Could AI help the brain to learn again after a stroke?

Long-term consequences of CVA

Strokes are the leading cause of disability occurring in adults. In 80% of cases, the cause is a clot formation in the blood vessels of the brain. The blocked vessels result in neuronal cell death, known as apoptosis. Each minute that passes results in the loss of 2 million brain neurons. So it is crucial to act quickly. 

Unfortunately, treatment in the acute stage does not prevent long-term functional impact, particularly in terms of arm and hand movement capacity, when the motor cortex has been damaged”, says Philippe Ciuciu, director of research at CEA and head of the MIND project team. “Rehabilitation through physiotherapy can reduce these motor function shortcomings, but doesn’t always eliminate them.”

BrainSync: an enterprising neural prosthesis project

Using AI to help the brain to learn again after a stroke is the ambition of the BrainSync project, which has joined forces with the joint MIND project team of Inria and CEA (NeuroSpin centre). What’s the aim? To develop an innovative method for upper limb function rehabilitation using neural prostheses. 

A large-scale initiative was launched at the end of 2024 and will last four and a half years, until 2029, with a budget of €5 million. This is one of the pivotal projects selected by the CEA ‘Audace’ initiative in response to the Government’s call to develop high-risk research projects for the ‘France 2030’ investment plan. 

In all, the project involves around thirty partners from nine research teams from Inria, CEA (Saclay and Grenoble), Paris-Saclay University (NeuroSpin), Grenoble-Alpes University, and several healthcare institutions (University Hospital of Ile-de-France AP-HP, Paris-Neurosciences University Hospital Group and the University Hospitals of Grenoble-Alpes and Saint-Etienne).

Combining neuroscience and artificial intelligence

At the core of this project lies Wimagine® technology, a neuroprosthesis developed by CEA-Leti to restore mobility to people suffering from motor disability caused by spinal cord injury. Unlike existing systems requiring sensors to be implanted in the brain tissue, this 4 cm² implant is placed on the surface of the cortex, like “sticking a plaster on the skull”, says Philippe Ciuciu. 

This technology allows us to measure electrical activity in the part of the brain that controls motor function (electrocorticography), thanks to 64 electrodes that are in contact with the dura mater, the fibrous membrane covering the brain. It presents several advantages, including long-term signal stability and virtually instantaneous synchronisation (less than 50 milliseconds from electrical signal decoding to motor action).

Associating electrical activity with intent of movement

“The difficulty here is that, in stroke victims, the cortex is damaged”, Philippe Ciuciu points out. 

"We don’t know at this stage whether the signals will be exploitable.” A major innovation for stroke victims.

But to reach that stage, the team must first determine the best site to place the Wimagine® implants on the cortex. To do this, from October 2025 they are to recruit 100 stroke patients in their recovery phase. They will take part in a non-invasive clinical trial based on multimodal neuroimaging at 7 Tesla (functional MRI - fMRI - to map active brain areas linked to movement intention, and structural MRI to locate damaged myelin bundles in the brain). 

“Bertrand Thirion, Demian Wassermann, Chaithya GR and myself from the MIND team are all involved in this study”, says Philippe Ciuciu. We are going to analyse the collected brain imagery data and then train various artificial intelligence predictive models, ranging from machine learning to deep learning, and including transformer and even foundation models based on large language models (LLM), so that they can link the functional signal recorded in fMRI to the intention of movement.”

Building robust AI models

In this context, Inria’s expertise is a valuable asset: it will allow us to build robust and explainable AI models with the aim of determining the best place to implant the Wimagine® neuroprostheses. The CEA-Leti researchers will then take over, to convert the brain signals measured by electrocorticography into motor commands in real time, using a Bluetooth protocol and real-time decoding algorithms, to enable the patient to activate an articulated glove or a cuff that facilitates gripping. This will allow the patient to grasp objects despite the hemiplegia. 

“We’re not there yet, but we intend to explore the possibility of restoring neuronal plasticity with the Wimagine® neuroprosthesis”, announces Philippe Ciuciu. “In other words, we hope to allow stroke victims, following a few months of rehabilitation assisted by the system, to partially regain motor function, even after they stop using it.”

This approach is part of a broader movement to combine neuroscience and artificial intelligence. The AI aspect helps to restore motor function to the brain, while understanding the brain makes it possible to build new bio-inspired AI models. A win-win combination.

BrainSync: multiple fields of expertise

This multidisciplinary project involves a total of 30 neurologists, neurosurgeons, neuroradiologists, rehabilitation doctors, applied mathematicians, artificial intelligence specialists and MRI physicians. 

Besides Philippe Ciuciu and his MIND team (CEA/Inria), BrainSync is also overseen by Myriam Edjlali-Goujon, a university professor and AP-HP hospital practitioner, Abdelmadjid Hihi, a researcher at CEA-Leti, Marina Reyboz, director of research at CEA-List and Mikaël Mazighi, director of the Neurovasc University Hospital Federation.

Another asset: an outstanding technological environment. The project benefits in particular from two new-generation 7 Tesla imagers and the Iseult MRI, the most powerful in the world with a magnetic field of 11.7 Tesla. This extraordinary equipment will make it possible to produce images of the human brain with unequalled precision. 

Researchers are to deal with another aspect, again under the BrainSync project. They are going to look at the joint or differentiated representation in the brain of the learning and decision-making functions, through a task of exploration/exploitation in a changing environment that reflects the reality of day-to-day life. They will aim to gauge whether humans are likely to demonstrate mental flexibility in the face of change. To do this, they are to develop new functional imaging with high spatio-temporal resolution on the Iseult MRI, in order to assess not only the mappings of these two functions - learning and decision-making - but also their temporal response profile, a challenge in functional MRI.

Find out more 

• Des chercheurs tentent de concevoir une neuroprothèse capable de détecter une "intention de mouvement" pour les victimes d’AVC (podcast), France Info, 10/3/2025.
• France 2030 : Kick-off de BrainSync, le projet structurant porté par Philippe Ciuciu (NeuroSpin) et Myriam Edjlali (APHP), Institut des Sciences du Vivant Frédéric Joliot (CEA), 29/11/2024.
• À la recherche de la perfection : l'imagerie cérébrale dans toute sa splendeur, plenary conference by Philippe Ciuciu (video), GRETSI, 14/10/2023.
• En direct de NeuroSpin à Paris-Saclay, dans et avec les cerveaux de l'IA (podcast), France Inter, 12/4/2023.