2025Activity​​ reportProject-TeamCAMIN

7.1.1 i-GRIP

• Keywords:​​
  Handicap, Computer vision, Persons​​​‌ attendant, Exoskeleton, Detection
• Scientific​ Description:
  Detection of object​‌ grasping intention and automatic​​ selection of grasp type​​​‌ for shared control of​ (neuro)prostheses.
• Functional Description:
  From​‌ a video stream of​​ hands and objects, i-GRIP​​​‌ detects the intention to​ grasp one of them​‌ and identifies the grip​​ the hand should adopt​​​‌ to appropriately seize it​ based on the approaching​‌ movement. i-GRIP will enable​​ intuitive and low cognitive​​​‌ load control of hand​ movement assistive devices (exoskeletons,​‌ functional electrical stimulation, prosthetics).​​
• Publication:
  hal-04141469
• Contact:
  Etienne​​​‌ Moullet

7.1.2 SMS

• Name:​
  Software for Manual Segmentations​‌
• Keyword:
  Image segmentation
• Functional​​ Description:
  Software that combines​​​‌ various ways of segmenting​ images, whether by drawing​‌ or placing points.
• Contact:​​
  Baptiste Faraud

7.1.3 TARGETTRACK​​​‌

• Name:
  TARGETTRACK
• Keywords:
  Motor​ reeducation, Health
• Functional Description:​‌
  Software for playfully visualizing​​ actions performed by users,​​​‌ with the aim of​ offering rehabilitation.
• Contact:
  Baptiste​‌ Faraud

7.1.4 IMUSEF

• Keywords:​​
  Inertial module unit, Adaptive​​​‌ algorithms, Human Movement Analysis​
• Scientific Description:
  Modular embedded​‌ framework for real-time control​​ of Functional Electrical Stimulation​​​‌ in closed loop with​ sensor feedback.
• Functional Description:​‌
  IMUSEF enables the collection​​ of data from various​​​‌ sensors, including inertial measurement​ units, to detect interactions​‌ with the user or​​ evaluate events or information​​​‌ about the user, such​ as the orientation of​‌ a segment or a​​ joint angle. With this​​​‌ information, a decision-making algorithm​ selected from a modular​‌ panel can detect events​​ and control actuators such​​​‌ as an electrical stimulator​ to assist movement.
• Release​‌ Contributions:
  1.0 : First​​ fully functional version application​​​‌ focused with stimulator control,​ sensor data acquisition and​‌ analysis and decisional algorithm​​ (https://www.mdpi.com/545320). 2.0 : Full​​​‌ rework as a modular​ framework, allowing selection of​‌ decisional algorithm, stimulator system,​​ sensors etc. and optional​​ communication and control in​​​‌ real time from a‌ remote fully featured Graphical‌​‌ User Interface through autonomous​​ hotspot wifi.
• News of​​​‌ the Year:
  New software‌ features were designed and‌​‌ integrated in preparation for​​ new experimentations, along with​​​‌ adaptations to conform to‌ new platform changes and‌​‌ upgrades on related hardware​​ and firmware. The improvements​​​‌ aimed at preparing this‌ platform and its capabilities‌​‌ to allow for standardized​​ and generalized use.
• URL:​​​‌
  https://­gitlab.­inria.­fr/­camin-soft/­IMUSEF/­IMUSEF_APP
• Publication:
  lirmm-02305548, version‌ 1
• Contact:
  Ronan Le‌​‌ Guillou
• Participant:
  4 anonymous​​ participants

7.2.1 AI-Hand platforms​​

AI-HAND first clinical trial​​​‌ (CT1) experimental platform to‌ assess the capacity to‌​‌ modulate stimulation intensity to​​ adjust grasping force

Participants:​​​‌ Baptiste Faraud, Christine‌ Azevedo, François Bailly‌​‌.

The content of​​ this experimental platform consists​​​‌ in two main aspects‌ : the control modalities‌​‌ (Fig. 3, top)​​ and the Target-Track software​​​‌ in static mode (Fig.‌ 3, bottom). Users‌​‌ included in CT1, with​​ tetraplegia without functional electrical​​​‌ stimulation) can use several‌ control modalities to adjust‌​‌ a virtual grasping force​​ which is displayed through​​​‌ the Target-Track software. The‌ force adjustment is performed‌​‌ thanks to two distinct​​ control inputs, "+" and​​​‌ "-", respectively. The Target-Track‌ software provides a visual‌​‌ feedback to the user​​ (through the display of​​​‌ the current grasping force‌ achieved through the control‌​‌ inputs) and a reference​​ value (static or dynamic)​​​‌ to be tracked by‌ the user. The overall‌​‌ objective is to evaluate​​ the performances of the​​​‌ several control modalities to‌ achieve force adjustments.

This‌​‌ image depicts, on the​​ top, examples of control​​​‌ modalities such as ear-based,‌ microphone or joystick. The‌​‌ user operates these devices​​ to send control signals​​​‌ to a software to‌ control a serious game.‌​‌

The​​​‌ Target-Track software has two‌ modes: a static gauge‌​‌ (Fig. 3 bottom) and​​ a dynamic serious game​​​‌ (Fig. 4).

The‌ image shows a video‌​‌ game with a UFO​​ at the center of​​​‌ the screen. The UFO‌ is surrounded by vertical‌​‌ blue columns. The UFO​​ appears to be navigating​​​‌ through the vertical columns.‌ One of them is‌​‌ highlighted in red.

Force estimation of‌​‌ fingers during hand grasping​​ movements

Participants: Jean De​​​‌ Gheldere, Christine Azevedo‌, François Bailly.‌​‌

As a part of​​ the AI-HAND project, in​​​‌ the scope of the‌ indirect measurement of hand‌​‌ grasping forces, we continued​​ the development of the​​​‌ robotic platform able to‌ directly assess grasp forces‌​‌ in various configurations. This​​ platform will help us​​​‌ assess hand dynamics, particularly‌ to calibrate the indirect‌​‌ measurement methods, by simulating​​ rigid or soft contacts​​​‌ while measuring the force‌ developed by such movements‌​‌ and grasping patterns.

The​​ system consists of two​​​‌ parts: a ’robot’ part‌ and an ’interface’ part.‌​‌ The robot is an​​ assembly of two torque-controlled​​​‌ axes, incorporating a capstan‌ drive mechanism, enabling backlash-free‌​‌ transmission between the motors​​​‌ and the ’interface’ subsystem.​ The interface is a​‌ 4-bar mechanism, which can​​ reach a wide range​​​‌ of configurations, and transmit​ the forces from the​‌ hand to the robot​​ rigidly.

The device was​​​‌ fully prototyped (Fig. 5​), and is now​‌ in the last development​​ stage. It integrates custom​​​‌ made torque sensors that​ are also in their​‌ final phase of development​​ (Fig. 6). Most​​​‌ of the technical choices​ are now validated, and​‌ we are working on​​ the device's reliability and​​​‌ deployability for future experiments.​

The image shows the​‌ mechanical assembly of the​​ device placed on a​​​‌ perforated workbench. The assembly​ includes a black and​‌ orange structure with various​​ components. There are wires​​​‌ connecting these parts.

CAD assembly featuring a​‌ cylindrical shaft and various​​ components attached along its​​​‌ length.

7.2.2​​​‌ IMUSEF modular platform for​ research experimentations with FES)​‌

Participants: Le Guillou Ronan​​, Azevedo Christine,​​​‌ Gasq David [CHU Toulouse]​.

This platform, developped​‌ in the team initially​​ for FES-assisted cycling with​​​‌ SCI participants, evolved into​ a modular platform to​‌ support diverse research experimentations​​ on various pathologies (e.g.​​​‌ Stroke, SCI, Cerebral Palsy​ etc.). A desired Functional​‌ Electrical Stimulator can be​​ controlled by this platform​​​‌ through a chosen decisional​ algorithm using sensor feedback.​‌ New sensors, algorithms and​​ electrical stimulators or other​​​‌ actionners can be added​ in a modular way.​‌ This embedded platform, upgraded​​ to become fully autonomous​​​‌ and wearable, allows for​ ambulatory experimentations.

The lack​‌ of devices available for​​ practical use in daily-life​​​‌ conditions in Stroke survivors​ led to work with​‌ the Hospital of Toulouse​​ in previous years and​​​‌ the acquisition of significant​ know-how on the design​‌ requirements, user feedback, and​​ characteristics needed for high​​​‌ acceptability in daily usage.​ Through collaboration with Toulouse​‌ Tech Transfer, the French​​ Society for Acceleration of​​​‌ Technological Transfer (SATT) of​ Toulouse, this acquired knowhow​‌ was deposited this year​​ as a e-Soleau Intellectual​​​‌ Property envelope. In preparation​ for future experimentations, various​‌ new features and upgrades​​ were added, hardware, firmware​​​‌ and software wise. New​ capabilities of this platform​‌ will allow future generalization​​ and standardization of its​​​‌ usage as a shared​ platform for other experimentations​‌ in the team.

The​​ image shows three parts​​​‌ of an assistive device,​ designed to assist grasping​‌ for individuals with limited​​ prehension capabilities. Part A​​​‌ shows a single board​ computer stored in a​‌ waist bag. Part B​​ is a wearable sleeve​​​‌ garment with an electrode​ array that sticks to​‌ the forearm skin, enabling​​ evoked muscle contractions through​​​‌ electrical currents. Part C​ is a small handheld​‌ device with buttons and​​ a screen, used for​​​‌ controlling the system and​ its modes of activations.​‌ The setup is presented​​ worn by a person​​​‌ which is using it​ to grasp a spoon​‌ in a clinical testing​​ situation.

8.1.1 Influence‌​‌ of myelo-architecture on direct​​ cortical response evoked by​​​‌ electrical stimulation

Participants: Clotilde‌ Turpin, Olivier Rossel‌​‌, Félix Schlosser-Perrin,​​ Riki Matsumoto [Neurology Dpt​​​‌ Kyoto Hospital, Japan],‌ Emmanuel Mandonnet [Neurosurgery Dpt‌​‌ APHP, Paris], Sam​​ Ng [Neurosurgery Dpt Montpellier​​​‌ Hospital], Hugues Duffau‌ [Neurosurgery Dpt Montpellier Hospital]‌​‌, François Bonnetblanc.​​

The measurement of evoked​​​‌ potentials (EPs) by direct‌ electrical stimulation (DES) during‌​‌ brain surgery allows identifying​​ structural or anatomical connectivity​​​‌ in real time, while‌ aiming to preserve it‌​‌ (39, 19​​, 20). Classically,​​​‌ the evoked response is‌ composed of an early‌​‌ positive component, denoted P0,​​ occurring after the DES​​​‌ artifact when measurable, followed‌ by a more robust‌​‌ negative deflection, denoted N1.​​ As DES initially activates​​​‌ larger elements, and given‌ its early onset, P0‌​‌ is assumed to reflect​​ a summation of highly​​​‌ synchronized action potentials (APs).‌

Cyto-myelo architecture varies across‌​‌ brain regions, particularly between​​ primary areas (motor M1,​​​‌ sensory S1), the premotor‌ cortex and the more‌​‌ associative areas such as​​ the Broca, Wernicke, and​​​‌ other association areas. Precentral‌ motor cortex and S1‌​‌ are distinguished by the​​ presence of larger diameter,​​​‌ heavily myelinated fibers. Due‌ to the specific characteristics‌​‌ of electrical stimulation, these​​ architectural variations should be​​​‌ reflected in cortical responses‌ evoked by DES.

Among‌​‌ the first elements to​​ be activated, pyramidal axons​​​‌ (particularly abundant in the‌ precentral motor and primary‌​‌ somatosensory (S1) areas) generate​​ high-amplitude signals, are more​​​‌ readily excitable, and propagate‌ action potentials more rapidly.‌​‌ This is expected to​​ result in a shorter​​​‌ latency but higher-amplitude P0‌ component, producing a steeper‌​‌ downward slope in the​​ signal (relaxation slope) in​​​‌ motor and somatosensory cortices‌ compared to more associative‌​‌ regions such as Broca’s​​ area, Wernicke’s area, and​​​‌ other associative cortices. We‌ sought to test this‌​‌ hypothesis.

DES was administered​​ directly to different regions​​​‌ of the cortex while‌ recording DCRs (direct cortical‌​‌ responses) in 10 patients.​​ The shapes of the​​​‌ first components P0 and‌ N1 of the signals‌​‌ were analyzed.

The downward​​ slope of the first​​​‌ component (P0) of the‌ signal is statistically greater‌​‌ for responses recorded in​​ precentral motor cortex and​​​‌ S1 than that of‌ EPs recorded in more‌​‌ associative areas (Fig. 8​​).

Anatomical features of​​​‌ (pre) motor and somatosensory‌ explain the response and‌​‌ the increased slope of​​​‌ the P0 component in​ these regions, compared to​‌ associative areas. The first​​ component of DES-evoked responses​​​‌ reflects myelo architecture in​ particular. This could form​‌ the basis of an​​ electrodiagnostic method using the​​​‌ evoked response or to​ better distinguish between specific​‌ areas intra-operatively.

