HAL-HCL
Not a member yet
19000 research outputs found
Sort by
Contralateral R1 response in blink reflex in patients with amyotrophic lateral sclerosis
No full textInternational audienceObjective: This study aimed to compare the frequency of blink reflex's contralateral R1 responses (R1') between patients with amyotrophic lateral sclerosis (ALS), non-ALS motor deficit patients, and healthy volunteers.Methods: A total of 120 participants were prospectively recruited: 40 with ALS, 40 with a non-ALS motor deficit, and 40 healthy volunteers. Blink reflexes were recorded from orbicularis oculi muscles following supraorbital nerve stimulation.Results: R1' was more frequent in the ALS group (42.5 %) compared to healthy volunteers (12.5 %, p = 0.00588), and compared to non-ALS patients (7.5 %, p = 0.000789). Bilateral R1' was observed only in ALS patients (22.5 %). No clinically significant difference was found in the latencies or amplitudes of the R1, R2, or R1' responses among groups. R1' was more frequent in ALS patients with pseudobulbar affect (71.4 %) compared to those without (36.4 %).Conclusions: The higher frequency of R1' in ALS highlights its potential role in distinguishing ALS from other motor disorders. Its sensitivity was low, but bilateral R1' was specific to ALS. The higher frequency of R1' among ALS patients with pseudobulbar affect potentially reflects corticobulbar neuron degeneration.Significance: The R1', especially when bilateral, could serve as an additional diagnostic biomarker for ALS, although its clinical relevance should be considered within the broader diagnostic context
Défis et avancées en apprentissage profond pour la segmentation de lésions d'AVC sur scanner CT : exploitation des caractéristiques images
No full textIschemic stroke, affecting 12.2 million people worldwide, is caused by a clot blocking blood flow to a part of the brain, leading to tissue necrosis. Two imaging modalities are used for its management: Magnetic Resonance Imaging (MRI) (expensive, requiring specific equipment and long acquisition times) and X-ray computed tomography (CT scan) (cheaper, faster, but offering lower-quality images with less contrast for stroke lesions). This thesis aims to automatically segment stroke lesions from CT scan images, a modality often overlooked due to its complexity compared to MRI. Two main objectives are addressed: (i) identifying the challenges and limitations of stroke lesion segmentation on CT scans, and (ii) improving the performance of automatic segmentation algorithms, focusing on lesion-to-healthy tissue contrast, evaluated using the Fisher ratio. A dataset of 108 patients is used, including an MRI at inclusion, a CT scan 24 hours later, and an MRI six days later. A reference mask of the final lesion is available. An initial analysis of state-of-the-art methods (U-Net, cGAN, Mask R-CNN, UNETR) shows that U-Net is the most effective when considering both performance and computational cost. As initial results suggested a performance plateau due to contrast limitations, a method was developed to establish a threshold below which images provide no relevant information for training. By excluding images with a Fisher ratio <0.05 (60% of the data), the models perform as well as when trained on the entire dataset. To overcome the limitations of neural networks, we explored new training approaches, notably curriculum learning, which progressively introduces training data based on a difficulty metric. In addition to contrast, lesion size and a metric derived from the loss function were tested as difficulty metrics, improving segmentation performance. To better understand the mechanisms of neural networks, we developed a method to represent the features extracted by the models based on neuron activation strength. This approach revealed disparities between the two imaging modalities and fostered discussions with medical staff on methods to guide the extraction of specific features. Finally, we designed a pretext task for self-supervised learning focused on segmenting anatomical regions related to stroke lesions defined by the ASPECT score. Compared to tasks like image rotation classification and inpainting, this method shows promising results for segmentation, particularly when fewer annotated data are available. This work deepens the understanding of the challenges of stroke lesion segmentation on CT