Thèse Identification des Signatures Neuronales des Capacités Cognitives Infracliniques à la Phase Aigue du Coma Unlock H/F - Doctorat.Gouv.Fr

Établissement : Université de Toulouse École doctorale : CLESCO - Comportement, Langage, Éducation, Socialisation, Cognition Laboratoire de recherche : ToNIC-Toulouse NeuroImaging Center (UMR 1214) Direction de la thèse : Stein SILVA ORCID 0000000196907413 Début de la thèse : 2027-09-01 Date limite de candidature : 2026-11-16T23:59:59 UNLOCK vise à transformer la gestion des troubles de la conscience (DoC) grâce à une neurotechnologie entièrement au chevet du patient. Son objectif est de détecter et de suivre longitudinalement la cognition cachée préservée - du traitement sensoriel élémentaire à la compréhension de la parole - tout en améliorant le diagnostic, le pronostic et la personnalisation thérapeutique. Une ambition centrale est d'identifier la conscience cachée, c'est-à-dire la dissociation motrice cognitive ou DCM, et de rétablir la communication entre les patients, les familles et les équipes cliniques à l'aide d'interfaces cerveau-ordinateur (ICM) personnalisées couplées à une IA générative.
Le projet intégrera des données multimodales continues acquises auprès du patient capturant l'activité cérébrale, la dynamique physiologique et les interactions cerveau-corps, en utilisant des modulations sensorielles, auditives et tactiles multisensorielles dans les stimuli naturalistes, y compris la parole et la vidéo. L'apprentissage par renforcement en boucle fermée adaptera ensuite ces stimuli au profil neuronal de chaque patient. Ce cadre permettra un décodage en temps réel à un seul essai à plusieurs niveaux de la hiérarchie cognitive, de la perception à l'intégration de haut niveau et au traitement du langage.
UNLOCK combinera des ICM multimodaux, une IRM cérébrale mobile à faible champ assistée par l'IA (0,064 T), des algorithmes de décodage cérébral individualisés et une IA explicable pour la fusion de données multimodales. Ces technologies soutiendront le phénotypage cognitif dynamique, la surveillance continue et la prédiction de la réponse au traitement. Lorsque des réponses neuronales fiables sont détectées, l'ICM s'interface avec de larges modèles de langage (LLM) pour faciliter la communication patient-soignant et réduire l'isolement profond associé à la DCM.
En unifiant les neurosciences, la neuroimagerie, l'apprentissage automatique et la médecine de soins intensifs, le projet UNLOCK va au-delà de l'évaluation comportementale statique vers un cadre adaptatif de médecine de précision pour le coma et le DoC. Le projet couvre la trajectoire aiguë à chronique des lésions cérébrales et peut redéfinir la façon dont la conscience est détectée, surveillée et engagée thérapeutiquement.
L'initiative UNLOCK a reçu le prix INSERM Neurotechnologies BOOSTER 2025 (S Silva), reconnaissant son originalité et son potentiel de transformation.
A coma is a brain state induced by acute, severe brain injury, marked by unarousable unresponsiveness.'1 Survival rates among coma patients at hospital discharge are extremely low, with most deaths resulting from the withdrawal of life-sustaining treatments, based on assumptions of poor neurological outcomes. However, recent longitudinal studies and neural marker identification reveal that many patients are misdiagnosed, with a significant portion potentially capable of survival1. Even more critically, among these patients with disorders of consciousness (DoC), about 25% of patients appear to display Cognitive Motor Dissociation (CMD)2,3, exhibiting hidden awareness only detectable through brain activity. All these findings underscore the potential for survival and unrecognized communication capabilities in coma patients during the acute phase, highlighting the urgent need for improved diagnostic and monitoring tools. Standard behavioral assessments-crucial for medical decision-making-are hindered by fragmented, isolated objective and subjective evaluations, further complicated by sensory and motor impairments and residual sedation. Currently, coma patients are presumed unable to communicate, leaving them entirely dependent on caregivers. This fosters a culture of therapeutic nihilism, stalling progress in diagnosis and treatment, while imposing significant emotional and logistical burdens on families and healthcare systems.