The figure​​ is composed of 3​​​‌ parts. The part A​ is an image showing​‌ a map of the​​ human brain with highlighted​​​‌ cortical regions and markers​ positionned at recording sites,​‌ each tagged with patient​​ numbers (P1 to P10).​​​‌ Part B is a​ graph of amplitude versus​‌ time showing an average​​ waveform of evoked potential​​​‌ and illustrating the different​ metrics. Part C presents​‌ multiple amplitude versus time​​ graphs for brain responses​​​‌ recorded at the various​ markers for the subjects.​‌

8.1.2 noCNN : No-brain-shift​​​‌ and Comprehensive Neurosurgical Navigation​ using computer vision

Participants:​‌ Paul André, Emilie​​ Ouraou, François Bailly​​​‌, François Bonnetblanc,​ Emmanuel Mandonnet [Neurosurgery Dpt​‌ APHP, Paris], Sam​​ Ng [Neurosurgery Dpt Montpellier​​​‌ Hospital], Hugues Duffau​ [Neurosurgery Dpt Montpellier Hospital]​‌.

Brain neurosurgeons have​​ a particular need for​​​‌ reliable images to plan​ and guide their procedures.​‌ This is especially true​​ because brain shift occurs​​​‌ immediately upon opening the​ dura-mater due to the​‌ decrease in intracranial pressure​​ and the outflow of​​​‌ cerebrospinal fluid. Since the​ brain is composed of​‌ soft tissue, this shift​​ is exacerbated when the​​​‌ surgeon must perform an​ excision, sometimes involving volumes​‌ exceeding 100 cm³ in​​ the case of tumor​​​‌ surgery. In the field​ of clinical imaging, intraoperative​‌ photos or videos of​​ brain surgeries are common​​​‌ but rarely used in​ conjunction with computer vision​‌ approaches.

During neurosurgery, Magnetic​​ Resonance Imaging (MRI) are​​​‌ considered before the operation​ to allow planification and​‌ neuro-navigation, but do not​​ take into account the​​​‌ brain shift that can​ be up to 2​‌ cm.

In order to​​ find the correspondence between​​​‌ intra-operative images and MRIs,​ we aim to perform​‌ a non-rigid registration. Intra-operative​​ images are in 2D,​​​‌ while MRIs are in​ 3D, so in order​‌ to have data of​​ the same dimensions, we​​​‌ project the MRI onto​ a plane that approximately​‌ represents the camera's position​​ in the real world.​​​‌

From images showing the​ brain surface in these​‌ two different modalities, we​​ set up a pipeline,​​​‌ shown in Figure 9​ to pre-process them and​‌ extract features in order​​ to perform registration. The​​ three elements that can​​​‌ be found in both‌ intraoperative images and preoperative‌​‌ MRI (without contrast agent​​ and with gadolinium injection)​​​‌ are vessels, sulci, and‌ gyri.

In order to‌​‌ segment these elements in​​ the intraoperative image, we​​​‌ have planed to train‌ a U-Net, a convolutional‌​‌ network that works well​​ with few data and​​​‌ is suitable for medical‌ applications 37. The‌​‌ training will incorporate constraints​​ giving advantage to the​​​‌ anatomical accuracy into the‌ cost function, such as‌​‌ ClDice 38. In​​ order to train this​​​‌ model, we are currently‌ building a dataset containing‌​‌ image/segmentation pairs with around​​ ten classes, which is​​​‌ more than what is‌ commonly done. We have‌​‌ currently segmented 65 images​​ manually, requiring around 4​​​‌ hours per image. Each‌ image will be double-checked‌​‌ by healthcare experts.

To​​ segment the MRIs, we​​​‌ first process them with‌ HD-BET (brain extraction tool)‌​‌ and then Freesurfer in​​ order to extract the​​​‌ brain and to better‌ remove the dura mater,‌​‌ which may still be​​ present after preprocessing with​​​‌ skullstrip tools. This also‌ allows us to normalize‌​‌ the voxel values. Then​​ we display the MRI​​​‌ in 3DSlicer's 3D view,‌ we manually position a‌​‌ camera in a view​​ similar to the camera​​​‌ in the real world,‌ and we generate an‌​‌ image representing the brain​​ surface from the T1​​​‌ MRI without contrast agent,‌ and another with the‌​‌ same viewpoint from the​​ T1 MRI with gadolinium.​​​‌ This step requires human‌ intervention, but could be‌​‌ automated using a FLAIR​​ image to determine the​​​‌ position of the tumor‌ and therefore deduce the‌​‌ position of the craniotomy​​ and the surgeon. In​​​‌ the MRI images, the‌ vessels are extracted by‌​‌ thresholding, while the gyri​​ and sulci are segmented​​​‌ manually.

Using these segmented‌ images, we will now‌​‌ perform a non-rigid registration​​ in order to find​​​‌ the correspondence between the‌ T1 MRI and the‌​‌ real brain during the​​ operation.

In addition, we​​​‌ plan to publish a‌ dataset containing the images‌​‌ and manual segmentations, as​​ well as a section​​​‌ of the MRI corresponding‌ to the craniotomy, over‌​‌ the upcoming year.

An​​ internal communication on this​​​‌ "Action exploratoire" was published‌ this year.

The image‌​‌ illustrates the process for​​ brain surface image creation​​​‌ and surgical planning. It‌ involves using deep learning‌​‌ (U-Net) for automated segmentation​​ of intraoperative images into​​​‌ vessels, gyri, and sulci.‌ MRI scans (T1 and‌​‌ T1CE) are used to​​ extract regions of interest​​​‌ and perform vessel, gyri,‌ and sulci segmentation. These‌​‌ segmentations are combined to​​ compute non-rigid registration and​​​‌ estimate brain surface deformation.‌ Style transfer is applied‌​‌ to create a detailed​​ brain surface image for​​​‌ surgical planning. Data augmentation‌ is used to improve‌​‌ segmentation accuracy.

8.1.3 Style​​ Transfer application to better​​​‌ guiding of brain surgery‌

Participants: Emilie Ouraou,‌​‌ Paul André, François​​ Bailly, François Bonnetblanc​​​‌, Ronan Le Guillou‌, Emmanuel Mandonnet [Neurosurgery‌​‌ Dpt APHP, Paris],​​​‌ Sam Ng [Neurosurgery Dpt​ Montpellier Hospital], Hugues​‌ Duffau [Neurosurgery Dpt Montpellier​​ Hospital].

Before neurosurgery,​​​‌ a pre-operatory MRI is​ acquired to allow the​‌ surgeon to plan the​​ surgery. However, because of​​​‌ brain deformations called brain-shift​ occurring at the start​‌ and during surgery, these​​ MRIs do not precisely​​​‌ represent what the surgeon​ will encounter during the​‌ operation. The main objective​​ of this project is​​​‌ to help surgery planning​ by providing a more​‌ accurate view of the​​ brain's configuration. Our main​​​‌ concern is the blood​ vessels, since we want​‌ the surgeon to be​​ able to clearly identify​​​‌ the regions where extra​ caution is required. Indeed,​‌ the surgeon has to​​ be very careful not​​​‌ to cut any big​ blood vessel trying to​‌ access the tumor, but​​ currently they can only​​​‌ be seen pre-operatively by​ extracting 2D slices from​‌ the 3D MRI.

For​​ a start, we want​​​‌ to link the pre-operative​ MRI with the intra-operative​‌ view (obtained via intra-operatory​​ images), initially without considering​​​‌ brain-shift. The goal is​ to transform the MRI​‌ into a realistic visual​​ representation corresponding to the​​​‌ cerebral surface where surgery​ is going to take​‌ place, using style transfer.​​ To do so, we​​​‌ seek to translate segmented​ images back into realistic​‌ intra-operative images using generative​​ models, namely Pix2pix and​​​‌ CycleGAN. Both are derivatives​ of the Generative Adversarial​‌ Network model types, but​​ while Pix2pix uses paired​​​‌ data (combinations of images​ and its associated segmentation,​‌ which limits the number​​ of available data due​​​‌ to the scarcity of​ segmentations available), CycleGAN works​‌ with unpaired data (images​​ and segmentations do not​​​‌ need to be from​ the same patient). On​‌ the other hand, Pix2pix​​ is often more respectful​​​‌ of the structures represented.​

Our first step focuses​‌ on intra-operative images. We​​ manually segment these images​​​‌ and train a generative​ model to learn the​‌ transformation from segmentation maps​​ back into real images.​​​‌ The main novelty of​ this work lies in​‌ the number of classes​​ chosen: while the literature​​​‌ generally considers three classes​ (background, parenchyma and vessels),​‌ we chose to also​​ include gyri and sulci,​​​‌ believing it should improve​ the precision of our​‌ model.

This step will​​ also be useful to​​​‌ train an automatic segmentation​ model. Due to the​‌ scarcity of segmented data​​ and the time burden​​​‌ of manually labeling each​ image (around four hours​‌ per image), segmented data​​ are rare and thus​​​‌ training datasets are limited.​ However, our style transfer​‌ model could be applied​​ to generate realistic-looking images​​​‌ from synthetic segmentations, in​ order to artificially increase​‌ to amount of data​​ at disposal to train​​​‌ an automatic segmentation model.​

The next step is​‌ to segment the pre-operative​​ MRIs and use those​​​‌ segmentations in the style​ transfer network, attempting to​‌ convert MRI segmentations into​​ photorealistic intra-operative images. This​​​‌ new representation could help​ non-expert surgeons better visualize​‌ MRI data and improve​​ surgery planning. In future​​​‌ steps, we could also​ generate augmented-reality images which​‌ would, for example, display​​ deep vessels not currently​​ visible to the surgeon​​​‌ but which require special‌ attention.

Figure 10 presents‌​‌ examples of the style​​ transfer currently achieved.

The​​​‌ image depicts a flowchart‌ of a style transfer‌​‌ model for medical imaging,​​ specifically for brain scans.​​​‌ It involves different stages‌ and components (input images‌​‌ and manual segmentation, generative​​ Models)

8.2.1 Optimal estimation of‌​‌ forearm muscle activations evoked​​ by implanted electrical stimulation​​​‌ in complete quadriplegia

Participants:‌ Maggioni Valentin, Schegg‌​‌ Pierre, Guiho Thomas​​, Faraud Baptiste,​​​‌ Azevedo Christine, Bailly‌ François.

Muscle activity‌​‌ assessment is crucial to​​ understand both natural and​​​‌ evoked human movements. The‌ most common method to‌​‌ measure it is through​​ surface electromyography (EMG). However,​​​‌ this method presents significant‌ caveats, such as cross-talk‌​‌ from adjacent muscles, or​​ its inability to detect​​​‌ deep-muscles’ contractions. A possible‌ solution to address these‌​‌ limitations is multimodal estimation,​​ through the use of​​​‌ a musculoskeletal (MSK) model‌ leveraging kinematic and EMG‌​‌ data. In particular, it​​ is possible to formulate​​​‌ an optimal estimation problem‌ to blend these quantities‌​‌ and estimate the activation​​ of the model’s muscles.​​​‌

This work aims to‌ apply this approach to‌​‌ a hand model developed​​ in the context of​​​‌ functional electrical stimulation (FES)‌ of arms’ nerves to‌​‌ restore hand movements in​​ individuals with complete quadriplegia​​​‌ via implanted epineural electrodes,‌ to accurately estimate the‌​‌ muscle activation of FES-induced​​ hands movements.

This study​​​‌ combines kinematic and EMG‌ data from the AGILIS‌​‌ and AGILISTIM experimental protocols​​ in a single optimal​​​‌ estimation problem that was‌ adapted specifically for the‌​‌ context of FES-induced hand​​ movements, by considering different​​​‌ objective functions and simulation‌ parameters depending on whether‌​‌ the radial or median​​ nerve is stimulated. The​​​‌ objective functions that permit‌ to exploit both kinematic‌​‌ and EMG data simultaneously​​ mathematically takes the form​​​‌ of a cost function‌ to be minimized and‌​‌ expressed as:

Where​​ $n_{mu\mathrm{s‌}cle\mathrm{s‌}}$ and $k_{j\mathrm{o‌‌}int\mathrm{s}}$ are the number of​​​‌ muscles and joints of‌ the musculoskeletal model respectively,‌​‌ $a_n^{*}$ and​​ $a_n$ are the​​​‌ muscle activation derived from‌ the EMG measurements and‌​‌ estimated by the optimal​​​‌ estimation problem respectively, ${\mathrm{q}}_k^{*}$ and ${\mathrm{q‌}}_k$ are the joint​​ kinematics measured experimentally and​​​‌ estimated by the optimal​ estimation problem respectively, ${\mathrm{\omega ‌}}_a$ and ${\omega }_{\mathrm{q}}$ are the weight associated​​​‌ with the activation and​ kinematic terms respectively.