scans and explores improvement avenues, paving the way for numerous future developments.L'accident vasculaire cérébral ischémique (AVC), touchant 12,2 millions de personnes dans le monde, est causé par un caillot bloquant la circulation sanguine dans une partie du cerveau, entraînant une nécrose des tissus. Deux modalités d'imagerie sont utilisées pour sa prise en charge : l’Imagerie par Résonance Magnétique (IRM) (coûteuse, avec un équipement spécifique et un temps d'acquisition long) et le scanner à rayons X (moins cher, plus rapide, mais offrant des images de moindre qualité, avec des lésions d’AVC moins contrastées). Cette thèse vise à segmenter automatiquement les lésions d'AVC à partir d'images de scanner, une modalité souvent négligée en raison de sa complexité par rapport à l'IRM. Deux axes sont abordés : (i) identifier les limites et défis de la segmentation des lésions d’AVC sur scanner, (ii) améliorer les performances des algorithmes automatiques, en se concentrant autour du contraste lésion-tissu sain, évalué via le ratio de Fisher. Un jeu de données comprenant 108 patient·es est utilisé, avec une IRM à l’inclusion, un scanner à 24 heures, et une IRM à 6 jours. Un masque de référence des lésions finales est disponible. Une première analyse des méthodes de l’état de l’art (U-Net, cGAN, Mask R-CNN, UNETR) montre que le U-Net est plus efficace, performances et coût computationnel combinés. Les premiers résultats suggérant un plafonnement des performances lié au contraste, une méthode a été développée pour définir un seuil en dessous duquel les images n’apportent pas d’information pertinente. En excluant les images avec un ratio de Fisher <0.05 (60% des données), les modèles performent aussi bien qu’avec l’ensemble des données. Pour dépasser les limites des réseaux de neurones, nous avons exploré des approches d’entraînement, notamment l’apprentissage curriculaire, qui introduit les données progressivement selon une métrique de difficulté. En plus du contraste, la taille des lésions et une métrique issue de la fonction de coût ont été testées comme métrique de difficulté, améliorant les performances de segmentation. Pour mieux comprendre les mécanismes des réseaux neuronaux, nous avons développé une méthode de représentation des caractéristiques extraites par les modèles, basée sur la force d'activation des neurones. Cela a permis de révéler les disparités entre les deux modalités d'imagerie et d’engager des discussions avec le personnel médical sur des méthodes pour guider l’extraction de caractéristiques spécifiques. Enfin, nous avons conçu une tâche prétexte pour l'apprentissage auto-supervisé, axée sur la segmentation de zones anatomiques liées aux lésions d'AVC définies par le score ASPECT. Comparée à une tâche de classification par rotation et remplissage d’images, cette méthode montre des résultats prometteurs pour la segmentation, surtout lorsque moins de données annotées sont disponibles. Ce travail approfondit la compréhension des difficultés liées à la segmentation des lésions d'AVC sur scanner et explore des pistes d’amélioration, ouvrant de nombreuses perspectives pour l’avenir
Brain dynamics of the interplay between auditory selective attention and working memory during melody encoding
No full textInternational audienceWorking memory and attention are jointly needed in most everyday life tasks and activities. They have however mostly been studied separately. Here we investigate how auditory working memory – in a delayed-matching-to-sample task – and selective attention interact using a recently introduced paradigm (MEMAT, for MEMory and ATtention) and MEG (Magneto-encephalographic) recordings in twenty-two participants. We manipulate the difficulty of the memory task and attentional filtering. When memory task difficulty increases, accuracy decreases, CNV (Contingent Negative Variation) amplitude in anticipation of the melody to encode increases, and the decrease in alpha power during encoding and maintenance in a left fronto-temporal network is reduced. When attentional filtering difficulty increases, accuracy decreases, the amplitude of the sustained evoked response during encoding increases, whereas the differential processing of relevant and irrelevant sounds in auditory areas is less pronounced,and frontal theta power during encoding and maintenance is higher. In the left auditory cortex, we directly observe the result of the interaction between auditory memory and attention: the facilitation of relevant sound processing in the easy filtering condition reduces when the memory task difficulty increases. This pattern mirrors the observed behavioral effects: a smaller difference in accuracy between attention conditions when the memory task difficulty increases. Overall brain dynamics highlight reciprocal influences of working memory and selective attention processes, in keeping with shared cognitive resources between them
