation. Follow the publication links for the full papers and their study-specific limitations.
Assessment and learning
2026
A narrative review of AI-driven stroke rehabilitation systems through the lens of human motor learning
Taghi, Z.; Serrien, D. J.; Fonseca, L.; Deshpande, N.
Examines how rehabilitation systems support human motor learning and how their models relate to learning mechanisms, including the comparatively underexplored learning involved in therapists’ assistance.
Read the paper · Research context
Abstract
Introduction Stroke represents a leading global cause of disability, often causing motor impairments that diminish quality of life. Neurorehabilitation that leverages human motor learning (HML) theories is crucial for post-stroke recovery. Therapists guide repetitive practice that supports re-learning, and they adjust assistance to individual needs and progress. Robot-assisted rehabilitation has advanced this approach, and recent work shows AI-driven systems can improve adaptability to patient behavior beyond earlier technologies. However, notably few systems aim to explicitly replicate therapist assistance from the perspective that physical assistance is a motor skill in itself.
Methodology This narrative review examines advances in AI-driven stroke rehabilitation, analyzing how systems facilitate HML within patients and how their models approximate HML mechanisms. By breaking down the four core HML processes to their essentials, using Marr’s tri-level hypothesis, we compare machine learning models used within rehabilitation systems to the HML processes.
Results Many of the reviewed systems appear to primarily facilitate use-dependent and sensory-prediction error-based learning, with limited facilitation of reinforcement learning or strategy-based learning. Explicit modeling of therapist HML within control frameworks appears relatively rare. Implicitly, many of the reviewed AI systems functionally represent one or two HML processes.
Conclusion Current research often considers HML primarily in patients, whereas therapists’ own HML likely underpins the robustness and adaptability of clinical assistance. Interpreting the reviewed rehabilitation systems through this lens highlights opportunities for therapist-inspired multi-process controllers, improved benchmarking with clinical scales, longitudinal retention studies, and AI-driven closed-loop neuromodulation to enhance personalization, adaptability, and outcomes, and to support clinical translation into routine practice.
2025
The evidence for biomechanical and physiological parameters as biomarkers to discriminate between individuals with and without non-specific neck pain using sensor devices: A systematic review with meta-analysis
Dagli, A. C.; Yazgan Dagli, B.; Al-Yahya, E.; Fonseca, L.; Caleb-Solly, P.
Synthesises evidence for sensor-derived movement and physiological measures in non-specific neck pain. Their potential for assessment still requires further validation.
Read the paper · Reading route
Abstract
Neck pain is among the most prevalent musculoskeletal conditions worldwide. The underlying cause mostly remains unidentified, classified as non-specific neck pain. Pain can alter movement patterns and physiological responses, suggesting that certain biomechanical and physiological changes may serve as objective biomarkers for non-specific neck pain. In recent years, growing interest in sensor technologies has enabled accurate and objective measurement of these changes. This is the first review to systematically summarise current evidence on the capability of biomechanical and physiological parameters, measured via sensors, to differentiate individuals with non-specific neck pain from asymptomatic controls, and evaluate their discriminative performance. Comprehensive searches of six databases (CINAHL, MEDLINE, EMBASE, AMED, IEEE Xplore, PEDro), grey literature, and reference lists (inception to August 20, 2025) yielded 53 observational studies for qualitative synthesis, with meta-analysis on 27. Meta-analysis indicates robust evidence linking non-specific neck pain with reduced neck range of motion, impaired joint position error, decreased step length and gait speed, reduced sway area, increased electromyographic activity of the sternocleidomastoid muscle, and reduced heart rate variability. Narrative findings reported altered neck movement speed, acceleration, and smoothness during functional tasks (e.g., reach and lifting). Classification studies showed high discriminative performance using machine learning and statistical techniques, with accuracies of 71.9–90%, sensitivities of 76.3–100%, and specificities of 77.6–90%, especially for gait and electromyography parameters. The findings highlight biomechanical and physiological alterations in non-specific neck pain that can serve as objective biomarkers. Clinically, these insights could offer support to enhance assessment and inform rehabilitation strategies.
Perspective This comprehensive review synthesises current evidence on physiological and biomechanical parameters as biomarkers in non-specific neck pain. While these parameters show promise for pain classification, their utility as biomarkers requires further evaluation and validation of their discriminative power for improved assessment and inform rehabilitation strategies.
Movement interfaces
2022
A Residual Movement Classification Based User Interface for Control of Assistive Devices by Persons With Complete Tetraplegia
Fonseca, L.; Guiraud, D.; Hiairrassary, A.; Fattal, C.; Azevedo-Coste, C.
A single inertial sensor and a user-specific classifier turn residual shoulder movement into commands for an assistive device.
Abstract
Objective: Complete tetraplegia can deprive a person of hand function. Assistive technologies may improve autonomy but needs for ergonomic interfaces for the user to pilot these devices still persist. Despite the paralysis of their arms, people with tetraplegia may retain residual shoulder movements. In this work we explored these movements as a mean to control assistive devices.
Methods: We captured shoulder movement with a single inertial sensor and, by training a support vector machine based classifier, we decode such information into user intent.
Results: The setup and training process take only a few minutes and so the classifiers can be user specific. We tested the algorithm with 10 able body and 2 spinal cord injury participants. The average classification accuracy was 80% and 84%, respectively.
Conclusion: The proposed algorithm is easy to set up, its operation is fully automated, and achieved results are on par with state-of-the-art systems.
Significance: Assistive devices for persons without hand function present limitations in their user interfaces. Our work presents a novel method to overcome some of these limitations by classifying user movement and decoding it into user intent, all with simple setup and training and no need for manual tuning. We demonstrate its feasibility with experiments with end users, including persons with complete tetraplegia without hand function.