The​‌ kinematic data was obtained​​ through markerless motion capture​​​‌ using webcams and the​ Mediapipe API. The pipeline​‌ used to convert the​​ webcam video feed into​​​‌ usable kinematic data was​ developed between 2024 and​‌ 2025 and is described​​ in a paper that​​​‌ was published in Sensors​ (15). A​‌ graph of the experimental​​ setup used to obtain​​​‌ kinematic and EMG data​ in the AGILIS and​‌ AGILISTIM experimental setup, as​​ well as the max​​​‌ position and muscle recruitment​ measured for specific articulations​‌ and muscles for two​​ stimulation configurations are shown​​​‌ in Fig. 11.​ Additionally, EMOK, a software​‌ made to ease the​​ usage of the motion​​​‌ capture method was developed​ and uploaded on the​‌ BIL software database.​​

To ensure a correct​​​‌ estimation, a musculoskeletal hand​ model was developed based​‌ on an open-source literature​​ model and personalized for​​​‌ quadriplegic individuals. This includes​ a simplification of the​‌ model by removing all​​ muscles innervated by the​​​‌ ulnar nerve, which is​ not stimulated by an​‌ implant and an adaptation​​ of the passive forces​​​‌ at each finger articulation,​ which are personalized to​‌ each individual based on​​ kinematic measurements.

The optimal​​​‌ estimation method, combined with​ the personalized hand musculoskeletal​‌ model allowed us to​​ estimate the muscle activation​​​‌ of deep muscle for​ some of the AGILISTIM​‌ measurements, and some of​​ the first results of​​​‌ this work were presented​ in a scientific conference​‌ (“Congrès de la société​​ de Biomécanique 2025”) and​​​‌ published in the conference​ paper 16. Some​‌ of the estimated muscle​​ activation from an AGILISTIM​​​‌ measurement are shown Figure​ 12. More measurements​‌ from both the AGILIS​​ and AGILISTIM experimental protocols,​​​‌ as well as measurements​ from the Clinical Trial​‌ 2 of the AI-HAND​​ European project will additionally​​​‌ be studied with this​ method as part of​‌ future works.

The image​​ illustrates the setup for​​​‌ analyzing hand movement and​ muscle activity using cameras​‌ and surface electromyography (eMG).​​ Diagrams depict the kinematics​​​‌ of wrist, thumb, and​ finger movements, along with​‌ muscle recruitment data for​​ various forearm muscles. The​​​‌ bar graphs detail maximum​ joint positions and muscle​‌ activity levels during different​​ hand tasks. A labeled​​​‌ diagram of the hand's​ muscles is also included,​‌ identifying key muscles involved​​ in these actions.

Bar graph​​​‌ of the mean forearm​ muscle activation obtained with​‌ the optimal control problem​​ solution and the measured​​​‌ values.

8.2.2‌​‌ Evaluation of muscle recruitment:​​ Decomposition of Evoked Electromyographic​​​‌ Signals

Participants: Olivier Rossel‌, Maria Fernanda Paes‌​‌ Leme, Thomas Guiho​​, François Bailly,​​​‌ Christine Azevedo.

The‌ image illustrates a scientific‌​‌ study of neural stimulation​​ and muscle electrical activity​​​‌ recording. Panel A illustrates‌ the neural stimulation setup,‌​‌ with electrodes placed on​​ the nerve and EMG​​​‌ recordings obtained from muscles‌ of the hand and‌​‌ forearm. Panel B displays​​ electromyography (EMG) signals from​​​‌ three distinct muscles. Panel‌ C shows muscle recruitment‌​‌ levels (in percentage) for​​ two stimulation configurations. Panel​​​‌ D displays the corresponding‌ recorded EMG signals for‌​‌ these two stimulation configurations​​ where the resulting EMG​​​‌ is a waited sum‌ of EMG components, whose‌​‌ amplitudes are proportional to​​ the muscle recruitment. Panel​​​‌ E summarizes the underlying‌ hypothesis and analysis framework‌​‌ (in text and equation)​​ : the recorded signals​​​‌ are assumed to be‌ linear combinations of EMG‌​‌ components with amplitudes proportional​​ to the muscle recruitment​​​‌ levels. The objective is‌ to identify the contributing‌​‌ EMG components and their​​ activation levels. The method​​​‌ consist in semi–non-negative matrix‌ factorization approach applied to‌​‌ records.

This project addresses the‌​‌ challenge of estimating muscle​​ recruitment levels through the​​​‌ analysis of evoked electromyographic‌ (EMG) signals. These signals,‌​‌ recorded using surface electrodes,​​ are inherently complex due​​​‌ to the superposition of‌ activities originating from multiple‌​‌ muscles (Fig. 13).​​ The recorded EMG signal​​​‌ contains information related both‌ to the recruitment level‌​‌ and to the specific​​ combination of EMG components​​​‌ associated with each muscle.‌ In this work, we‌​‌ assume that the observed​​ signal can be modeled​​​‌ as a linear combination‌ of individual muscle EMG‌​‌ components, weighted by their​​ respective recruitment levels (Fig.​​​‌ 13.E).

This direct‌ modeling approach enables the‌​‌ generation of synthetic EMG​​ signal databases, which are​​​‌ used as benchmark data‌ to quantitatively evaluate the‌​‌ performance and efficiency of​​ the proposed signal decomposition​​​‌ methods.

The image displays‌ three panels related to‌​‌ electromyography (EMG) signal analysis.​​ Panel A shows the​​​‌ synthetic EMG benchmark signal‌ and the corresponding estimated‌​‌ signal (black and colored​​​‌ traces, respectively). Panel B​ presents the results obtained​‌ using the standard non-negative​​ matrix factorization (SNNMF), including​​​‌ the extracted signal waveforms​ and their associated activation​‌ patterns (histograms) for the​​ different stimulation configurations. Panel​​​‌ C shows the results​ of the modified SNNMF.​‌ As in Panel B,​​ the extracted waveforms and​​​‌ their corresponding activation patterns​ are displayed, but with​‌ improved signal separation and​​ activation estimates that more​​​‌ closely match the benchmark​ data.

The Standard Non-Negative​​​‌ Matrix Factorization (SNNMF) approach​ was evaluated and demonstrated​‌ satisfactory decomposition performance. However,​​ as illustrated in Fig.​​​‌ 14.B, the initial​ factorization may generate artifacts​‌ in the estimation (​​$W$ columns may represent​​​‌ mixtures of several EMG​ contribution, with excessive energy,​‌ $H$ lines may be​​ correlated, reflecting a dependency​​​‌ between estimated recruitments, $\mathrm{W}$ columns may be correlated,​‌ reflecting a dependency between​​ estimated waves).

These limitations​​​‌ compromise the physiological interpretation​ of the results, particularly​‌ with respect to accurately​​ distinguishing individual muscle contributions.​​​‌

The proposed method, developed​ during a M1 internship,​‌ is based on a​​ reformulation of the classical​​​‌ SNNMF, where the decomposition​ is obtained by minimizing​‌ an enriched cost function.​​ In addition to the​​​‌ reconstruction error term, regularization​ terms have been added​‌ to penalize the energy​​ of the matrix $\mathrm{W‌}$ and reduce correlations between​ the columns of $\mathrm{W‌}$ and between the rows​​ of $H$, thus​​​‌ ensuring greater independence of​ the extracted components. The​‌ optimization of this regularized​​ cost function led to​​​‌ the derivation of new​ analytical expressions for the​‌ gradients, which were then​​ used to define the​​​‌ update rules.

These modified​ approaches show promising results​‌ on synthetic benchmark data,​​ as illustrated in Fig.​​​‌ 14. Quantitative evaluation​ using these synthetic benchmarks​‌ further demonstrates their effectiveness.​​ Promising results were also​​​‌ obtained on real EMG​ recordings. Overall, the proposed​‌ methods show strong potential​​ for real-time muscle recruitment​​​‌ analysis.

8.2.3 Processing of​ complex surface EMG signals​‌ - AI-Hand project

Participants:​​ Hamdi Amani, Maggioni​​​‌ Valentin, Guiho Thomas​, Guiraud David [Neurinnov]​‌.

Analysis of the​​ electrical signals generated by​​​‌ muscles (EMGs) in response​ to neural stimulation makes​‌ it possible to evaluate​​ the selectivity of the​​​‌ stimulation and thus estimate​ its potential in terms​‌ of functional rehabilitation. These​​ EMGs, which are mainly​​​‌ captured by electrodes placed​ on the surface of​‌ the skin in human,​​ contain rich but complex​​​‌ information. The relative distance​ of the electrodes from​‌ the target muscles alters​​ the accuracy of the​​​‌ measurements and often leads​ to the recording of​‌ activity from several muscles​​ (composite EMGs). The decomposition​​​‌ of these signals is​ an important challenge that​‌ must enable: i. the​​ correct monitoring of the​​​‌ target muscles, ii. deciding​ on the selectivity of​‌ the stimulation parameters investigated,​​ and iii. identifying the​​ most selective parameters. To​​​‌ address this issue, time-frequency‌ analysis work based on‌​‌ the Meyer wavelet transform​​ has been undertaken in​​​‌ recent years within the‌ team. The objective of‌​‌ this project was to​​ further this work by:​​​‌

• Comparing the performance of‌ Meyer wavelets with other‌​‌ wavelet families;
• Facilitating the​​ analysis of composite EMG​​​‌ signals through the automation‌ of wavelet decomposition.

To‌​‌ achieve this, existing MATLAB​​ scripts—based on Meyer wavelets—were​​​‌ refined to improve the‌ processing of a dataset‌​‌ from a clinical trial​​ investigating the impact of​​​‌ arm nerve stimulation on‌ restoring wrist and hand‌​‌ function in four quadriplegic​​ individuals.

8.2.4 Optimal Control​​​‌ Framework for Personalized FES-cycling‌ in Individuals with Spinal‌​‌ Cord Injury

Participants: Coelho-Magalhães​​ Tiago, Azevedo Christine​​​‌, Bailly François.‌

We introduce an optimal‌​‌ control framework to enhance​​ Functional Electrical Stimulation (FES)-cycling​​​‌ for individuals with spinal‌ cord injury. The work‌​‌ integrated a stimulation-aware muscle​​ model, calibrated using experimental​​​‌ data from an SCI‌ participant, and demonstrated the‌​‌ feasibility of tracking a​​ 42 rpm cadence and​​​‌ a 20 W power‌ target through optimized pulse-duration‌​‌ control of six muscles​​ 23. The approach​​​‌ achieved an average power‌ output of 17.53 ±‌​‌ 7.87 W and a​​ velocity of 4.09 ±​​​‌ 0.36 rad/s (Fig. 15‌), highlighting the potential‌​‌ of using trajectory optimization​​ to improve the personalization​​​‌ and effectiveness of FES-cycling‌ protocols 25.

The‌​‌ image contains three polar​​ plots showing optimized stimulation​​​‌ patterns, crank angular velocity‌ and power output.

8.2.5 Optimized‌ Stimulation Patterns for FES-assisted‌​‌ Cycling Using an Experimentally​​ Identified Physiological Muscle Model​​​‌

Participants: Coelho-Magalhães Tiago,‌ Azevedo Christine, Bailly‌​‌ François.

This work​​ presents a numerical-optimal-control–compliant muscle​​​‌ model accounting for electrically‌ evoked contractions and its‌​‌ application to FES-assisted cycling.​​ The model incorporates calcium–troponin​​​‌ complex dynamics to more‌ accurately reproduce the nonlinear‌​‌ force response of muscle​​ under electrical stimulation, and​​​‌ it is calibrated using‌ subject-specific isometric torque data‌​‌ collected from an individual​​ with spinal cord injury.​​​‌ The individualized muscle parameters‌ are identified through a‌​‌ trajectory optimization procedure implemented​​ with Bioptim, ensuring compatibility​​​‌ with gradient-based optimal control.‌ Using the calibrated model,‌​‌ predictive simulations of FES-driven​​ pedaling are performed in​​​‌ which stimulation pulse duration‌ are optimized to achieve‌​‌ a target cadence and​​ power output (Fig. 16​​​‌). Results demonstrate coordinated‌ agonist–antagonist activation patterns and‌​‌ realistic pedaling dynamics, highlighting​​ the potential of using​​​‌ optimal control for improving‌ FES cycling performance and‌​‌ contributing toward more effective,​​ personalized rehabilitation strategies.

The​​​‌ image depicts a study‌ on FES-assisted cycling with‌​‌ four sub-figures. (a) Shows​​ the musculoskeletal model used​​​‌ in the study. (b)‌ Features a polar plot‌​‌ of an optimized stimulation​​​‌ patterns over a 360-degree​ cycle. (c) Displays a​‌ polar plot of crank​​ angular velocity over a​​​‌ cycle. (d) Presents a​ polar plor of power​‌ output measured in watts​​ across the cycle. The​​​‌ plots use various colors​ and lines to indicate​‌ different data sets and​​ stages of movement.

8.2.6​ Numerical-Optimal-Control-Compliant Muscle Model for​‌ Electrically Evoked Contractions

Participants:​​ Coelho-Magalhães Tiago, Azevedo​​​‌ Christine, Bailly François​.