Functionalization of chitosan with a polycarboxylic macrocycle yields injectable hydrogel with pH and salts responsiveness
No full textInternational audienceThe development of injectable hydrogels that respond to physiological stimuli represents a promising strategy for a range of biomedical applications, although the precise tuning of gelation kinetics, mechanical stability, and biocompatibility remains a significant challenge. This study presents a pH- and osmolarity-responsive injectable hydrogel, formulated from a combination of chitosan and chitosan functionalized with a macrocyclic polycarboxylate. The functionalization of chitosan significantly modifies the electrostatic charges along the polymer backbone, enabling fast gelation under physiological conditions. The gelation process is driven by pH neutralization and osmolarity increase, where electrostatic interactions between the zwitterionic chitosan and unmodified chitosan generate a dynamic, entangled network strengthened by both electrostatic crosslinking and hydrophobic interactions. Rheological and structural analyses reveal the possibility of fine-tuning the gelation kinetics and mechanical properties by altering the ratio of zwitterionic chitosan to conventional chitosan in the formulation. Here, the optimized 67:33 % ratio achieved favorable compromise between rapid gelation and stability in physiological media. Different microscopy experiments, including conventional scanning electronic microscopy and live imaging, have confirmed the porous architecture of the hydrogels. In vivo experiments confirmed the injectability, biocompatibility, and biodegradability of the hydrogels, with gradual degradation observed by magnetic resonance imaging and fluorescence imaging over time in healthy mice after subcutaneous administration. Additionally, preclinical safety assessments in rabbits demonstrated good local tolerance to both single and repeated subcutaneous injections, with no systemic toxicity observed. These findings support the potential for broad biomedical future applications, including drug delivery, wound healing, local metal uptake and tissue regeneration
High-dose chemotherapy with autologous haematopoietic stem cell transplantation in patients with isolated vitreoretinal lymphoma: a LOC network study
No full textInternational audienceDespite its indolent evolution, vitreoretinal lymphoma (VRL) has a poor prognosis due to a major risk of relapse in the central nervous system (CNS) and may necessitate aggressive therapy. However, the use of high-dose chemotherapy with autologous stem cell transplantation (HCT-ASCT) is poorly documented. We retrospectively analysed from the French LOC network database the adult immunocompetent patients treated with HCT-ASCT for isolated VRL. Thirty-eight patients underwent consolidation with HCT-ASCT for isolated VRL between 2008 and 2019 after induction chemotherapy. Twenty patients had primary VRL, and 18 had an isolated VRL relapse of a primary CNS lymphoma. Three patients underwent HCT-ASCT in first-line treatment, 24 in second-line treatment, and 11 in subsequent lines. At HCT-ASCT, the median age was 61 years, and the median KPS was 90. Thirty-two patients (84%) received high-dose thiotepa-based HCT. One patient (3%) died from HCT-ASCT toxicity. Nineteen (50%) patients relapsed after HCT-ASCT, including 17 cases occurring in the brain. The median progression-free survival, brain-free survival and overall survival from HCT-ASCT were 96, 113 and 92 months, respectively. HCT-ASCT represents an effective therapeutic strategy for select VRL patients, with a tolerable safety profile. However, the risk of subsequent brain relapse remains significant
Real-world multicentre study of cefiderocol treatment of immunocompromised patients with infections caused by multidrug-resistant Gram-negative bacteria: CEFI-ID.
