Portail HAL de l'Université Picardie Jules Verne
Not a member yet
32425 research outputs found
Sort by
Density estimates for a (non)local variational model with degenerate double-well potential
No full textInternational audienc
Shoulder rotators isokinetic profile according to instability and/or sport specificity: Implications for rehabilitation
No full textInternational audienceObjectives: To (i) describe the shoulder rotator muscle profiles across healthy, unstable, and athletic contexts, and (ii) to compare the usual Peak Torque (PT-method) and the Angular Range (AR-method).Design: Retrospective analysis of data collected cross-sectionally.Setting: Hospital. Each participant participated in isokinetic evaluations of the shoulder rotator muscles at 60°.s-1 (concentric/eccentric) for both shoulders.Participants: 239 participants of 24.7 (7.5) years INTERVENTION: None.Main outcome measures: We recorded the PT and AR mean torque by 10°, and we calculated the antagonist/agonist ratios. We used a two-way repeated measures ANOVA with a correction for multiplicity to compare laterality (i.e., side-to-side) and contexts (i.e., no-overhead sports healthy, no-overhead sports with unstable shoulder, overhead sports healthy and overhead sports with unstable shoulder RESULTS: Concentric PT of external rotators were significantly lower in no-overhead athletes with shoulder instability than healthy no-overhead athletes (p=0.007) and than healthy overhead athletes (p=0.029). The AR highlighted significant (p<0.05) lower concentric external and internal rotator muscles strength: i) in no-overhead athletes with shoulder instability than in healthy no-overhead athletes; ii) in healthy overhead athletes than in healthy no-overhead athletes; iii) in no-overhead athletes with shoulder instability than in healthy overhead athletes. No significant difference was observed in the eccentric modality or in the PT/AR ratios. Significant side-to-side differences (p<0.05) between dominant and non-dominant sides were reported by both PT and AR methods.Conclusion: Only the concentric muscle profiles differed across context groups. The AR allowed for a more precise detection of shoulder muscle adaptations by identifying unique muscle signatures in the moment-angle relationship. These 10° angular range measurements offer complementary information and enhance the clinical utility of isokinetic profiling compared to the traditional PT-method
On the legal foundations of green bonds
No full textInternational audienceThis paper presents one of the first systematic analyses of the legal foundations of green bond issuance. We find that French and Scandinavian civil law origins are positively associated with green bond issuance, whereas English common law origin is negatively associated. These results are robust across alternative measures of green bond issuance, and two quasi-natural experiments. We further show that stronger creditor-rights protection (consistent with stakeholder theory) significantly enhances green bond issuance, while greater financial liberalization (consistent with financial liberalization theory) significantly reduces it. Finally, we find that the positive impact of green bond issuance on renewable energy generation and environmental project financing is stronger under French, Scandinavian, and German civil law origins, and weaker under English common law. Taken together, these findings contribute to debates on firms' ethical responsibilities in addressing climate change and carbon dioxide emissions, and highlight the critical role of legal systems in financing the climate transition
Advanced integrated human in vitro approach to explore gut-brain barrier interactions with gut microbiota metabolites and pesticides
No full textInternational audienceThe gut microbiota (GM) plays a central role in host barrier homeostasis. Alterations in GM due to chronic dietary pesticide residues can have far-reaching consequences on human health. Although in vitro models are increasingly used to study organophosphate effects, especially Chlorpyrifos (CPF), accurately modeling complex GM-host interactions remains challenging. Here we investigated how CPF-altered GM metabolites influence epithelial and brain barrier integrity. An integrated human-relevant in vitro approach combining three complementary models: the Simulator of the Human Intestinal Microbial Ecosystem (SHIME®) under control and CPF-exposed conditions to evaluate both short-term and long-term effects (15 and 30 days), a Caco-2-based intestinal barrier (IB) and a blood-brain barrier (BBB) co-culture. Two parallel experimental setups (SHIME-IB and SHIME-BBB) were developed: SHIME®-derived microbial metabolites and pesticide residues supernatants were directly applied to each barrier model for 24 h. Confocal imaging revealed discontinuous localization of tight junction proteins (occludin, claudin-5 and ZO-1) without concomitant increase in FITC-dextran apparent permeability or overt cytotoxicity. This structural disorganization occurred despite unchanged transcriptional expression of tight junctions, except for a significant reduction of CLDN5 mRNA in the BBB at CPF15, suggesting early molecular signals. In contrast, we showed a selective decrease in P-GP (P-glycoprotein) expression in both barriers, confirmed at the protein level. Concurrently, IL-8 (Interleukin-8) secretion increased markedly in IB, particularly at CPF15, highlighting its potential as an early biomarker chemokine-driven inflammatory activation. This integrated approach provides an ethical model to study environmental contaminants' effects on human barriers, revealing microbiota-mediated disruption by CPF