A published study​‌ adapted an existing physiological​​ muscle model—designed to predict​​​‌ force generation in response​ to electrical stimulation—to make​‌ it compatible with gradient-based​​ numerical optimal control (​​​‌13). The activation​ dynamics was reformulated to​‌ allow stimulation sequences that​​ vary over time, enabling​​​‌ the simulation of more​ complex FES-assisted movements. Model​‌ parameters were identified using​​ electrically evoked isometric torque​​​‌ data from three individuals​ with spinal cord injury.​‌ Using an optimal control​​ framework, the work demonstrated​​​‌ accurate prediction of knee​ torque and the feasibility​‌ of optimizing stimulation patterns​​ in simulation (Fig. 17​​​‌). This proof of​ concept highlights the potential​‌ of physiological muscle model–based​​ control to personalize and​​​‌ improve functional electrical stimulation​ strategies.

The image depicts​‌ an overview of the​​ methodology used in the​​​‌ paper cited in this​ section (equations, methodology and​‌ an illustration of the​​ musculoskeletal model used in​​​‌ the study.)

8.2.7 Musculotendon Parameters​‌ of the Human Upper​​ Limb: a Scoping Review​​​‌ and Dataset Aggregation

Participants:​ Schegg Pierre, Maggioni​‌ Valentin, Faraud Baptiste​​, Bailly François.​​​‌

We conducted a PRISMA-ScR-compliant​ scoping review of musculotendon​‌ parameters for 50 upper​​ limb muscles, aggregating data​​​‌ from 107 studies and​ 3,742 participants. The database​‌ covers seven parameters, reported​​ across 50 muscles, with​​​‌ physiological cross sectional area​ documented in all 50​‌ muscles, maximal isometric force​​ in 19 muscles, optimal​​​‌ fiber length in 49,​ fiber length in 48,​‌ pennation angle in all​​ 50, tendon slack length​​​‌ in 40, and contraction​ velocity in one. We​‌ provide statistical values of​​ parameters for each muscle,​​​‌ alongside interactive visualization tools.​ To facilitate data exploration,​‌ we developed a dedicated​​ website for visualization and​​​‌ navigation, complemented by open-source​ software offering advanced filtering​‌ and visualization capabilities. The​​ study also discusses biaises​​​‌ and gaps in the​ litterature and makes recommendations​‌ on how to address​​ them. We anticipate that​​​‌ this research and the​ accompanying tools will serve​‌ the biomechanical modeling community​​ by providing both an​​​‌ initial guess for musculoskeletal​ model calibration and a​‌ validation tool. Early results​​ of the study were​​ orally presented at the​​​‌ 50ème Congrès de la‌ Société de Biomécanique by‌​‌ Pierre Schegg and the​​ full length article was​​​‌ submitted to Sports Medicine.‌ In conjunction with this‌​‌ review, we developed open-source​​ software to visualize and​​​‌ filter the musculotendon parameter‌ data. This tool facilitated‌​‌ the analysis and discussion​​ in our article and​​​‌ is designed to be‌ the primary interface for‌​‌ researchers engaging with the​​ database. The software is​​​‌ published on GitLab for‌ open access. Additionally, we‌​‌ created a companion website​​ that displays the data​​​‌ in a browser-accessible format‌ without the filtering options‌​‌ available in the software.​​ The website is a​​​‌ more accessible option since‌ it can be viewed‌​‌ through a browser while​​ the software requires a​​​‌ python environment to be‌ set up.

8.2.8 Upper-limb‌​‌ Musculoskelatal Model Calibration

Participants:​​ Schegg Pierre, Bailly​​​‌ François.

This research‌ focuses on musculoskeletal model‌​‌ calibration and personalization. We​​ investigated the use of​​​‌ static optimization techniques applied‌ to pre-recorded kinematic and‌​‌ electromyographic (EMG) data to​​ calibrate musculotendon parameters of​​​‌ the model. Specifically, we‌ examined how the quantity‌​‌ of input data influences​​ calibration accuracy. We proposed​​​‌ a method to reduce‌ the amount of experimental‌​‌ data needed to accurately​​ identify musculotendon parameters of​​​‌ the model, which led‌ to a poster presentation‌​‌ at the 20th International​​ Symposium on Computer Methods​​​‌ in Biomechanics and Biomedical‌ Engineering (Barcelona, Spain) 27‌​‌. Additionally, we are​​ exploring the statistical and​​​‌ physical characteristics of recorded‌ motion that contribute to‌​‌ effective calibration. This preliminary​​ work is conducted in​​​‌ the context of the‌ B-IRD ANR grant (PI‌​‌ François Bailly) and will​​ be further developed next​​​‌ year with the beginning‌ of Kloé Bonnet's PhD‌​‌ thesis in late 2025.​​

8.2.9 Optimization of Bicycle​​​‌ Designs

Participants: Otmani Sabrina‌, Murray Andrew [Dayton‌​‌ University, OH, USA],​​ Azevedo Christine, Bailly​​​‌ François.

This year‌ marks a milestone for‌​‌ this project in which​​ we work towards the​​​‌ improvement of FES-assisted pedaling‌ performance for people with‌​‌ spinal cord injury (SCI).​​ Our approach, which consists​​​‌ in optimizing the bicycle‌ structural design to maximize‌​‌ the power throughput at​​ the crank at constant​​​‌ speed, led to two‌ journal publications this year.‌​‌

The first article (​​18) explores the​​​‌ customization and optimization of‌ three distinct bicycle drive‌​‌ mechanisms (Fig. 18),​​ leveraging an individual’s biomechanical​​​‌ data to maximize pedaling‌ power throughput. Our approach‌​‌ utilizes torque/velocity/position relationships of​​ the hip and the​​​‌ knee, so that the‌ kinematics of the optimized‌​‌ designs allow the user​​ to pedal with maximized​​​‌ joint torques and thus,‌ enhance the power produced‌​‌ at the crank. The​​ method is applied to​​​‌ the cases of two‌ users with significantly distinct‌​‌ anthropometries, showing noticeable changes​​ in the drive mechanisms​​​‌ and demonstrating its effectiveness‌ for personalizing bicycle designs.‌​‌ The study highlights the​​ importance of considering individual​​​‌ biomechanical factors, showing that‌ even slight variations in‌​‌ design can lead to​​ changes in the cycling​​​‌ kinematics, resulting in improved‌ performance. Simulation results also‌​‌ show increased mean power​​​‌ throughput for more complex​ drive mechanisms compared to​‌ a classical one, regardless​​ of the user profile.​​​‌ This suggests that such​ designs should be capable​‌ of accommodating a range​​ of cyclists, from recreational​​​‌ users to high-performance athletes,​ as well as individuals​‌ and athletes with motor​​ impairments. These findings highlight​​​‌ the potential of biomechanically-informed,​ personalized bicycle drive mechanisms.​‌ Such systems can optimize​​ pedaling efficiency and enhance​​​‌ performance across diverse user​ groups.

The second article​‌ (17) presents​​ a novel crank-pedal mechanism​​​‌ designed to optimize pedal-path​ kinematics. The goal of​‌ the design is to​​ maximize power throughput by​​​‌ utilizing torque-generating capabilities produced​ by individual riders. The​‌ dimensions of the design​​ are determined through an​​​‌ optimization algorithm that modifies​ the crank length, pedal​‌ shape, and frame geometry.​​ The optimization uses the​​​‌ joint position, velocity, and​ torque relationships of a​‌ user (Fig. 19).​​ As such, the solver​​​‌ can take advantage of​ musculoskeletal motions that generate​‌ sustained large torques. The​​ approach is tested in​​​‌ simulation with data from​ two user profiles, demonstrating​‌ similarities and variations that​​ morphologies can produce in​​​‌ the design. In both​ cases, the optimized designs​‌ with the new crank-pedal​​ mechanism improved the mean​​​‌ crank power during a​ crank revolution by approximately​‌ 15% compared to a​​ traditional personalized bicycle. These​​​‌ results suggest that altering​ the dimensions of a​‌ bike using biomechanical data​​ in the design process​​​‌ could have a significant​ impact on the pedaling​‌ performance of individuals, ranging​​ from everyday users to​​​‌ athletes and individuals with​ motor impairments.

The image​‌ contains three diagrams illustrating​​ how the human lower​​​‌ limb in interaction with​ the bicycle is mathematically​‌ modeled (angles, dimensions, relationships).​​

The​‌ image shows how the​​ solver leverages the muscular​​​‌ capacities of the users​ by overlaying the optimized​‌ trajectories (hip and knee​​ joints) with colormaps of​​​‌ the muscular capacities.

8.2.10 Scoping review​‌ on the use of​​ optimal control in FES​​​‌

Participants: Bailly François,​ Begon Mickael [Université de​‌ Montréal, Canada], Co​​ Kevin [Université de Montréal,​​​‌ Canada], Moissonet Florent​ [Université de Genève, Switzerland]​‌.

Following the PRISMA​​ guidelines, a search was​​​‌ conducted up to February​ 2024 using the combined​‌ keywords “FES”, “optimal control”​​ or “fatigue” across five​​​‌ databases (Medline, Embase, CINAHL​ Complete, Web of Science,​‌ and ProQuest Dissertations $\&$ Theses Citation Index) 12​​​‌. Inclusion criteria included​ the use of optimal​‌ control with FES for​​ healthy individuals and those​​​‌ with neuromuscular disorders. Among​ the 44 included studies,​‌ half were in silico​​ and half in vivo,​​ involving 87 participants, predominantly​​​‌ healthy young men. Twelve‌ different motor tasks were‌​‌ investigated, with a focus​​ on single-joint lower-limb movements.​​​‌ These studies principally used‌ simple FES models, modulating‌​‌ pulse width or intensity​​ to track joint-angle.

Optimal​​​‌ control-driven FES can deliver‌ precise motions and reduce‌​‌ fatigue. Yet clinical adoption​​ is slowed down by​​​‌ the lack of consensus‌ about modeling, inconvenient model‌​‌ identification protocol and limited​​ validation. Additional barriers include​​​‌ insufficient open- science practices,‌ computational performance reporting and‌​‌ the availability of customizable​​ commercial hardware. Comparative FES​​​‌ model studies and longitudinal‌ trials with large cohorts,‌​‌ among other efforts, are​​ required to improve the​​​‌ technology readiness level. Such‌ advances would help clinical‌​‌ adoption and improve patient​​ outcomes.

8.3.1 AI-Hand project:‌ Restoring upper-limb functions in‌​‌ individuals with tetraplegia

The​​ AI-Hand European project (EIC​​​‌ pathfinder), focuses on the‌ development of an active‌​‌ implantable medical device (AIMD)​​ for neural stimulation -​​​‌ supported by Neurinnov company‌ - to restore wrist‌​‌ and hand function in​​ individuals with tetraplegia 28​​​‌. In this framework,‌ the project is divided‌​‌ into two successive phases,​​ a phase dedicated to​​​‌ the development/refinement of the‌ approach that will immediately‌​‌ be followed by a​​ clinical phase by 2026​​​‌ (first implantation in people‌ with tetraplegia). The CAMIN‌​‌ team, tasked with coordinating​​ the project, focuses on​​​‌ two key aspects: 1)‌ in-depth study of user‌​‌ stimulation control strategies and​​ 2) support for the​​​‌ development of the implanted‌ solution (by directly managing‌​‌ trials through preclinical testing​​ and being an integral​​​‌ part of clinical trials).‌ Regarding preclinical trials, progress‌​‌ for 2025 is mainly​​ focused on finalizing the​​​‌ processing of previous experiments‌ in human 22,‌​‌ 11 the processing of​​ biological signals and samples​​​‌ collected during acute (PLASTICISTIM)‌ and chronic (AI-Hand) experiments‌​‌ on pigs conducted in​​ 2024.

CT1 non invasive​​​‌ clinical trial

Participants: Azevedo‌ Christine, François Bailly‌​‌, Baptiste Faraud.​​

The first non-invasive Clinical​​​‌ Trial 1 (CT1) was‌ dedicated to assessing the‌​‌ usability of various piloting​​ modalities future implant users​​​‌ will have access to‌ in the medical device‌​‌ developed by NEURINNOV. During​​ this trial, CAMIN was​​​‌ in charge of analyzing‌ the capacity of participants‌​‌ with upper limb paralysis​​ to use the piloting​​​‌ modalities to modulate a‌ command. 10 participants were‌​‌ recruited (USSAP Perpignan Rehabilitation​​ Center). An experimental platform​​​‌ was developed to allow‌ the users to interact‌​‌ with the serious game​​ TARGETTRACK via the control​​​‌ modalities developed by NEURINNOV‌ (see Fig. 20).‌​‌ As described in the​​ "New Platform" section, two​​​‌ games were available: the‌ first one consisted in‌​‌ a simple gauge. The​​ “+/-” inputs turn the​​​‌ needle to the right‌ or left until it‌​‌ reaches the dotted area.​​ The second game consisted​​​‌ in driving the position‌ of a moving character‌​‌ in a cluttered environment​​ with the “+/-“ control​​​‌ inputs to avoid collisions,‌ reflecting the capacity of‌​‌ the user to interact​​ dynamically with the control​​​‌ modalities (joystick or voice‌ control).