Neurotechnologies hold immense promise to revolutionize healthcare and dramatically enhance patient quality of life. Recent BCI research using noninvasive solutions revealed hidden cognition in DoC patients by decoding signals linked to perception, attention, and mental imagery5, even enabling yes/no communication. Yet, BCIs face major hurdles6. Pre-implementation data-such as the structural integrity of language-processing or arousal brain networks -remain limited. Neural signals in DoC patients can be weak or inconsistent, risking underestimation of their preserved cognitive abilities. A recent large-scale study2 underscores this challenge: only 25% of patients with confirmed awareness were correctly detected by BCI, highlighting the urgent need for more robust and reliable solutions. BCIs also demand individualized calibration to train AI algorithms, a process that can be too demanding for severely compromised patients. Due to these unresolved methodological barriers, most research has been limited to single-session studies using one modality (e.g., visual, auditory, or motor) or basic feasibility trials, rather than longitudinal approaches. This raises critical ethical concerns: offering hope to patients only to withdraw it after testing leaves them without a sustainable solution. Moreover, current BCI paradigms rely on simplistic, highly controlled stimuli to generate clear, interpretable signals for assessment and diagnosis. Communication modes are often static, confined to binary (yes/no) responses-an approach with limited real-world applicability. To capture the complexity of human cognition and assess higher-level functions, there is a need to move beyond these constrained paradigms by incorporating more sophisticated stimuli (e.g., speech, videos). Likewise, evolving toward more advanced communication modes will unlock new assistive technologies, better aligning with patients' needs.
BCI paradigms will benefit from integrating peripheral bodily signals, specifically by targeting cardiac and respiratory phases that optimize stimulus processing. In patients with emerging consciousness, this novel approach will enable volitional control over stimulus delivery through respiratory modulation, thereby improving overall intervention acceptance. The UNLOCK project sets out to revolutionize the management of patients with DoC and related disorders by delivering an innovative, bedside, and cost-effective neurotechnology approach. This groundbreaking solution enables real-time, longitudinal assessment of physiological and cognitive abilities-from basic perception to advanced speech comprehension. By enhancing diagnostic and prognostic precision, UNLOCK bridges critical gaps in current practices, aiming to restore communication-via Generative-AI based Neurotechnology -between patients, caregivers, and families, thereby alleviating the isolation experienced by individuals with covert consciousness. Moreover, it integrates personalized therapies to accelerate recovery and promote cognitive engagement. The UNLOCK will run for 36 months, with tasks grouped in 4 research and 1 structural Work Packages (WP). Data collection will commence on September 2026, upon completion of required ethics and reglementary approvals.
WP1. Project management and coordination (Co-coord. S Silva; Partner ToNIC; M De Lucia Brain-Body and Consciousness). Aims. 1. To monitor and control the overall project progress and ensure that deliverables and milestones are achieved in time. 2. To manage communication and coordinate the work between the WP (partners, Health Impact organization, stakeholders). 3. To ensure project reporting and monitoring strictly following Health Impact and INSERM requirements. An international SAB will provide support (L Naccache, N Shiff, A Thibaut). Description. Task 1.1. Project management. Task 1.2. Organizing project Kick-off meeting. Task 1.3. Risk management. Task 1.4. Relation with Health Impact. Deliverables. Project startup meeting (M2); Periodic progress reports (M12, 24, 36); Public reports on projects (M12, 24, 36).
