HAL-Paris 13
Not a member yet
33893 research outputs found
Sort by
Targeting high circulating dipeptidyl peptidase 3 in circulatory failure
No full textInternational audienceCirculating dipeptidyl peptidase 3 is a new biomarker linked to circulatory failure prognosis and pathophysiology and is a potential actionable therapeutic target. In this short review intended for the clinician, a question-and-answer format provides key insights on the nature of this biomarker and the therapeutical potential of its targeted inhibition in critically ill patients
Care management and determinants of day 14 mortality in severely ill children aged under 5 years subsequent to hypoxaemia diagnosed using routine pulse oximetry in primary care: evidence from the AIRE project
No full textInternational audienceBackground The Amélioration de l'Identification des détresses Respiratoires de l'Enfant (AIRE) project introduced the routine use of pulse oximetry (PO) into Integrated Management of Childhood Illness (IMCI) consultations within primary health centres (PHCs) in Burkina Faso, Guinea, Mali and Niger. We analysed how severe cases were managed and 14-day mortality by hypoxaemia severity. Methods All children aged under 5 years attending IMCI consultations integrating PO use at 16 research PHCs and classified as severe cases (severe IMCI cases or severe hypoxemia: SpO 2 <90%) were eligible for referral and enrolled in a 14-day prospective cohort with parental consent. Referral decisions, admissions, access to oxygen therapy and Kaplan-Meier probability of death were compared by hypoxaemia severity. An adjusted mixed-effects Cox regression model with a random effect for PHC estimated adjusted ORs (aORs) and 95% CIs of mortality by day 14. Results From July 2021 to July 2022, 1998 severe cases were enrolled, including 10.6% aged <2 months; 7.1% had severe hypoxaemia, and 10.5% had moderate hypoxaemia (90%≤oxygen saturation≤93%). By day 14, 625 (31.3%) were referred, 463 (23.2%) hospitalised, and 95 children (4.8%) had died. Referral decisions, hospitalisations and oxygen therapy rates were significantly higher for severe hypoxaemic cases (83.8%, 82.3% and 34.5%, respectively) than for moderate hypoxaemic cases (32.7%, 26.5% and 7.1%, respectively) and cases without hypoxaemia (26.3%, 17.5% and 1.4%, respectively). Similarly, day 14 mortality rates were 26.1%, 7.5% and 2.3%, respectively. The aORs for mortality were severe hypoxaemia (9.34, 95% CI 5.08 to 17.16), moderate hypoxaemia (2.32, 95% CI 1.16 to 4.64), age <2 months (3.68, 95% CI 1.67 to 8.13), severe malaria (2.02, 95% CI 1.03 to 3.97) and living in Niger (4.06, 95% CI 1.41 to 11.67). Conclusion Regardless of severity, hypoxaemia was common among outpatients screened using PO and meeting criteria for severity. Its presence was associated with mortality risk. Incorporating PO within IMCI prompted care management of severely hypoxaemic cases, but hospital referrals and access to oxygen remain sub-optimal and are crucial levers for reducing under-five mortality. Study registration number PACTR202206525204526 registered retrospectively on 15 June 2022
Transfert inter-dialectal et généralisation zéro-shot en arménien
No full textInternational audienceWe evaluate lemmatization, POS-tagging, and morphological analysis across four Armenian varieties : Classical Armenian (CA), Modern Eastern Armenian (MEA), Modern Western Armenian (MWA), and the under-documented Getashen dialect (G). Three model families are compared : RNNs, mDeBERTa, and GPT-4-Turbo-2024-04-09. RNNs perform best in supervised settings, mDeBERTa captures morphology across standards, while GPT-4-Turbo shows superior zero- and few-shot transfer to Getashen. These results highlight in-context learning as a powerful strategy for cross-dialectal and low-resource NLP.Cet article évalue trois tâches de TAL (lemmatisation, étiquetage morphosyntaxique, analyse morphologique) sur quatre variétés de l'arménien : arménien classique, oriental moderne, occidental moderne et le dialecte de Getashen. Trois familles de modèles sont comparées : réseaux récurrents (RNN), transformer multilingue (mDeBERTa) et modèle génératif (GPT-4-Turbo-2024-04-09). Les RNNs excellent en supervision complète, mDeBERTa modélise efficacement la morphologie, tandis que GPT-4-Turbo surpasse les autres en transfert zéroet few-shot vers Getashen. Ces résultats soulignent l'efficacité de l'apprentissage in-context pour le TAL des langues peu dotées et des variétés dialectales
Improving risk stratification and detection of early HCC using ultrasound-based deep learning models
No full textInternational audienceBackground & Aims: Hepatocellular carcinoma (HCC) surveillance programs are suboptimal. We aimed to design an ultrasound-based deep learning model for HCC risk stratification (STARHE-RISK) and early-stage HCC detection (STARHE-DETECT) in patients with compensated advanced chronic liver disease (cACLD).Methods: This prospective multicentric study included 403 adult patients with cACLD of all causes enrolled in a surveillance program for at least 6 months without prior history of HCC. STARHE-RISK was trained on ultrasound cine clips of the non-tumoral liver parenchyma using two classes: cases (n = 152 patients with early-stage HCC; 137/152 [82%] male; median age 63 years) and controls (n = 170 patients without HCC at inclusion and during a subsequent 1-year follow-up; 120/170 [71%] male; median age 69 years). STARHE-DETECT was trained on tumour ultrasound cine clips. The training/validation and testing sets were stratified according to potential confounders, and 50 patients who were balanced in both groups were allocated to the independent testing set based on sample size calculation. Statistical analysis included classification and detection metrics.Results: STARHE-RISK achieved good prediction performances in the testing set with a 0.72 accuracy (95% CI 0.57–0.84) and an odds ratio of 6.6 (95% CI 1.9–22.7; p = 0.003). The combination of STARHE-RISK and the FASTRAK score, a multi-aetiology HCC risk stratification score, achieved a higher specificity (0.86 [95% CI 0.65–0.97]) and