2022
Frequency-Domain sEMG Classification Using a Single Sensor
Stefanou, T.; Guiraud, D.; Fattal, C.; Azevedo-Coste, C.; Fonseca, L.
Explores how information from a single surface electromyography sensor can support classification of movements and muscle contractions.
Abstract
Working towards the development of robust motion recognition systems for assistive technology control, the widespread approach has been to use a plethora of, often times, multi-modal sensors. In this paper, we develop single-sensor motion recognition systems. Utilising the peripheral nature of surface electromyography (sEMG) data acquisition, we optimise the information extracted from sEMG sensors. This allows the reduction in sEMG sensors or provision of contingencies in a system with redundancies. In particular, we process the sEMG readings captured at the trapezius descendens and platysma muscles. We demonstrate that sEMG readings captured at one muscle contain distinct information on movements or contractions of other agonists. We used the trapezius and platysma muscle sEMG data captured in able-bodied participants and participants with tetraplegia to classify shoulder movements and platysma contractions using white-box supervised learning algorithms. Using the trapezius sensor, shoulder raise is classified with an accuracy of 99%. Implementing subject-specific multi-class classification, shoulder raise, shoulder forward and shoulder backward are classified with a 94% accuracy amongst object raise and shoulder raise-and-hold data in able bodied adults. A three-way classification of the platysma sensor data captured with participants with tetraplegia achieves a 95% accuracy on platysma contraction and shoulder raise detection.
2019
Assisted Grasping in Individuals with Tetraplegia: Improving Control through Residual Muscle Contraction and Movement
Fonseca, L.; Tigra, W.; Navarro, B.; Guiraud, D.; Fattal, C.; Bó, A.; Fachin-Martins, E.; Leynaert, V.; Gélis, A.; Azevedo-Coste, C.
Investigates residual muscle contractions and upper-arm movement as input methods for robotic and stimulation-assisted grasp.
Abstract
Individuals who sustained a spinal cord injury often lose important motor skills, and cannot perform basic daily living activities. Several assistive technologies, including robotic assistance and functional electrical stimulation, have been developed to restore lost functions. However, designing reliable interfaces to control assistive devices for individuals with C4–C8 complete tetraplegia remains challenging. Although with limited grasping ability, they can often control upper arm movements via residual muscle contraction. In this article, we explore the feasibility of drawing upon these residual functions to pilot two devices, a robotic hand and an electrical stimulator. We studied two modalities, supra-lesional electromyography (EMG), and upper arm inertial sensors (IMU). We interpreted the muscle activity or arm movements of subjects with tetraplegia attempting to control the opening/closing of a robotic hand, and the extension/flexion of their own contralateral hand muscles activated by electrical stimulation. Two groups were recruited: eight subjects issued EMG-based commands; nine other subjects issued IMU-based commands. For each participant, we selected at least two muscles or gestures detectable by our algorithms. Despite little training, all participants could control the robot’s gestures or electrical stimulation of their own arm via muscle contraction or limb motion.
2018
Investigating Upper Limb Movement Classification on Users with Tetraplegia as a Possible Neuroprosthesis Interface
Fonseca, L.; Bo, A.; Guiraud, D.; Navarro, B.; Gelis, A.; Azevedo-Coste, C.
Compares calibration approaches for a single-IMU interface used to control postures of a robotic hand.
Abstract
Spinal cord injury (SCI), stroke and other nervous system conditions can result in partial or total paralysis of individual’s limbs. Numerous technologies have been proposed to assist neurorehabilitation or movement restoration, e.g. robotics or neuroprosthesis. However, individuals with tetraplegia often find difficult to pilot these devices. We developed a system based on a single inertial measurement unit located on the upper limb that is able to classify performed movements using principal component analysis. We analyzed three calibration algorithms: unsupervised learning, supervised learning and adaptive learning. Eight participants with tetraplegia (C4C7) piloted three different postures in a robotic hand. We achieved 89% accuracy using the supervised learning algorithm. Through offline simulation, we found accuracies of 76% on the unsupervised learning, and 88% on the adaptive one.
Stimulation and activity
2022
Activating effective functional hand movements in individuals with complete tetraplegia through neural stimulation
Azevedo Coste, C.; William, L.; Fonseca, L.; Hiairrassary, A.; Andreu, D.; Geffrier, A.; Teissier, J.; Fattal, C.; Guiraud, D.
A proof-of-concept neuroprosthesis study combining selective nerve stimulation with an interface controlled through voluntary actions.
Abstract
Individuals with complete cervical spinal cord injury suffer from a permanent paralysis of upper limbs which prevents them from achieving most of the activities of daily living. We developed a neuroprosthetic solution to restore hand motor function. Electrical stimulation of the radial and median nerves by means of two epineural electrodes enabled functional movements of paralyzed hands. We demonstrated in two participants with complete tetraplegia that selective stimulation of nerve fascicles by means of optimized spreading of the current over the active contacts of the multicontact epineural electrodes induced functional and powerful grasping movements which remained stable over the 28 days of implantation. We also showed that participants were able to trigger the activation of movements of their paralyzed limb using an intuitive interface controlled by voluntary actions and that they were able to perform useful functional movements such as holding a can and drinking through a straw.
2017
Cycling with Spinal Cord Injury: A Novel System for Cycling Using Electrical Stimulation for Individuals with Paraplegia, and Preparation for Cybathlon 2016
Bo, A. P. L.; da Fonseca, L. O.; Guimaraes, J. A.; Fachin-Martins, E.; Paredes, M. E. G.; Brindeiro, G. A.; de Sousa, A. C. C.; Dorado, M. C. N.; Ramos, F. M.
Describes an FES-cycling system and preparation for the Cybathlon competition, connecting control and instrumentation with a functional activity.