Automatic speech recognition (ASR) and natural language processing (NLP) play key roles in advancing human–technology interactions, particularly in healthcare communications. This study aims to enhance French-language online mental health platforms through the adaptation of the QuartzNet 15 × 5 ASR model, selected for its robust performance across a variety of French accents as demonstrated on the Mozilla Common Voice dataset. The adaptation process involved tailoring the ASR model to accommodate various French dialects and idiomatic expressions, and integrating it with an NLP system to refine user interactions. The adapted QuartzNet 15 × 5 model achieved a baseline word error rate (WER) of 14%, and the accompanying NLP system displayed weighted averages of 64.24% in precision, 63.64% in recall, and an F1-score of 62.75%. Notably, critical functionalities such as ‘Prendre Rdv’ (schedule appointment) achieved precision, recall, and F1-scores above 90%. These improvements substantially enhance the functionality and management of user interactions on French-language digital therapy platforms, indicating that continuous adaptation and enhancement of these technologies are beneficial for improving digital mental health interventions, with a focus on linguistic accuracy and user satisfaction.
Sepsis remains a major global health concern, causing high mortality rates, prolonged hospital stays, and substantial economic burdens. The accurate prediction of clinical outcomes, such as mortality and length of stay (LOS), is critical for optimizing hospital resource allocation and improving patient management. The present study investigates the potential of machine learning (ML) models to predict these outcomes using a dataset of 1492 sepsis patients with clinical, physiological, and demographic features. After rigorous preprocessing to address missing data and ensure consistency, multiple classifiers, including Random Forest, Extra Trees, and Gradient Boosting, were trained and validated. The results demonstrate that Random Forest and Extra Trees achieve high accuracy for LOS prediction, while Gradient Boosting and Bernoulli Naïve Bayes effectively predict mortality. Feature importance analysis identified ICU stay duration (ICU_DAYS_OBS) as the most influential predictor for both outcomes, alongside vital signs, white blood cell counts, and lactic acid levels. These findings highlight the potential of ML-driven clinical decision support systems (CDSSs) to enhance early risk assessment, optimize ICU resource planning, and support timely interventions. Future research should refine predictive features, integrate advanced biomarkers, and validate models across larger and more diverse datasets to improve scalability and clinical impact.
(1) Background: Many familial diseases are caused by genetic accidents, which affect both the genome and its epigenetic environment, expressed as an interaction graph between the genes. We wish to show the existence of a form of robustness for such interaction networks in the case of three familial diseases: Familial angioedema; Osteogenesis imperfecta; Biliary atresia; (2) Methods: The update of the gene states at the vertices of the interaction graph (1 if a gene is activated, 0 if it is inhibited) can be done in multiple ways, well studied over the last 20 years: parallel, block-parallel, sequential, block-sequential, random, etc. (3) Results: We study three particular interaction graphs, linked to the familial diseases chosen as examples. They have in common a subgraph that is activated in an intricate manner, a new update mode, of which we study the robustness; (4) Conclusions: Comparison of the interaction modes of the three familial diseases studied, which are of different etiology, shows visible common points in the regulatory graphs of the genes involved, in particular a structural robustness with respect to their expression clock.
Many familial diseases are caused by genetic accidents, which affect the genome and its epigenetic environment, summarized as an interaction network between genes. We wish to study the existence or absence of robustness for such genetic interaction networks centered on the gene SP1 and involved in three familial diseases: familial angioedema, osteogenesis imperfecta, and biliary atresia. The updating of gene states at the vertices of the interaction graph of the genetic network (1 if a gene is activated, 0 if it is inhibited) can be performed in multiple ways that have been well-studied over the last 20 years: parallel, block-parallel, sequential, block-sequential, random, etc. We add to these classic updating modes two new ones, the intricate and the state-dependent. We have studied the robustness of three particular interaction graphs related to the familial diseases chosen as examples. The comparison of the interaction graphs and dynamics of the chosen familial diseases of different etiology shows common points in their interaction graphs and similarities in their dynamics according to their expression clock.