Publications scientifiques de l'Université de Reims Champagne-Ardenne
Not a member yet
27404 research outputs found
Sort by
Dual-energy CT for peritoneal metastases: When the right question meets the right technology
No full textInternational audienc
Targeting bacterial motility: Prospective therapeutic countermeasures
No full textInternational audienc
Electrical properties reconstruction from MRI data: theoretical and numerical aspects
No full textInternational audienceThe paper presents an inverse problem for the reconstruction of the electrical properties of the human brain from Magnetic Resonance Imaging -data. We establish a rigorous mathematical framework for the formulation of the inverse problem and provide an identifiability result. To achieve a robust numerical solution, the contrast source inversion method is successfully applied to both academic and realistic configurations. The finite element method is used to discretize the boundary value problems, allowing complex geometries and realistic boundary conditions to be handled accurately
Segmentation en EEG néotanal — Apprentissage sur des représentations temps-fréquence
No full textNational audienceL’électroencéphalogramme (EEG) est utilisée en routine clinique pour le diagnostic et le suivi des nouveau-nés, notamment en cas de suspicion d’encéphalopathie anoxo-ischémique (HIE). L’interprétation de l’EEG pour la prise en charge du patient nécessite de localiser des figures physiologiques et/ou pathologiques dans le signal. Ceci nécessite un expert formé, disponible immédiatement et pendant toute la période de suivi. La conception d’outils semi-automatiques de détection / segmentation de motifs est ainsi un enjeu majeur pour améliorer les temps de diagnostic et d’intervention.L’apprentissage profond ouvre la voie au développement de tels outils. Néanmoins, dans le contexte clinique considéré, les motifs recherchés sont très peu nombreux dans des données de grande taille. L’utilisation brute du signal sous forme de série temporelle a fait l’objet d’une étude précédente qui a notamment montré les limites d’une telle approche (1). Nous explorons, dans ce travail, une stratégie alternative qui consiste à représenter les donnéesEEG dans un paradigme temps-fréquence. Nous étudions notamment dans quelle mesure ce changement de représentation est à même de permettre une segmentation efficace de motifs en EEG.Le signal EEG est subdivisé en fenêtres indépendantes de 2 560 points (10 secondes). Une transformée de Pseudo Wigner-Ville lissée (2) (avec un fenêtrage de Hamming en temps et en fréquence : 51 points en temps, 101 en fréquence) est appliquée pour extraire le module du spectre. Les images obtenues sont tronquées pour conserver l’information fréquentielle de 0 à 32 Hz, et sous-échantillonnées pour aboutir à une dimension de 256×32. Seules les fenêtres contenant une annotation de motif (encoche frontale, pointe, dysrythmie lente antérieure) sont utilisées. Les instants correspondant à une encoche frontale sont marqués comme cas positif, les autres comme cas négatifs. Chaque canal du signal EEG est considéré indépendamment.Nous proposons une architecture similaire à U-Net (3), adaptée pour la segmentation d’images de spectre temps-fréquence. Chaque étape implémente une Depthwise Separable Convolution 2D (4) (à padding égal), suivi d’une couche de max-pooling, et d’un sous-échantillonnage. La fonction de perte Focal Loss (5) est utilisée. Les signaux EEG sont issus de l’étude ancillaire LyTONEPAL (6), pour des patients ayant souffert d’une HIE sévère à modérée, dans leurs 6 premières heures de vie. L’étude est multi-centrique. Le montage utilisé dans tous les cas est le même, en référence bipolaire, longitudinal. Le signal est échantillonné à 256 Hz (un filtre passe-bas à 30 Hz est appliqué).La validation expérimentale de notre modèle sur les données EEG montre que l’information relative aux encoches frontales dans le signal brut est utilisable dans la représentation temps-fréquence pour la segmentation (67,16% AUC). La nature différente des modes de données invite à une comparaison plus précise entre série temporelle et de représentation spectrale
A real-world patient MR Angiography (MRA) Database for Brain Vasculature Segmentation and Modelling in Predictive Simulation for the Planning of Interventional Neuroradiology procedures (PreSPIN)
No full textNational audienc
Supervisors' need-related behaviors and employees' functioning: A person-centered investigation