odds ratio (8.9 [95% CI 2.1–38.3; p = 0.004]) for predicting a patient at high risk of HCC development. STARHE-DETECT achieved a 0.67 mAP10, a 0.68 sensitivity (95% CI 0.47–0.85), and a 0.82 specificity (95% CI 0.69–0.91) for detecting early-stage HCC.Conclusions: STARHE-RISK and STARHE-DETECT achieved robust performances for HCC risk stratification and early-stage HCC detection, respectively. They could become valuable surveillance tools and pave the way for a risk-based personalised surveillance program.Impact and implications: STARHE-RISK is a reliable ultrasound-based deep learning model for hepatocellular carcinoma (HCC) risk stratification in patients with compensated advanced chronic liver disease and can be associated with complementary scores integrating clinical and blood parameters. STARHE-DETECT could become a complementary tool to visual assessment for radiologists and sonographers in HCC surveillance. Both models are based on simple and easy-to-perform ultrasound cine clip acquisitions. This study paves the way for a risk-based personalised surveillance program that will not ultimately rely on a single test but rather on a combination of approaches mixing clinical, biological, and radiological data
Addressing Explicit Weight Bias in Medical Students: Contribution of Demographics and Educational Factors
No full textInternational audienceObjective This study assesses explicit weight bias (EWB) among French medical students and its association with demographic factors and educational tools. Methods A cross-sectional study assessed EWB among 1,635 students from two universities between June and September 2024. The Anti-Fat Attitudes Questionnaire measured three dimensions: dislike, fear of fat, and belief in weight controllability (willpower). Academic and demographic data were collected, with scores adjusted for sex. Two educational tools were evaluated: a podcast addressing weight bias ("Augusta’s Oath") and a clinical rotation in a nutrition department. Results On a 1-to-9 scale where higher scores indicate stronger weight bias, medical students showed a moderate overall score (2.7 +/- 1.1), with low levels of dislike toward individuals with obesity (1.9 +/- 1.2), moderate belief in weight controllability (willpower, 3.1 +/- 1.9), and high levels of personal fear of gaining weight (4.2 +/- 2.1), indicating a persistent presence of explicit weight bias. Men exhibited higher EWB than women. Students who listened to the podcast had significantly lower willpower scores than non-listeners (2.6 +/- 1.5 vs. 3.1 +/- 1.7; p < 0.01) and were half as likely to score above 4 i.e. explicitly expressing bias (59/470; 12% vs. vs. 279/1164; 24%; OR: 0.50, 95% CI: [0.33; 0.74]; p < 0.01). Nutrition rotations were also associated with slightly lower willpower scores (2.8 +/- 1.7 vs. 3.0 +/- 1.7, p < 0.01). Neither educational exposure was associated with dislike or fear scores. Conclusion EWB, especially fear of getting fat and willpower beliefs, is prevalent among French medical students. Educational podcasts show promise in reducing specific biases, offering tools to combat weight stigma in medical education
UNHaP: Unmixing Noise from Hawkes Processes
No full textInternational audiencePhysiological signal analysis often involves identifying events crucial to understanding biological dynamics. Many methods have been proposed to detect them, from handcrafted and supervised approaches to unsupervised techniques. All these methods tend to produce spurious events, particularly as they detect each event independently. This work introduces UNHaP (Unmix Noise from Hawkes Processes), a novel approach addressing the joint learning of temporal structures in events and the removal of spurious detections. By treating the event detection output as a mixture of structured Hawkes and unstructured Poisson events, UNHaP efficiently unmixes these processes and estimates their parameters. This approach significantly enhances event distribution characterization while minimizing false detection rates on simulated and real data.</div
Multipole expansion for dispersion forces
No full textInternational audienceLight-matter interaction models invariably rely on the multipole expansion of the electromagnetic potentials generated by complex charge distributions. These multipoles are typically taken to be traceless; however, for a correct evaluation of dispersion forces at all distances, the validity of this assumption has to be checked carefully. Here, we revisit the concept of dispersion forces on an atom near a dielectric surface from the perspective of macroscopic quantum electrodynamics and find that, beyond the quadrupole, the multipoles cannot always be taken as fully traceless. In particular, we show that the trace of the octupole moment contributes to Casimir-Polder interactions beyond the electrostatic regime
Chapitre II. Une numérisation inachevée et inégale des loisirs et de la culture
No full textInternational audienc
MEFA-MS: Attention-Based U-Net for Pedestrian and Vehicle Detection
No full textInternational audienceA Multimodal image Early Fusion with Attention (MEFA) module can be employed for visual object detection, leveraging the robust capabilities of well-established monomodal computer vision models. However, it exhibits poor inference time under real-world usage conditions. The present paper introduces a novel module, referred to as MEFA-MS. This module proposes a U-Net encoder-decoder architecture incorporating spatial and channel attention mechanisms. It aims to reduce inference time while maintaining high precision for pedestrians and vehicles detection. We conduct a comparative study between the proposed MEFA-MS module and the predecessor MEFA module. The proposed approach entails a 2% improvement in mAP50, accompanied by a 50% reduction in inference time for pedestrian and vehicle detection on the DENSE dataset, employing the RT-DETR object detector. Moreover, we demonstrate that the system exhibits superior performance in all weather conditions, achieving a 4% reduction in false positives compared to the previous MEFA module