Millimeter-wave high frequency 5G (26 GHz) electromagnetic fields do not modulate human brain electrical activity
No full textInternational audienceThe deployment of 5G networks utilizing millimeter-wave frequencies such as 26 GHz has raised concerns aboutpotential neurophysiological effects. However, no controlled studies have investigated the impact of 26 GHzexposure on human brain electrical activity.We conducted a randomized, triple-blind crossover study in 31 healthy young adults (18 men, 14 women,mean age 26.1 ± 5.2 years). Participants underwent two sessions (real and sham exposure) separated by oneweek, with 26.5-min exposure to 26 GHz electromagnetic fields at 2 V/m. EEG activity was recorded before,during, and after exposure. Power spectral density was computed for delta (1–4 Hz), theta (4–8 Hz), alpha (8–12Hz), and beta (12–35 Hz) frequency bands. Statistical analysis employed mixed-effects models with baselinecorrection, examining exposure effects across temporal phases and electrode clusters.No significant modulation of EEG frequency bands was observed during eyes-closed conditions following 26GHz exposure. Mixed-effects modeling revealed no significant main effects or interactions for exposure conditionsacross all frequency bands and electrode clusters.This first controlled investigation of 26 GHz 5G effects on human EEG activity found no detectable alterationsin brain electrical activity under regulatory-compliant exposure conditions. These findings contribute importantpreliminary safety data for 5G millimeter-wave technology deployment, though further research across diversepopulations and exposure scenarios remains warranted
Missed Opportunity for Fibrinolysis in ST-Segment–Elevation Myocardial Infarction: The Nationwide France-PCI Registry
No full textInternational audienceBackground Timely reperfusion is a key quality-of-care target in ST-segment–elevation myocardial infarction (STEMI). When primary percutaneous coronary intervention (PPCI) cannot be delivered within 120 minutes from first medical contact (FMC), guidelines recommend fibrinolysis as early as possible within 12 hours of symptom onset in eligible patients. We quantified missed opportunities for fibrinolysis in a nationwide STEMI network. Methods We analysed consecutive STEMI patients enrolled ≤24 hours from symptom onset in the France-PCI registry (2014–2022). FMC was approximated by the first diagnostic ECG. Initial reperfusion was classified as timely PPCI (FMC-to-device ≤120 min), delayed PPCI (>120 min), or fibrinolysis. Among delayed PPCI, eligibility for fibrinolysis required no oral anticoagulant, no prior stroke, and no documented contraindication. We evaluated temporal trends, regional variation, and outcomes. Results Among 19,472 patients, 12,633 (64.9%) underwent timely PPCI, 5,895 (30.3%) delayed PPCI, and 944 (4.8%) fibrinolysis. Timely PPCI increased over time, whereas fibrinolysis declined. Among delayed PPCI, 3,279/5,895 (55.6%) presented within the prespecified early-presenter window (symptom-to-ECG ≤3 h) and met our strict fibrinolysis-eligibility criteria, yet underwent delayed PPCI; this proportion remained stable across years, with marked regional heterogeneity. Fibrinolysis use was favoured by mobile intensive care units, helicopter transport, and longer distance to PPCI centres, whereas older age was associated with delayed PPCI without fibrinolysis. Conclusions In this national STEMI network, more than half of delayed PPCI in eligible early presenters represented a persistent missed-fibrinolysis gap. Routine audit of delayed PCI and missed fibrinolysis as system-level quality metrics should guide time-based pre-hospital triage and align reperfusion with guideline-recommended targets
When One’s Secret Inwardness becomes a Commodity: Kierkegaardian and Marxist critical perspectives on Religious Trauma
No full textInternational audienc
Predicting early progression in advanced hepatocellular carcinoma treated with atezolizumab-bevacizumab: A multimodal deep learning approach
No full textInternational audienceBackground & Aims: Atezolizumab plus bevacizumab represents the current standard first-line treatment for advanced HCC, although individual therapeutic response varies considerably and remains challenging to predict. We developed a multimodal artificial intelligence framework combining clinical data with computed tomography imaging to predict six-month progression-free survival in patients undergoing atezolizumab-bevacizumab therapy.Methods: We conducted a retrospective analysis of 62 patients with advanced HCC receiving first-line atezolizumab-bevacizumab at a tertiary referral centre (University of Palermo, Italy). A deep learning architecture was developed, integrating a convolutional neural network for CT image analysis (arterial, venous, and delayed phases) with a multilayer perceptron for structured clinical variables. The model was trained to predict progression-free survival status at six months. Performance metrics included accuracy, precision, recall, F1-score, balanced accuracy, and area under the receiver operating characteristic curve. Results were benchmarked against conventional logistic regression using clinical variables only.Results: At a median follow-up of 15.9 months, median overall survival reached 24.3 months (95% CI: 15.9-38.0). Twenty-three patients (37.1%) experienced disease progression or death within