No full textInternational audienceUsing a recent tripartite representation of supervisors’ behaviors rooted in Self-Determination Theory (SDT), this person-centered study sought to identify the different configurations (profiles) of supervisors’ need-supportive, need-thwarting, and need-indifferent behaviors as perceived by their subordinates. It also investigated how these supervisory profiles were related to subordinates’ levels of need satisfaction, affective organizational commitment, performance, and burnout. Latent profile analyses were estimated in a sample of 807 employees (Mage = 39.17; 42.7% women) recruited via the Prolific Academic crowdsourcing platform. Our results revealed five distinct profiles of supervisors’ need-related behaviors: (1) globally unfavorable, (2) globally moderately unfavorable and specifically indifferent, (3) globally moderately favorable, (4) globally unfavorable and specifically thwarting, and (5) globally unfavorable and specifically inconsistent. In line with our expectations, these profiles shared distinct associations with all outcomes. Moreover, our results showed that subordinates’ specific perceptions of their supervisor need-related behaviors mattered, beyond their global perception of their supervisor, when predicting employee functioning. These findings underline the importance of considering global and specific levels of need-related behaviors when seeking to understand supervisory profiles and their associations with employee functioning
Commentary on “Is gaitrite system sensitive in discriminating gait pattern of subjects affected by Charcot Marie tooth? A pilot study”
No full textInternational audienc
A population pharmacokinetic study of ampicillin therapy in hospitalized foals
No full textInternational audienceBackground: Pharmacokinetic studies on ampicillin in foals are limited, underscoring a relevant gap in knowledge, particularly regarding the treatment of critically ill neonatal foals. Hypothesis/Objectives:To evaluate the adequacy of the ampicillin dosing regimen in hospitalized foals and, if necessary, suggest alternative dosing strategies to achieve effective ampicillin concentrations. Animals: Data were collected from 12 hospitalized foals.Methods: Foals were treated with 20 mg/kg ampicillin intravenously every 6 h. Plasma samples were obtained within the first 48 h, and the minimum inhibitory concentration (MIC) was determined for pathogen-positive patients. Ampicillin concentrations were quantified using liquid chromatography-tandem mass spectrometry. A population pharmacokinetic model was developed using a nonlinear mixed-effects approach (stochastic approximation expectation-maximization or SAEM algorithm), and a pharmacodynamic evaluation of different dose regimens was conducted using Monte Carlo simulations. Results:A 2-compartment model with first-order elimination was selected. Age significantly influenced peripheral compartment volume and clearance. The model demonstrated excellent internal validation, with 97% of observed values within prediction intervals and robust stability, as confirmed by bootstrap and visual predictive checks. Pharmacodynamic simulations indicated that a dosage regimen of 20 mg/kg every 6 h achieved optimal PTA (≥90%) for MICs of 0.06-0.25 mg/L with a 50% fT > MIC target. For higher MICs or a 100% fT > MIC target, more frequent dosing (q4h) and higher doses (30-40 mg/kg) were necessary. Conclusions and clinical importance:The results from the simulations highlight the clinical importance of adjusting dosing regimens based on foal characteristics and MIC to ensure effective treatment, especially in critically ill foals.</div
Unraveling the Multiple Biocontrol Mechanisms of Trichoderma spp. in the Protection of Grapevines Against Botrytis cinerea
No full textInternational audienceBotrytis cinerea, the causal agent of grey mold in grapevine, remains one of the most economically important pathogens in viticulture and a key target for sustainable biocontrol strategies. This study evaluated the antagonistic potential of seven Trichoderma isolates (T1–T7), collected from the rhizosphere of grapevine in Morocco, using a combination of in vitro and in planta assays designed to capture multiple direct and indirect modes of action. The isolates exhibited variable levels of antagonism through competition, volatile organic compounds, extracellular metabolites, and elicitation responses. Preliminary in planta assays on detached grape berries further demonstrated that all selected isolates reduced lesion development, with preventive applications yielding the strongest protection. Overall, the study highlights the complementary and strain-specific mechanisms underlying Trichoderma & B. cinerea interactions and underscores the importance of isolate selection and application timing for the development of effective and environmentally friendly grey mold management strategies. These findings provide a mechanistic basis for the future evaluation of promising isolates under vineyard conditions