A person in‌​‌ a wheelchair (the participant)​​​‌ is seated at the​ center of a desk,​‌ with two other people​​ seated on either side.​​​‌ On the desk, there​ is one laptop and​‌ a main screen, both​​ displaying similar information: a​​​‌ clear pathway surrounded by​ blue vertical columns. A​‌ UFO is displayed on​​ the clear pathway. Photos​​​‌ of the two modalities​ (joystick and module of​‌ the voice control are​​ also displayed)

Acute​ animal experiments - PLASTICISTIM​‌ project

Participants: Baum Jonathan​​, Guiho Thomas,​​​‌ Azevedo Christine, Chamot-Nonin​ Manon [Neurinnov], Guiraud​‌ David [Neurinnov].

The​​ PLASTICISTIM project complements AI-Hand​​​‌ by exploring, as a​ proof of concept, the​‌ combined use of peripheral​​ nerve stimulation (PNS) and​​​‌ spinal cord stimulation (SCS)​ to restore motor function​‌ after spinal cord injury.​​ While PNS enables direct​​​‌ motor activation, SCS has​ been shown to facilitate/strengthen​‌ degraded voluntary commands and​​ support progressive motor recovery,​​​‌ suggesting a potentially synergistic​ effect. In June–July 2024,​‌ the project investigated the​​ combination of transcutaneous SCS​​​‌ (t-SCS) with epineural PNS​ in three pigs. Sessions​‌ of sole PNS stimulation​​ were performed before and​​​‌ after combining PNS with​ t-SCS using state of​‌ the art stimulation devices​​ —Neurinnov’s benchtop stimulator connected​​​‌ to CorTec’s epineural cuff​ electrodes on one side​‌ (PNS) and Pajunk’s Stim2go​​ stimulator on the other​​​‌ side (t-SCS). Muscular responses​ were assessed via custom​‌ implanted EMG needles.

EMG​​ data processing began in​​​‌ early 2025, with an​ initial focus on PNS-evoked​‌ signals to assess the​​ reliability of the PNS​​​‌ stimulation chain but also​ to evaluate the short-term​‌ effect of tSCS on​​ PNS responses by comparing​​​‌ PNS-evoked muscular responses before​ and after tSCS. First​‌ only pure EMG signals​​ — i.e. noise-free signals​​​‌ — were processed to​ compute recruitment curves reflecting​‌ muscles activation following progressive​​ increases in stimulation intensity​​​‌ (Fig. 21). In​ parallel with this work,​‌ a source separation algorithm​​ is currently implemented to​​​‌ enable the processing of​ all data — regardless​‌ of the different sources​​ of noise (measurement or​​​‌ physiological noise).

The image​ is in two parts.​‌ A top subfigure and​​ a bottom subfigure. The​​​‌ top subfigure shows graphs​ of electromyography (EMG) signals​‌ from four different muscles​​ obtained for increasing stimulation​​​‌ intensities: FDS, FCR, PT,​ and ECR. The top​‌ four graphs display voltage​​ (U) versus time (ms)​​​‌ for each muscle, with​ multiple signal traces in​‌ different colors — one​​ trace for every stimulation​​​‌ intensity. The bottom subfigure​ shows graph plotting the​‌ stimulation intensity (I) in​​ microamperes (µA) against the​​​‌ amplitude of the EMG​ signals (output voltage -​‌ U) in millivolts (mV),​​ showing the response of​​​‌ each muscle to varying​ stimulation intensities. The colors​‌ red, blue, green, and​​ purple represent FDS, FCR,​​​‌ PT, and ECR respectively.​

At the​​​‌ end of these acute‌ experiments, stimulated and non-stimulated‌​‌ (control) nerves were harvested​​ for histological purposes. Three​​​‌ features conditioning the impact‌ of stimulation on nerves‌​‌ were investigated: 1) global​​ morphometric structure of the​​​‌ nerve using Hematoxylin-eosin staining;‌ 2) identification and distribution‌​‌ of motor fibers within​​ the nerve via anti-ChAT​​​‌ immunostaining; and 3) characterization‌ of fiber diameter with‌​‌ toluidine blue staining —​​ myelin sheath thickness —​​​‌ (Fig. 22).

The‌ image shows three microscopic‌​‌ views of a tissue​​ sample. These views are​​​‌ corresponding to cross-sectional sections‌ of one of the‌​‌ stimulated peripheral nerves The​​ first, stained with hematoxylin-eosin,​​​‌ displays a cellular structure‌ with pink and red‌​‌ hues (the fascicles). The​​ second, labeled ChAT, shows​​​‌ a more subdued view‌ with brownish spots indicating‌​‌ enzyme activity specific to​​ motor neurons (enabling identification​​​‌ of the positioning of‌ these motor neurons). The‌​‌ third, stained with toluidin​​ blue, highlights the tissue​​​‌ in various tones of‌ blue, showing detailed cellular‌​‌ and structural elements. Among​​ these details, the thickness​​​‌ of the myelin sheath‌ surrounding these neurons is‌​‌ of particular interest.

The rationale behind​​ these histological studies is​​​‌ to enable comparison of‌ the 3D distribution of‌​‌ injected currents (so called​​ configurations of current) with​​​‌ the actual architecture of‌ the stimulated nerve in‌​‌ order to better understand​​ the evoked muscle responses​​​‌ (24). This‌ comparative study, made possible‌​‌ by the collection of​​ stimulated nerve samples —​​​‌ and therefore restricted to‌ animal testing — provides‌​‌ a rare opportunity to​​ better understand the links​​​‌ between stimulation, nerve architecture,‌ and muscle response. With‌​‌ this in mind, a​​ reconstruction of the relative​​​‌ position of the electrode‌ around the nerve and‌​‌ a comparison with the​​ recruitment curves obtained by​​​‌ stimulation is currently underway‌ (Fig. 23).

The‌​‌ image consists of two​​ main parts: a radar​​​‌ chart on the left‌ and a histological section‌​‌ on the right. The​​ radar chart shows six​​​‌ axes labeled STR1 to‌ STR6 - each of‌​‌ these corresponding to a​​ specific stimulation configuration with​​​‌ a single active side‌ parametered as Cathode turning‌​‌ progressively around the nerve​​ - starting from the​​​‌ first active site as‌ cathode STR1 to the‌​‌ sixth contact as cathode​​ STR6 (justifying the radar​​​‌ representation), each with values‌ ranging from 0 to‌​‌ 1. This scores corresponds​​ to normalized muscles activity​​​‌ (normalized by the maximum‌ amplitude of the corresponding‌​‌ EMG) and compares activation​​ of four diffrent muscles:​​​‌ FDS (blue), FCR (orange),‌ PT (yellow), and ECR‌​‌ (purple). The right side​​ features a labeled microscopic​​​‌ image corresponding to the‌ cross-section of the correspondig‌​‌ nerve, highlighting specific regions​​ with preferential muscle activation:​​​‌ FCR, PT, ECR, and‌ FDS, with boundaries marked‌​‌ by black and red​​ lines. The labeled regions​​​‌ correspond to the parameters‌ in the radar chart.‌​‌ The image depicts cellular​​ structures within a designated​​​‌ area (mainly fascicular anatomy‌ and motor neurons location).‌​‌

8.3.2 Correcting​ the Gait in Real-Time​‌ for Children with Cerebral​​ Palsy

Participants: Graffagnino Gabriel​​​‌, Gasq David,​ Patte Karine [Institut Saint-Pierre]​‌, Sijobert Benoît,​​ Azevedo Christine.

Gabriel’s​​​‌ thesis, funded by an​ INRIA-INSERM grant, focuses on​‌ pathological gait in children​​ with cerebral palsy (CP).​​​‌

CP is the most​ prevalent motor disorder in​‌ childhood and often results​​ in gait abnormalities that​​​‌ hinder mobility and diminish​ quality of life. Functional​‌ electrical stimulation (FES) has​​ demonstrated potential in enhancing​​​‌ gait in individuals in​ this population 29,​‌ however, its practical implementation​​ remains complex, as it​​​‌ requires monitoring various gait​ parameters and delivering personalized​‌ stimulation to different muscles​​ in order to correct​​​‌ various gait impairments. Recent​ advancements in real-time motion​‌ capture (MOCAP) and wearable​​ sensors now enable the​​ development of closed-loop, multi-channel​​​‌ FES systems.

In this‌ context, we developed a‌​‌ real-time, event-triggered multi-channel stimulation​​ protocol during treadmill walking,​​​‌ and assessed the feasibility‌ and responsiveness of it‌​‌ 26. The stimulation​​ was triggered by specific​​​‌ gait events (heel strike,‌ knee flexion, ankle dorsiflexion)‌​‌ detected through the MOCAP​​ system and administered via​​​‌ a multichannel electrical stimulator.‌ We emulated the real-time‌​‌ kinematic acquisition seen in​​ children with CP using​​​‌ data already acquired to‌ assess the technical feasibility.‌​‌ We reported different technical​​ outcomes including the latency​​​‌ between gait event detection‌ and triggering function calling‌​‌ in the algorithm, and​​ the latency between this​​​‌ function call and the‌ start message sent to‌​‌ the stimulator. The results​​ confirm the viability of​​​‌ the system, laying the‌ groundwork for future clinical‌​‌ application in the rehabilitation​​ of children with CP​​​‌ (Fig. 24).

The‌ image illustrates a process‌​‌ for detecting gait events​​ and triggering stimulation. It​​​‌ begins with motion capture‌ (a), showing a person‌​‌ walking on a treadmill​​ monitored by cameras. The​​​‌ captured data is analyzed‌ to detect gait events‌​‌ (b), represented by a​​ graph of knee flexion​​​‌ over time. This information‌ is processed by a‌​‌ triggering algorithm (c), which,​​ upon detecting a gait​​​‌ event (T1), initiates a‌ function call (T2), leading‌​‌ to the sending of​​ a start message (T3)​​​‌ to a stimulation trigger‌ device. The process involves‌​‌ complete tracking of a​​ walking individual's movement, data​​​‌ analysis, decision-making through an‌ algorithm, and triggering of‌​‌ a specific action.

An important contribution​​ this year was also​​​‌ to analyze complementary biomarkers‌ of gait in children‌​‌ with cerebral palsy to​​ insist on their complementarity​​​‌ 14.

8.3.3 Systematic‌ review to explore assistive‌​‌ devices designed to improve​​ upper-limb movements

Participants: Charlotte​​​‌ Le Goff, Pauline‌ Coignard [Association Approche],‌​‌ Charles Fattal [Centre Bouffard​​ Vercelli USSAP], Azevedo​​​‌ Christine.

Among people‌ with disabilities resulting from‌​‌ chronic illnesses, accidents, or​​ aging, upper limb (UL)​​​‌ motor impairments are particularly‌ common and hinder independence‌​‌ in activities of daily​​ living (ADL). Assistive technology​​​‌ devices offer promising solutions,‌ but their diversity and‌​‌ level of maturity remain​​ variable. In the context​​​‌ of PEPR O2R ASSISTMOV‌ project, we carried a‌​‌ systematic review to explore​​ assistive devices designed to​​​‌ improve UL movement in‌ people with disabilities, in‌​‌ a functional assistance context.​​ A systematic search was​​​‌ conducted according to PRISMA‌ guidelines, with rigorous study‌​‌ selection. Several databases were​​ searched, including PubMed, Pascal,​​​‌ IEEExplore, and EBSCO. The‌ inclusion criteria were as‌​‌ follows: a study of​​​‌ an UL assistive device​ used in ADLs or​‌ whose concept was transposable​​ to this use, and​​​‌ published in English. Forty-five​ studies were selected, the​‌ majority published after 2010,​​ and involving single testing​​​‌ sessions. Of these, 46.7​$\%$ involved patients with​‌ chronic stroke, with a​​ particular focus on distal​​​‌ deficits (hand, wrist, fingers).​ The most frequently studied​‌ devices are mechatronic exoskeletons​​ and gripping gloves. Although​​​‌ numerous prototypes have been​ developed, few are currently​‌ available on the market,​​ limiting their accessibility for​​​‌ users. We concluded that​ existing technologies offer benefits​‌ in terms of functional​​ autonomy and quality of​​​‌ life, but still face​ constraints related to ergonomics,​‌ cost, and portability. Assistive​​ devices for daily living​​​‌ activities represent a promising​ but still limited technological​‌ field. This review highlights​​ the importance of user-centered​​​‌ development and the need​ to strengthen the methodological​‌ robustness of future studies​​ to prioritize innovative, modular,​​​‌ and accessible solutions 21​.

8.3.4 Grasping intention​‌ estimator

Participants: Moullet Etienne​​ [INRIA WILLOW], Azevedo​​​‌ Christine, Bailly François​, Justin Carpentier [INRIA​‌ WILLOW].