six months. The multimodal model attained an AUC of 0.86, with accuracy of 80.1%, specificity 84.7%, sensitivity 68.9%, F1-score 67.0%, and balanced accuracy 76.8%. Conversely, traditional logistic regression produced an AUC of 0.67 (95% CI: 0.54-0.78), with neoplastic portal vein invasion demonstrating a trend toward significance (OR: 2.79, 95% CI: 0.98-7.92, p=0.053) in multivariable analysis.Conclusions: A multimodal artificial intelligence approach combining CT imaging with clinical data shows promising performance in predicting early disease progression among HCC patients treated with atezolizumab-bevacizumab. Although preliminary and requiring external validation, these results suggest that deep learning frameworks may improve risk stratification and facilitate personalized treatment decision-making in advanced hepatocellular carcinoma
A multimodal artificial intelligence approach for predicting therapeutic decisions in hepatocellular carcinoma: integrating clinical and imaging data
No full textInternational audienceBackground & Aims: Multidisciplinary management of hepatocellular carcinoma (HCC) involves complex decision-making that integrates tumor burden, liver function, and performance status. While multidisciplinary team approaches improve patient outcomes, they remain resource-intensive and not universally accessible. We developed a multimodal deep learning framework that integrates unstructured clinical text reports and computed tomography (CT) imaging to predict therapeutic decisions in HCC patients.Methods: We retrospectively analyzed 240 unstructured clinical reports with corresponding CT images from 204 patients evaluated at a tertiary referral center between September 2020 and November 2024. Therapeutic decisions were classified into seven categories: liver transplantation, surgical resection, percutaneous ablation, transarterial treatments, systemic therapy, best supportive care, and continued follow-up. We developed a multimodal architecture combining BioBERT for clinical text analysis with a convolutional neural network for CT image feature extraction. The model was trained using an 80:20 split and evaluated through accuracy, sensitivity, specificity, precision, F1-score, and balanced accuracy metrics. Results: The study cohort comprised 204 patients (mean age 71.6±10.3 years, 71.2% male). Distribution of therapeutic decisions included: transplantation (3.8%), resection (5%), ablation (7%), transarterial treatments (8.7%), systemic therapy (32%), best supportive care (22%), and follow-up (21.2%). The multimodal model achieved robust performance metrics in replicating expert clinical decisions: 95.5% overall accuracy, 95.0% sensitivity, 94.0% precision, 94.4% F1-score, 95.0% balanced accuracy, and 84.7% specificity. Confusion matrix analysis revealed excellent discrimination across all therapeutic classes with minimal misclassification between treatment categories. Conclusions: This multimodal artificial intelligence framework, integrating unstructured clinical reports and radiological imaging, accurately predicts therapeutic decisions in HCC with performance close to 95%. Although prospective validation and enhanced interpretability are needed, this approach shows promise as a clinical decision support tool for standardizing care, optimizing multidisciplinary team workflows, and facilitating evidence-based treatment selection in HCC management
Deep Learning for Automated Classification of Liver Tumours on Digitalized Liver Biopsies: An Unsupervised and Semi- Supervised Approach
No full textInternational audienceBackground&Aims: Automated classification of hepatic tu- mours on liver biopsies remains challenging due to morpholog- ical overlaps. We developed deep learning models to discrimi- nate hepatocellular carcinoma (HCC), cholangiocarcinoma (CCA), and metastatic lesions on digitalized whole slide images (WSI).Methods: We retrospectively analysed 210 liver biopsy samples from 204 patients (2020-2025) at the University of Palermo, in- cluding HCC (n=84), intrahepatic CCA (n=34), hepatocholangiocar- cinoma (n=4), liver metastases (n=65), and non-neoplastic sam- ples (n=23). WSI were digitized in NDPI format. Initially, we ap- plied unsupervised learning: WSI were segmented into 512 × 512px patches, deep features extracted using pre-trained ResNet-50, di- mensionality reduction performed with UMAP, and clustering ex- ecuted with HDBSCAN followed by agglomerative merging. Subse- quently, we developed a semi-supervised Mean Teacher model us- ing 30% expert-annotated WSI and 70% unlabeled data with con- sistency regularization.Results: Unsupervised clustering achieved low purity: Cluster 0 (42% HCC, 11% CCA, 47% negative), Cluster 1 (35% HCC, 16% CCA, 48% negative), demonstrating inadequate class discrimination. The semi-supervised model achieved markedly superior performance: 97.85% training accuracy and 89.24% validation accuracy. Starting from limited annotations (30%), the model effectively generated high-quality pseudo-labels for the unlabeled majority, learning ro- bust features despite class imbalance and successfully discriminat- ing between hepatic tumour types.Conclusions: Semi-supervised learning significantly outper- formed unsupervised approaches for hepatic tumour classification. This model shows promise as a first-line screening tool in liver biopsies to optimize diagnostic workflow and guide immunohisto- chemical panel selection, serving as an assistant for junior pathol- ogists and quality assurance instrument in low-volume centres