No full textInternational audienceThe increase in the population of immunocompromised patients due to advances in management of end-stage diseases and transplants poses challenges in treating infections caused by multi-drug resistant (MDR) pathogens. Cefiderocol (FDC), a siderophore cephalosporin, has shown efficacy against carbapenem-resistant Gram-negative bacteria. This retrospective multicentre study investigated the real-world use of FDC in 114 immunocompromised adults treated for MDR infections in 12 French hospitals (June 2020-November 2023). Clinical and microbiological outcomes, including infection cure, relapse, as well as mortality, and resistance acquisition, were assessed at days 28 and 90. Antibiotic prescription compliance with current guidelines was investigated. At day 28, clinical success was achieved in 53.3% of cases, and overall mortality was 37.7%, consistent with other studies (33-37%). Infection-related mortality accounted for 25.4%. Relapse occurred in 17.5% of patients by day 28, rising by an additional 9.8% among survivors by day 90. Resistance acquisition was observed in two cases at day 28 (Pseudomonas aeruginosa and Stenotrophomonas maltophilia) and in three additional cases by day 90. FDC was used as monotherapy in 49.1% of cases, with a median treatment duration of 10 days. Nearly 25% of strains collected in FDC-treated patients were susceptible to best-practice alternatives. These findings highlight FDC's utility in difficult-to-treat infections, particularly S. maltophilia, but the high relapse rate and resistance acquisition underscore the need for careful monitoring, adherence to guidelines, and reconsideration of empirical use to prevent resistance and improve outcomes in fragile populations
MRI management of NMOSD and MOGAD: Proposals from the French Expert Group NOMADMUS
No full textInternational audienceBackground: Currently, there are no available recommendations or guidelines on how to perform MRI monitoring in the management of neuromyelitis optica spectrum disorder (NMOSD) and myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD). The issue is to determine a valuable MRI monitoring protocol to be applied in the management of NMOSD and MOGAD, as previously proposed for the monitoring of multiple sclerosis. Objectives: The objectives of this work are to establish proposals for a standardized and feasible MRI acquisition protocol, and to propose control time points for systematic MRI monitoring in the management of NMOSD and MOGAD. Methods: A steering committee composed of 7 neurologists and 5 neuroradiologists, experts in NMOSD and MOGAD from the French group NOMADMUS, defined 8 proposals based on their expertise and a review from the literature. These proposals were then submitted to a Rating Group composed of French NMOSD / MOGAD experts. Results: In the management of NMOSD and MOGAD, a consensus has been reached to perform systematic MRI of the brain, optic nerve and spinal cord, including cauda equina nerve roots, at the time of diagnosis, both without and after gadolinium administration. Moreover, it has been agreed to perform a systematic MRI scan 6 months after diagnosis, focusing on the area of interest, both without and after gadolinium administration. For long-term follow-up of NMOSD and MOGAD, and in the absence of clinical activity, it has been agreed to perform gadolinium-free MRI of the brain (+/- optic nerves) and spinal cord, every 36 months. Ideally, these MRI scans should be performed on the same MRI system, preferably a 3T MRI system for brain and optic nerve MRI, and at least a 1.5T MRI system for spinal cord MRI. Conclusions: This expert consensus approach provides physicians with proposals for the MRI management of NMOSD and MOGAD. (c) 2024 The Authors. Published by Elsevier Masson SAS. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/
Wear factor comparison between single and dual mobility cup in total hip arthroplasty
No full textInternational audienceThe dual mobility cup (DMC) was developed to decrease wear, increase stability, and restore patients’ hip range of motion compared to the traditional single mobility cup (SMC) in total hip arthroplasty (THA). A rejuvenation of the THA population is increasing the risk of wear, and more personalised follow-up of patients is required. The aim of this study was to estimate DMC-THA wear during daily life activities based on 3D motion capture data. Using a musculoskeletal and a stem–liner contact model, a wear factor was computed for 16 participants with SMC-THA and compared to virtual DMC-THA. The force–velocity factor was estimated for slow, self-selected, and fast gait, as well as sit-to-stand and stand-to-sit activities. A significantly lower factor was found for DMC-THA than for SMC-THA during gait at the different speeds and during sit-to-stand. In the stand-to-sit task, the potential advantage of DMC-THA was not demonstrated, likely because the liner kinematics was simulated during intervals in the activity when hip contact force and sliding velocity were high. This study suggests the potential lower risk of wear of the DMC-THA, based on 3D motion capture data and multibody dynamics simulations. It also showed that an in vivo risk of wear can be estimated to help stratifying patients