We have​​ developed i-GRIP, a framework​​​‌ that decodes grasp intention​ - specifically target object​‌ and grasp type -​​ from natural upper-limb kinematics​​​‌ during reaching within a​ known scene containing multiple​‌ candidate objects. We have​​ investigated its real-time operation​​​‌ and user adaptation when​ deployed as a grasp-assistance​‌ control interface in an​​ immersive virtual reality (VR)​​​‌ environment (Figs. 25 and​ 26). Two control​‌ modes were compared: natural​​ control, in which​​​‌ the virtual hand directly​ mirrored the user’s hand​‌ motion, and assisted control​​, in which finger​​​‌ motion was delegated to​ i-GRIP predictions while hand​‌ position remained user-driven. Twenty-two​​ healthy participants performed 3,300​​​‌ grasping trials (3,244 retained​ for analysis). The VR​‌ environment demonstrated ecological validity,​​ as movements produced under​​​‌ natural control exhibited completion​ durations comparable to those​‌ reported in physical-world settings.​​ Under assisted control, task​​​‌ success remained high (93–96%),​ despite longer and less​‌ smooth movements compared to​​ natural control. Mixed-effects analyses​​​‌ revealed robust learning effects​ across trials, with increasing​‌ odds of task success​​ and progressive reductions in​​​‌ movement duration and velocity​ peaks.

The image shows​‌ a person wearing a​​ VR headset, seated in​​​‌ a chair with their​ arm extended. They are​‌ interacting with a virtual​​ reality environment. In the​​​‌ VR scene, there are​ several household items on​‌ a table, including a​​ carton of milk, a​​​‌ bottle of mustard, a​ bottle of dish soap,​‌ and a jar of​​ tomato sauce. A virtual​​​‌ arm is reaching towards​ these items. There is​‌ also a button labeled​​ "start experiment" on the​​​‌ table. The person appears​ to be participating in​‌ an experiment within the​​ VR setting.

The​​​‌ image contains six sub-images​ labeled (A) through (F):​‌ (A) A 3D model​​ of a virtual hand​​ with a central positional​​​‌ marker on the palm.‌ (B) A virtual reality‌​‌ scene showing a table​​ with various objects including​​​‌ a sphere, bottles, and‌ a box. Instructions and‌​‌ status are displayed on​​ a screen. (C) Similar​​​‌ as the scene (B),‌ but when a button‌​‌ is pressed with the​​ virtual hand, the ghost​​​‌ outline of another hand‌ appears and performs a‌​‌ grasping action to be​​ reproduced by the participant.​​​‌ (D) In the same‌ virtual scene, a completion‌​‌ feedback is given to​​ the participant when the​​​‌ task has been performed‌ successfully by highlighting the‌​‌ hand in green color.​​ (E) In the same​​​‌ virtual scene, a failure‌ feedback is given to‌​‌ the participant when the​​ task has been performed​​​‌ wrongly by highlighting the‌ hand in red color.‌​‌ (F) In a similar​​ virtual scene, a yellow​​​‌ overlay is placed on‌ the object being targetted‌​‌ by the virtual hand​​ and depending on the​​​‌ approach trajectory of the‌ hand,grasping strategies are chosen‌​‌ (whether pinch or palmar​​ grasp) to control the​​​‌ opening and closing of‌ the hand.

10.1.1 Inria​​ associate team not involved​​​‌ in an IIL or‌ an international program

10.1.2 Visits​​​‌ of international scientists

10.1.3​‌ Visits to international teams​​

10.1.4​​ Horizon Europe

INRIA-INSERM​​​‌ Phd thesis grant (2023-26)‌

• Coordinator: Christine Azevedo (INRIA).‌​‌
• We obtained a grant​​ to finance the PhD​​​‌ thesis of Gabriel Graffagnino‌ between CAMIN and INSERM‌​‌ Tonic team (CHU Toulouse)​​ in collaboration with Institut​​​‌ Saint Pierre (Palavas).

ARC‌ FOUNDATION for Research Against‌​‌ Cancer (2022-2025)

• Coordinator: François​​ Bonnetblanc (INRIA), (collaboration with​​​‌ Pr Hugues Duffau (CHU‌ Montpellier) and Pr Emmanuel‌​‌ Mandonnet (APHP)). Guiding brain​​ tumor surgery in real​​​‌ time using electrophysiology

  During‌ the resection of brain‌​‌ tumors, the neuro-surgeons have​​ substantial imaging data allowing​​​‌ them to plan their‌ gesture upstream. However, during‌​‌ the actual surgical gesture,​​ in real time, this​​​‌ imaging becomes ineffective due‌ to the deformation of‌​‌ the brain (so called​​ brain shift). It is​​​‌ then possible to use‌ direct electrical stimulation of‌​‌ the brain in an​​ awake patient who cooperates​​​‌ with the neurosurgeon to‌ determine the functional areas‌​‌ and those which are​​ not. When patients are​​​‌ under general anesthesia this‌ possibility no longer exists.‌​‌ We have planned to​​ use the electrophysiology evoked​​​‌ by the DES of‌ the brain during brain‌​‌ surgery to diagnose and​​ determine the location the​​​‌ tumor and the anatomical‌ connectivity on-line in order‌​‌ to guide the surgery​​ in awake patients or​​​‌ under general anesthesia. This‌ work needs to go‌​‌ beyond the proof of​​ concept we have already​​​‌ performed, and necessitates addressing‌ and solving some methodological‌​‌ challenges. At a fundamental​​ level, this will also​​​‌ help to better understand‌ the electrophysiological effect of‌​‌ DES in order to​​ optimize its use.

AEx​​​‌ noCNN (2024-27)

• Coordinators :‌ François Bailly and François‌​‌ Bonnetblanc (INRIA).

• No-brain-shift and​​​‌ Comprehensive Neurosurgical Navigation using​ computer vision, funded by​‌ INRIA's Action Exploratoire program.​​

  Preoperative MRI is central​​​‌ to preparing for tumor​ resection in the brain.​‌ Throughout the operation, it​​ serves as a reference​​​‌ point for the surgeon​ to guide their actions​‌ using neuronavigation, a technology​​ that allows the surgeon’s​​​‌ tools used in the​ operating room to be​‌ represented in this MRI.​​ However, during the operation,​​​‌ the brain loses its​ initial shape, distorting this​‌ representation. In collaboration with​​ neurosurgeons, we are working​​​‌ to model this phenomenon​ in real time in​‌ order to correct neuronavigation​​ and better guide surgical​​​‌ procedures. More specifically, using​ computer vision and deep​‌ learning, the objective of​​ the noCNN project is​​​‌ to (i) accurately, automatically​ and continuously reconstruct the​‌ volume of the brain​​ exposed by craniotomy during​​​‌ neurosurgeries and (ii) reposition​ this volume deformed by​‌ the decrease in intracranial​​ pressure and resection in​​​‌ standard imaging (MRI).

PEPR​ O2R ASSISTMOV

• Coordinators :​‌ François Bailly and Christine​​ Azevedo.
• The integrated PI3​​​‌ project “ASSISTMOV”, made up​ of a multidisciplinary team​‌ in engineering and Social​​ Sciences and Humanities (SSH),​​​‌ targets the use case​ of assistive robotics for​‌ movement support for people​​ with disabilities. Through the​​​‌ development of a range​ of exoskeletons (for both​‌ lower and upper limbs),​​ the project aims to​​​‌ deliver a disruptive technology​ enabling smooth interaction that​‌ is robust across a​​ wide variety of environments​​​‌ and uses (from rehabilitation​ to everyday life). The​‌ project finances the PhDs​​ of Charlotte Le Goff​​​‌ (Association Approche) and Amina​ Ferrad (INRIA).

Handitech Lab​‌ Inria (HLI)

• Coordinators: Christine​​ Azevedo and Roger Pissard-Gibollet​​​‌ (INRIA).
• Developing technological solutions​ for and with people​‌ with handicap. INRIA cross-cutting​​ action.

ANR B-IRD (2024-28)​​​‌

• Coordinator: François Bailly (INRIA).​

  Biomechanically-Informed Rehabilitation Devices :​‌ Fast and reliable biomechanical​​ methods dedicated to assistive​​​‌ technologies.

Grasp-Again Project Maturation​ Société d’Accélération du Transfert​‌ de Technologies (SATT) (2024-25)​​

• Coordinators: David Gasq (CHU​​​‌ de Toulouse), Ronan Le​ Guillou (INRIA), Christine Azevedo​‌ (INRIA).
• In cooperation with​​ Toulouse Tech Transfer (SATT​​​‌ of Toulouse), the know-how​ acquired through the Prehens-Stroke​‌ and Grasp-Again clinical research​​ protocols on the usage​​​‌ and developement of a​ grasping assistance neuroprosthesis was​‌ officialy commited as intellectual​​ property as an e-Soleau​​​‌ enveloppe. Furthermore, through a​ "Project Maturation" dedicated funding,​‌ the Digital Medical Hub​​ (DMH) and Aguila Technologies​​​‌ were contracted to investigate​ market access considerations and​‌ future potential roadmaps.

ANR​​ JCJC AT-Reach (2025-29)

• Coordinator:​​​‌ Thomas Guiho (INRIA).

• Title:​ Computational models to optimize​‌ functional rehabilitation of upper-limb​​ functions: Relevance of pairing​​​‌ peripheral nerve and spinal​ cord stimulations.

  Bilateral loss​‌ of upper-limb functions after​​ complete cervical Spinal Cord​​​‌ Injury (SCI) dramatically impacts​ people ability to live​‌ independently. Although, considerable progress​​ has been made in​​​‌ the field of Spinal​ Cord Stimulation (SCS) for​‌ rehabilitation of motor functions​​ in recent years, the​​​‌ most significant results are​ still achieved through long-lasting​‌ clinical trials combining SCS​​ with task-specific physical exercises.​​​‌ Alongside these studies, recent​ clinical trials reported promising​‌ outcomes after performing peripheral​​ nerve stimulation (PNS) with​​ implanted electrodes which immediately​​​‌ restores functional – albeit‌ coarse – movements of‌​‌ the upper limb. The​​ AT-REACH proposal aims at​​​‌ paving the way to‌ the next generation of‌​‌ clinical protocols by investigating​​ the added-value of pairing​​​‌ PNS with SCS. The‌ AT-REACH multimodal approach will‌​‌ call for research beyond​​ state-of-the-art and test these​​​‌ hypotheses via the conception‌ of biomimetic computational models‌​‌ supported by animal experiments​​ in order to 1-​​​‌ elucidate SCS mechanisms of‌ action and 2- improve‌​‌ knowledge on spinal cord​​ neural networks in intact​​​‌ and neurologically impaired conditions.‌ Progress is expected at‌​‌ the crossroads of engineering,​​ neuroscience and rehabilitation medicine.​​​‌

11.1.1 Scientific events:​​ organization

11.1.2 Scientific events:​​​‌ selection

11.1.3 Journal

11.1.4 Invited talks

• Gabriel​​ Graffagnino FES Vienna 2025​​​‌ - Real-time gait event‌ detection using motion capture‌​‌ to control an electrical​​​‌ stimulator: Proof-of-concept - September​ 15th to 18th 2025​‌
• François Bonnetblanc : The​​ BCI & Neurotechnology Spring​​​‌ School, keynote speaker 2025,​ link(Participants from 140​‌ countries joined the Spring​​ School; More than 90,000​​​‌ people attended the live​ sessions; Over 550,000 views​‌ were recorded across the​​ 10 days of the​​​‌ event; 140 lectures were​ delivered)
• François Bailly was​‌ an invited speaker at​​ the Inria-Brasil Workshop on​​​‌ Digital Health, online, April​ 2025. link
• Christine Azevedo​‌ was an invited speaker​​ at the Inria-Brasil Workshop​​​‌ on Digital Health, online,​ April 2025. link
• Christine​‌ Azevedo and Roger Pissard-Gibollet​​ (SED Grenoble) presented HumanLab​​​‌ HLI actions during GDR​ Robotic seminar on Making​‌ robotics accessible to a​​ non-expert audience (May 2025).​​​‌
• Christine Azevedo presented her​ activities on science dissemination​‌ towards children during GDR​​ Robotic seminar on Making​​​‌ robotics accessible to a​ non-expert audience (May 2025).​‌
• Christine Azevedo and Roger​​ Pissard-Gibollet (SED Grenoble) presented​​​‌ HumanLab HLI actions during​ INRIA scientific Days "Handicap"​‌ in Paris (June 2025).​​
• Christine Azevedo presented AGILIS​​​‌ and AI-Hand projects actions​ during Program INRIA Quadrant​‌ annual meeting in Paris​​ (June 2025).
• Christine Azevedo​​​‌ presented AGILIS and AI-Hand​ projects at Kerpape Rehabilitiation​‌ Center (Rennes) (June 2025).​​
• Christine Azevedo and Charles​​​‌ Fattal presented Ai-Hand and​ Freewheels projects at a​‌ fundraising evening organized by​​ the USSAP endowment fund​​​‌ in Perpignan (November 2025).​
• Christine Azevedo presented HumanLab​‌ HLI actions during European​​ Week SEEPH "Handicap" in​​​‌ Rocquencourt (November 2025).
• Christine​ Azevedo presented HumanLab HLI​‌ actions in Saclay INRIA​​ Center (October 2025).
• Christine​​​‌ Azevedo presented HumanLab HLI​ during a webinar for​‌ INRIA Alumni association (November​​ 2025).
• Christine Azevedo and​​​‌ Charles Fattal presented AGILIS​ and AGILISTIM projects during​‌ the USSAP annual day​​ in Narbonne (November 2025).​​​‌
• François Bailly was an​ invited speaker at the​‌ workshop “Barriers and Facilitators​​ in FES Cycling: Bridging​​​‌ Clinical Insights and Technological​ Advances”, Rehabweek 2025, Chicago,​‌ USA. link
• François Bailly​​ was an invited speaker​​​‌ at the "Journées Nationales​ de la Recherche en​‌ Robotique" 2025, Rennes, France,​​ link
• Thomas Guiho was​​​‌ invited to give a​ talk on "Functional central​‌ and peripheral neural stimulation​​ for rehabilitation" at Grenoble​​​‌ Neurotechschool 2025 in Aussois.​