WP2. Real-time neurocognitive, somatosensory and neuromuscular naturalistic assessments for coma prognosis and BCI-readiness for communication (Coord. F Dehais, Partner ISAE-SUPAERO). Aims. 1. First time in coma deep phenotyping based on continuous multimodal data collection. 2. To identify a bedside and scalable roadmap for detecting CMD and assessing prerequisites for effective BCI solution to restore communication. Description. Prospective cross-sectional proof of concept (POC) (ancillary from funded PHRC-N ARISE): n = 50 patients at the acute phase of coma with 6 months follow up. Task 2.1. Continuous baseline EEG monitoring using low-cost, scalable dry electrodes devices. Task 2.2. Continuous neuromuscular assessment using wearable EMG, 3D-accelerometry and AI-empowered body movement recognition. Task 2.3. Unprecedented real-time, single-trial decoding across the full spectrum of cognitive processes using multisensory StAR and advanced closed-loop reinforcement learning. Task 2.4. Development of DCM Alert System based on baseline (arousal) EEG, multisensory StAR round the clock and portable brain MRI findings (cf. WP2). Deliverables. Structured research database of clinical, neurocognitive and neuromuscular biomarkers data (EBRAIN) (M24). Report on the acceptability of recording devices (M24). Identification of multimodal phenotype of patients with CMD (M30). Identification of multimodal signatures of BCI-readiness for communication (M30). CMD Alert system set-up (M36).
WP3. Portable MRI brain imaging standards and analysis. (Coord. S Silva; Partner ToNIC). Built upon AI-powered portable 0.064T MRI & workflow for image transfer, storage and evaluation from CUBE ANR project (P.I. S Silva). Aims. 1. Set up and provide an image core lab with continuous evaluation of all images acquired within the study of WP1 (POC study cohort from ARISE project, n = 50) 2. Assess structural integrity of key networks for BCI-readiness (arousal, language, vision). Description. Task 3.1. Create and maintain an image database. Task 3.2. Monitor image quality in the clinical trial. Task 3.3. Identify and classify structural prerequisites for BCI-based communication. Deliverables. Identification of structural signatures related to CMD readiness for BCI-based communication (M30). Final core image lab report (M36).
WP4. Brain-Body signatures (Coord. M De Lucia) Systematic evaluation of the impact of cardiac and respiratory rhythms and their coupling on sensory processing in acute coma. Identify phases and latency of bodily rhythms optimizing sensory processing across modalities including auditory and tactile stimuli. Leveraging these results for improved brain computer interface paradigms integrating bodily stimuli in patients with emerging consciousness.
WP5. Data integration and development of PoC algorithms. (Coord. S Achard, Partner CNRS-UGA-INRIA). Aim. 1. Define and implement data collecting procedures to produce pseudonymous datasets. 2. Organize the pooling of the data to a common cloud-based data lake with appropriate architecture, services for visualization and processing. 3. Perform unimodal and multimodal analysis to predict coma outcome. Description. Task 5.1. Data collection. Task 5.2. Data storage. Task 5.3. Data processing. Sparsity-promoting machine learning models such as optimization strategies with sparsity constraints and graph signal processing. Confounding variables will be taken into account carefully by pre-processing data (ISOMAP). Task 5.4. Ethical implications (POC testing devices into healthcare pathways). Deliverables. A statistical description and analysis of the complete cohort (M36). A report describing predictive models from multimodal data (M36). A dataset of physiological and clinical data that will be made publicly accessible on-demand (M30). A paper on ethical implications of the instruction of PoC testing in healthcare pathways (M36).

Le projet conviendra à un ou une étudiant/étudiante ayant une formation en bioingénierie, en sciences de la vie, en neurosciences ou dans un domaine connexe. Le ou la candidat.e idéal.e aura de l'expérience dans l'acquisition et l'analyse de l'EEG, un vif intérêt pour le développement méthodologique et un enthousiasme pour les données neurophysiologiques multimodales, les interactions cerveau-corps et les interfaces cerveau-machine. Des compétences en traitement du signal, en programmation et/ou en apprentissage automatique seraient précieuses, tout comme la capacité à travailler en collaboration dans un environnement de recherche clinique interdisciplinaire. Une bonne connaissance de l'anglais est requise. La connaissance du français n'est pas obligatoire.

• Gestion des données
• Electro-encéphalographie (EEG)
• Programmation
• Anglais
• Traitement du signal
• Machine learning
• Processing