11.1.5 Leadership within the​‌ scientific community

• Christine Azevedo​​ is member of the​​​‌ International Functional Electrical Stimulation​ Society (IFESS) society board.​‌

11.1.6 Scientific expertise

• Christine​​ Azevedo is member of​​​‌ Program INRIA Quadrant (PIQ)​ expert committee.
• François Bonnetblanc​‌ belongs to the college​​ of experts for the​​​‌ European Science Foundation and​ makes regular expertise for​‌ this institution.
• Christine Azevedo​​ is a member of​​​‌ the National Scientific Advisory​ Board of the Robotics​‌ Research Group (GdR 3072).​​

11.1.7 Research administration

• Christine​​​‌ Azevedo and Roger Pissard-Gibollet​ (SED Grenoble) coordinate the​‌ HanditechLab Inria.
• Thomas​​ Guiho is responsible for​​​‌ the “Neuroprostheses” teaching unit​ (Université de Montpellier, Dpt​‌ EEA). This unit is​​ an option common to​​​‌ all the masters of​ the Information and Communication​‌ Technologies (ICT) for health​​ training package.
• Jonathan Baum​​​‌ managed Inria Montpellier PhD​ seminars (with Anne Bernard​‌ from LEMON Inria team)​​
• Ronan Le Guillou was​​ a member of the​​​‌ organization commitee for the‌ 6-7 November 2025 meeting‌​‌ in Montpellier of the​​ Inria Thematic Network (Réseau​​​‌ Thématique) on Prototyping.
• Ronan‌ Le Guillou is a‌​‌ member of the organization​​ commitee for the 20-21​​​‌ January 2026 meeting in‌ Lyon of the Inria‌​‌ Thematic Network (Réseau Thématique)​​ named LLM4Prod and intended​​​‌ to help merge the‌ experiences of the various‌​‌ Inria centers in developing​​ and proposing Large Language​​​‌ Models (LLM) for the‌ various needs of Inria‌​‌ agents.
• François Bonnetblanc co-supervised​​ with Christophe Botella the​​​‌ local comittee for sustainable‌ development.

11.2.1 Teaching

• Gabriel Graffagnino‌ Sensory supplementation (1.5h) -‌​‌ Neuroprosthetics Teaching Unit, SNS​​ Master 2, University of​​​‌ Montpellier - Introduction to‌ Electronics (12h) - MEA3,‌​‌ Polytech Montpellier - Logic​​ Systems Programming (36h) -​​​‌ MEA3, Polytech Montpellier -‌ Signal Processing (18h) -‌​‌ MEA3, Polytech Montpellier
• Pierre​​ Schegg Licence STAPS, Montpellier​​​‌ University, France : “Movement‌ and Performance Analysis”, 8h,‌​‌ Licence Mention Education et​​ Motricité - 16h, Licence​​​‌ Mention Entraînement sportif -‌ 8h, Licence Mention Management‌​‌ du Sport.
• Pierre Schegg​​ Master 2 ICT for​​​‌ Health, Montpellier University, France,‌ Neuroprosthesis option: “Introduction to‌​‌ signal processing and Brain​​ Computer Interfaces”, 3h -​​​‌ “Introduction to biomechanics”, 3h‌ - “Introduction to Control‌​‌ Theory”, 3h - “Lab:​​ Signal Processing with Python”,​​​‌ 6h.
• Master ICT for‌ Health, Neuroprotheses option: Charles‌​‌ Fattal, “Neuroprosthesis and motor​​ support strategies after spinal​​​‌ cord injuries”, 3h, M2,‌ Montpellier University, France
• Master‌​‌ ICT for Health, Neuroprotheses​​ option: Olivier Rossel, “Modeling​​​‌ of the peripheral nervous‌ system”, 3h, M1, Montpellier‌​‌ University, France
• Master ICT​​ for Health, Neuroprotheses option:​​​‌ Ronan Le Guillou, “Control‌ basics and signal processing”,‌​‌ 3h, M1, Montpellier University,​​ France
• Master ICT for​​​‌ Health, Neuroprotheses option: Valentin‌ Maggioni, “Biomaterials and biocompatibility”‌​‌ and "Signal processing of​​ neural signals", 22h, M2,​​​‌ Montpellier University, France
• Master‌ Biologie Santé, Module Approches‌​‌ Bioniques Program Integrated Pathophysiology​​ Charles Fattal, 3h, M2,​​​‌ Montpellier University, France
• Master‌ Cognitive and integrated neuroscience,‌​‌ "Sensorimotor Deficiencies and palliative​​ strategies teaching unit": Thomas​​​‌ Guiho, “Implantable neuroprosthesis for‌ motor rehabilitation”, 4.5h, M2,‌​‌ Paul Sabatier University, Toulouse,​​ France
• State diploma of​​​‌ “hearing aid professional”: Jonathan‌ Baum, “office automation”, 33h,‌​‌ 1st year, audiocampus, Montpellier​​ University, France
• Pierre Schegg​​​‌ Master 2 Robotics, Montpellier‌ University, France: “Perception for‌​‌ Robotics”, 30h.

11.2.2 Supervision​​

11.2.3​‌ Juries

• Christine Azevedo was​​ president for the PhD​​​‌ thesis defense of Mathieu​ Celerier "Interaction Homme-Robot Physique​‌ Soutenue: Des Mouvements Inspirés​​ de l'Humain au Contrôle​​​‌ Sûr, Adaptable et Précis​ vers une Industrie centrée​‌ sur l'Humain." Montpellier University,​​ December 16th, 2025.
• Christine​​​‌ Azevedo was president for​ the PhD thesis defense​‌ of Abdelwaheb Hafs "Commande​​ prédictive par jeux différentiels​​​‌ pour l'assistance intuitive du​ mouvement lors d'une interaction​‌ humain-robot." Paris-Saclay University, December​​ 8th, 2025.
• Christine Azevedo​​ was reviewer for the​​​‌ PhD thesis defense of‌ Edouard Ferrand "Interfaçage d'une‌​‌ prothèse bidirectionnelle chez la​​ souris." Paris Saclays University,​​​‌ October 7th, 2025.
• Gabriel‌ Graffagnino Master's internship defense‌​‌ - SNS Master 1,​​ University of Montpellier
• Gabriel​​​‌ Graffagnino, Jonathan Baum, Olivier‌ Rossel and Pierre Schegg‌​‌ were members of the​​ juries assessing the work​​​‌ performed by ICT for‌ health Master’s students during‌​‌ their 2-month projects in​​ immersion in public laboratories​​​‌ or private companies.
• François‌ Bailly was member of‌​‌ the jury for the​​ selection of a maître​​​‌ de conférence for the‌ Université de Montpellier (section‌​‌ 74)

11.2.4 Educational and​​ pedagogical outreach

• Gabriel Graffagnino​​​‌ participated in the finals‌ for the Occitanie region‌​‌ of "My Thesis in​​ 180 seconds" (MT180) in​​​‌ March 28th 2025
• Thomas‌ Guiho participated in a‌​‌ round table discussion on​​ ethics in digital health​​​‌ at the University of‌ Montpellier in December 2025.‌​‌
• Jonathan Baum welcomed a​​ 1 week internship of​​​‌ an 8th grade children‌ this year

11.3.1 Productions (articles, videos,​​ podcasts, serious games, ...)​​​‌

• The latest video capsule‌ retracing the life of‌​‌ the AGILIS and AGILISTIM​​ projects is now online​​​‌ video
• G Graffagnino participated‌ in "Dance your PhD"‌​‌ - video
• The Ai-Hand​​ project was presented in​​​‌ a television report on‌ the program MAG SANTÉ,‌​‌ broadcast on France TV​​ on May 5, 2025.​​​‌
• Christine Azevedo gave an‌ interview to CORTEC, a‌​‌ partner company of AI-Hand,​​ sharing insights into the​​​‌ project. Two videos are‌ available online on their‌​‌ website. video
• Thomas Guiho​​ participated in a “Lab​​​‌ Santé” round table discussion‌ organized by the newspaper‌​‌ “midi libre” on the​​ theme:“Moving differently thanks​​​‌ to technology”
• Thomas Guiho‌ participated in a webinar‌​‌ on November 25 to​​ share feedback on Chiche!​​​‌ interventions.

11.3.2 Participation‌ in Live events

CHICHE‌​‌ program

• Thomas Guiho is​​ the Montpellier referent for​​​‌ the "1 Chercheur, 1‌ Classe : Chiche!" programme.‌​‌
• Thomas Guiho spoke to​​ students 2 second-year classes​​​‌ at Lycée Notre Dame‌ de la Merci in‌​‌ Montpellier on January 15th,​​ 2025 as part of​​​‌ the "1 Chercheur, 1‌ Classe : Chiche!" program.‌​‌
• Thomas Guiho spoke to​​ students 4 second-year classes​​​‌ at Lycée Philippe de‌ Girard in Avignon on‌​‌ January 31th, 2025 as​​ part of the "1​​​‌ Chercheur, 1 Classe :‌ Chiche!" program.
• Gabriel Graffagnino‌​‌ and Thomas Guiho spoke​​ to students in 4​​​‌ second-year classes at Lycée‌ Joseph Joffre in Montpellier‌​‌ on February 6th, 2025​​ as part of the​​​‌ "1 Chercheur, 1 Classe‌ : Chiche!" program.
• Clotilde‌​‌ Turpin, Gabriel Graffagnino and​​ Thomas Guiho spoke to​​​‌ students in 14 second-year‌ classes at Lycée Jean‌​‌ Vilar in Villeneuve-lès-Avignon on​​ April 7th, 2025 as​​​‌ part of the "1‌ Chercheur, 1 Classe :‌​‌ Chiche!" program.
• François Bonnetblanc​​ and Thomas Guiho spoke​​​‌ to students in 6‌ second-year classes at Lycée‌​‌ Déodat de Severac in​​ Ceret on December 9th,​​​‌ 2025 as part of‌ the "1 Chercheur, 1‌​‌ Classe : Chiche!" program.​​

Other interventions

• Thomas Guiho​​​‌ helped supply the prizes‌ for the winners of‌​‌ the Olympiad in Mathematics.​​​‌
• Christine Azevedo participated in​ a mediation/outreach activity around​‌ the project “Sport, Brain​​ and Nutrition”, in collaboration​​​‌ with Lycée Françoise Combes​ in Montpellier and Genopolys.​‌ The project aimed to​​ introduce a second-year class​​​‌ to the research being​ carried out in these​‌ fields, as well as​​ to careers in research.​​​‌ This involved a 1-hour​ classroom session followed by​‌ participation in the students’​​ final presentations in May​​​‌ over the course of​ a morning.
• Christine Azevedo​‌ gave introduction to programming​​ interventions using Thymio Robot​​​‌ (4 sessions of 1,5​ hour) in two 6th-grade​‌ classes at Collège Léon​​ Cordas.
• François Bailly gave​​​‌ introduction to programming interventions​ using Thymio Robot (half​‌ a day) in two​​ 6th-grade classes at Collège​​​‌ Léon Cordas, Montpellier.

11.3.3​ Others science outreach relevant​‌ activities

• Ten team members​​ participated in the 3-day​​​‌ hackathon FABRIKARIUM organized by​ the HumanLab Saint Pierre​‌ (LINK).
• Jonathan Baum, Baptiste​​ Faraud and Gabriel Graffagnino​​​‌ presented live demonstrations of​ EMG recordings for interactions​‌ at "Nuit méditerranéenne des​​ chercheuses 2025" - Mediterranean​​​‌ Researchers' Night MEDNIGHT -​ September 26th 2026.

International‌​‌ journals

• 11 articleC.​​Christine Azevedo Coste,​​​‌ T.Thomas Guiho,‌ F.Fernanda Ferreira,‌​‌ F.François Bailly,​​ B.Benjamin Degeorge,​​​‌ A.Antoine Geffrier,‌ D.David Andreu,‌​‌ J.Jacques Teissier,​​ D.David Guiraud and​​​‌ C.Charles Fattal.‌ Restoring hand grasping for‌​‌ patients with a complete​​ tetraplegia with selective epineural​​​‌ stimulation: a parsimonious thus‌ relevant clinical approach.‌​‌Journal of NeuroEngineering and​​ RehabilitationJanuary 2026HAL​​​‌DOIback to text‌
• 12 articleK.Kevin‌​‌ Co, M.Mickaël​​ Begon, F.François​​​‌ Bailly and F.Florent‌ Moissenet. Optimal control-driven‌​‌ functional electrical stimulation: A​​ PRISMA-ScR scoping review.​​​‌Computers in Biology and‌ Medicine2025HALDOI‌​‌back to text
• 13​​ articleT.Tiago Coelho-Magalhães​​​‌, C.Christine Azevedo‌ Coste and F.François‌​‌ Bailly. Numerical-Optimal-Control-Compliant Muscle​​ Model for Electrically Evoked​​​‌ Contractions.IEEE Transactions‌ on Medical Robotics and‌​‌ BionicsJuly 2025HAL​​DOIback to text​​​‌
• 14 articleG.Gabriel‌ Graffagnino, B.Benoît‌​‌ Sijobert, K.Karine​​ Patte, C.Christine​​​‌ Azevedo-Coste and D.David‌ Gasq. Importance of‌​‌ analyzing spasticity and co-activation​​​‌ as complementary biomarkers of​ gait in children with​‌ cerebral palsy.Clinical​​ Biomechanics132February 2026​​​‌, 106743HALDOI​back to text
• 15​‌ articleV.Valentin Maggioni​​, C.Christine Azevedo-Coste​​​‌, S.Sam Durand​ and F.François Bailly​‌. Optimisation and Comparison​​ of Markerless and Marker-Based​​​‌ Motion Capture Methods for​ Hand and Finger Movement​‌ Analysis.Sensors25​​42025, 1079​​​‌HALDOIback to​ text
• 16 articleV.​‌Valentin Maggioni, P.​​Pierre Schegg, T.​​​‌Thomas Guiho, C.​Christine Azevedo-Coste and F.​‌François Bailly. Optimal​​ estimation of forearm muscle​​​‌ activations evoked by implanted​ electrical stimulation in complete​‌ quadriplegia.Multidisciplinary Biomechanics​​ Journal50th congress of​​​‌ the...October 2025HAL​DOIback to text​‌
• 17 articleS.Sabrina​​ Otmani, A.Andrew​​​‌ Murray, C.Christine​ Azevedo Coste and F.​‌François Bailly. Data-driven​​ biomechanical optimisation of bicycle​​​‌ and crank-pedal design for​ maximizing power output.​‌Computer Methods in Biomechanics​​ and Biomedical Engineering2025​​​‌HALDOIback to​ text
• 18 articleS.​‌Sabrina Otmani, A.​​Andrew Murray, C.​​​‌Christine Azevedo Coste and​ F.François Bailly.​‌ Maximizing cycling efficiency: innovative​​ bicycle drive mechanisms tailored​​​‌ to individual muscular capacities​.Computer Methods in​‌ Biomechanics and Biomedical Engineering​​March 2025, 13​​​‌HALDOIback to​ text
• 19 articleC.​‌Clotilde Turpin, O.​​Olivier Rossel, F.​​​‌Félix Schlosser-Perrin, M.​Mathilde Carrière, R.​‌Riki Matsumoto, E.​​Emmanuel Mandonnet, H.​​​‌Hugues Duffau, S.​Sam Ng and F.​‌François Bonnetblanc. Gold​​ standard for estimation of​​​‌ propagation velocity from axono-​ or cortico-cortical evoked potentials?​‌ A case study.​​Clinical Neurophysiology179November​​​‌ 2025, 2111370HAL​DOIback to text​‌
• 20 articleC.Clotilde​​ Turpin, O.Olivier​​​‌ Rossel, F.Félix​ Schlosser-Perrin, R.Riki​‌ Matsumoto, E.Emmanuel​​ Mandonnet, S.Sam​​​‌ Ng, H.Hugues​ Duffau and F.François​‌ Bonnetblanc. Influence of​​ myelo-architectures on direct cortical​​​‌ response evoked by electrical​ stimulation.Clinical Neurophysiology​‌December 2025, 2111488​​HALDOIback to​​​‌ text

National journals

• 21​ articleC.Charlotte Le​‌ Goff, P.Pauline​​ Coignard, C.Christine​​​‌ Azevedo-Coste, F.Franck​ Geffard and C.Charles​‌ Fattal. Design and​​ use of devices to​​​‌ assist movement of the​ upper limb: review of​‌ the literature.Pratiques​​ en Médecine Physique et​​​‌ de Réadaptation (Bi-annuel)04​March 2025, 58-63​‌HALback to text​​

International peer-reviewed conferences

• 22​​​‌ inproceedingsC.Christine Azevedo​ Coste, T.Thomas​‌ Guiho, F.Fernanda​​ Ferreira, F.François​​​‌ Bailly, B.Benjamin​ Degeorge, A.Antoine​‌ Geffrier, V.Valentin​​ Maggioni, J.Jonathan​​​‌ Baum, D.David​ Andreu, J.Jacques​‌ Teissier, C.Charles​​ Fattal and D.David​​​‌ Guiraud. Epineural Electrical​ Stimulation for Grasp Recovery:​‌ 28-Day Study in 4​​ Tetraplegic Participants.FES​​​‌ 2025 - 15th Vienna​ Workshop on Functional Electrical​‌ Stimulation & 28th Annual​​ Conference of the International​​ Functional Electrical Stimulation Society​​​‌Vienne, AustriaSeptember 2025‌HALback to text‌​‌
• 23 inproceedingsT.Tiago​​ Coelho-Magalhães, C.Christine​​​‌ Azevedo Coste and F.‌François Bailly. Optimal‌​‌ Control Framework for Personalized​​ FES-cycling in Individuals with​​​‌ Spinal Cord Injury.‌FESWS 2025 - 15th‌​‌ Vienna International Workshop on​​ Functional Electrical Stimulation &​​​‌ 30 years IFESS Anniversary‌Vienna (AUSTRIA), AustriaSeptember‌​‌ 2025HALback to​​ text

Conferences without proceedings​​​‌

• 24 inproceedingsJ.Jonathan‌ Baum, C.-N.Chamot-Nonin‌​‌ Manon, O.Oppelt​​ Vera, D.David​​​‌ Guiraud and T.Thomas‌ Guiho. Towards Electrophysiological‌​‌ and Histological Mapping of​​ Upper Limb Nerves in​​​‌ Pigs Using Epineural Stimulation‌.FESWS 2025 -‌​‌ 15th Vienna International Workshop​​ on Functional Electrical Stimulation​​​‌ & 30 years IFESS‌ AnniversaryVienna, AustriaSeptember‌​‌ 2025HALback to​​ text
• 25 inproceedingsT.​​​‌Tiago Coelho-Magalhães, C.‌Christine Azevedo-Coste and F.‌​‌François Bailly. Optimized​​ Stimulation Patterns for FES-stimulated​​​‌ Cycling Using an Experimentally‌ Identified Physiological Muscle Model‌​‌.28th Annual Conference​​ of the International Functional​​​‌ Electrical Stimulation Society -‌ Rehabweek 2025Chicago, United‌​‌ StatesMay 2025HAL​​back to text
• 26​​​‌ inproceedingsG.Gabriel Graffagnino‌, D.David Gasq‌​‌, K.Karine Patte​​, B.Benoît Sijobert​​​‌ and C.Christine Azevedo‌ Coste. Real-time gait‌​‌ event detection using motion​​ capture to control an​​​‌ electrical stimulator: Proof-of-concept.‌FESWS 2025 - 15th‌​‌ Vienna International Workshop on​​ Functional Electrical Stimulation &​​​‌ 30 years IFESS Anniversary‌Vienne (Austria), AustriaSeptember‌​‌ 2025HALback to​​ text
• 27 inproceedingsP.​​​‌Pierre Schegg and F.‌François Bailly. Data‌​‌ efficient muscle parameter estimation:​​ Application to the Human​​​‌ Upper Limb.20th‌ International Symposium on Computer‌​‌ Methods in Biomechanics and​​ Biomedical EngineeringBarcelona, Spain​​​‌September 2025HALback‌ to text

Scientific book‌​‌ chapters

• 28 inbookC.​​Charles Fattal, F.​​​‌Fernanda Ferreira, D.‌David Guiraud and C.‌​‌Christine Azevedo Coste.​​ Neuroprothèses de stimulation motrice​​​‌ du membre supérieur tétraplégique‌.Pratiques en médecine‌​‌ Physique et de réadaptation​​March 2025HALback​​​‌ to text
• 29 inbook‌G.Gabriel Graffagnino,‌​‌ C.Christine Azevedo Coste​​, D.David Gasq​​​‌, B.Benoît Sijobert‌ and K.Karine Patte‌​‌. Usages de la​​ stimulation électrique dans le​​​‌ cadre de la rééducation‌ motrice fonctionnelle chez l'enfant‌​‌.Handicap neurologique central​​ de l'enfant : améliorer​​​‌ la fonction motrice des‌ membres inférieursAcquisitions en‌​‌ médecine physique et de​​ réadaptationMarch 2025,​​​‌ 128-136HALback to‌ text

Reports & preprints‌​‌

• 30 miscK.Kevin​​ Co, M.Mickaël​​​‌ Begon, F.François‌ Bailly and F.Florent‌​‌ Moissenet. Optimal control​​ driven functional electrical stimulation:​​​‌ A scoping review.‌2025HAL

Other scientific‌​‌ publications

• 31 inproceedingsS.​​Siddhiraj Banjac, C.​​​‌Charles Fattal, D.‌Denis Sablot, N.‌​‌Nadège Olivier, C.​​Catherine Leblond, J.​​​‌Julia Schmidt, C.‌Coraline Lethimonnier, C.‌​‌Carole Plantard, M.​​Marianne Vaugoyeau, S.​​​‌Sara Rivas-Lamello and A.‌Alice Guyon. Protocol‌​‌ for a randomized controlled​​​‌ trial to evaluate the​ efficacy of a meditative​‌ relaxation practice on quality​​ of life and stress​​​‌ reduction in multiple sclerosis​ patients.NeuroFrance 2025​‌ - 17th International Meeting​​ of the French Neuroscience​​​‌ SocietyMontpellier, FranceMay​ 2025HAL
• 32 inproceedings​‌S.Siddhiraj Banjac,​​ C.Charles Fattal,​​​‌ D.Denis Sablot,​ N.Nadège Olivier,​‌ C.Catherine Leblond,​​ J.Julia Schmidt,​​​‌ C.Coraline Lethimonnier,​ C.Carole Plantard,​‌ M.Marianne Vaugoyeau,​​ S.Sara Rivas-Lamello and​​​‌ A.Alice Guyon.​ Randomized controlled trial to​‌ evaluate the efficacy of​​ a meditative relaxation practice​​​‌ on quality of life​ and stress reduction in​‌ multiple sclerosis patients.​​CONFERENCE ON MULTIPLE SCLEROSIS​​​‌Paris, FranceJune 2025​HAL
• 33 inproceedingsJ.​‌Jonathan Baum, V.​​Valentin Maggioni, L.​​​‌Lucie William, F.​François Bailly, T.​‌Thomas Guiho, C.​​Charles Fattal, D.​​​‌David Guiraud and C.​Christine Azevedo Coste.​‌ Analyzing the responses evoked​​ by multi-contact epineural electrical​​​‌ stimulation to restore upper-limb​ functions in complete tetraplegia​‌.REHABWEEK 2025Chicago​​ (Illinois), United StatesMay​​​‌ 2025HAL
• 34 inproceedings​T.Tiago Coelho-Magalhães,​‌ C.Christine Azevedo-Coste and​​ F.François Bailly.​​​‌ Optimized Stimulation Patterns for​ FES-stimulated Cycling Using an​‌ Experimentally Identified Physiological Muscle​​ Model.28th Annual​​​‌ Conference of the International​ Functional Electrical Stimulation Society​‌ - Rehabweek 2025Chicago,​​ United StatesMay 2025​​​‌HAL
• 35 inproceedingsN.​Nicolas Huchet, C.​‌Côme Butin, H.​​Henri Boutard, C.​​​‌Christophe Braillon, Y.​Yann Corbel, S.​‌Sébastien Cariou, C.​​Charlie Dreano, C.​​​‌Christian Fromentin, J.​Jean Forest, A.​‌Arthur van Haaren,​​ R.Roger Pissard-Gibollet and​​​‌ C.Christine Azevedo Coste​. BIONICOHAND an opensource​‌ and collaborative myoelectric hand​​ prothesis.IFESS 2025​​​‌ - 28th Annual Conference​ of the International Functional​‌ Electrical Stimulation SocietyChicago,​​ United States2025HAL​​​‌
• 36 inproceedingsP.Pierre​ Schegg and F.François​‌ Bailly. Data efficient​​ muscle parameter estimation: Application​​​‌ to the Human Upper​ Limb.20th International​‌ Symposium on Computer Methods​​ in Biomechanics and Biomedical​​​‌ EngineeringBarcelona (ES), Spain​September 2025HAL