Enhancing gastroenterology with multimodal learning: the role of large language model chatbots in digestive endoscopy DOI Creative Commons
Yuanyuan Qin, Julie Chang, Li Li

et al.

Frontiers in Medicine, Journal Year: 2025, Volume and Issue: 12

Published: May 21, 2025

Introduction Advancements in artificial intelligence (AI) and large language models (LLMs) have the potential to revolutionize digestive endoscopy by enhancing diagnostic accuracy, improving procedural efficiency, supporting clinical decision-making. Traditional AI-assisted endoscopic systems often rely on single-modal image analysis, which lacks contextual understanding adaptability complex gastrointestinal (GI) conditions. Moreover, existing methods struggle with domain shifts, data heterogeneity, interpretability, limiting their applicability. Methods To address these challenges, we propose a multimodal learning framework that integrates LLM-powered chatbots imaging patient-specific medical data. Our approach employs self-supervised extract clinically relevant patterns from heterogeneous sources, enabling real-time guidance report generation. We introduce domain-adaptive strategy enhance model generalization across diverse patient populations Results discussion Experimental results multiple GI datasets demonstrate our method significantly improves lesion detection, reduces variability, enhances physician-AI collaboration. This study highlights of LLM-based advancing gastroenterology providing interpretable, context-aware, adaptable AI support endoscopy.

Language: Английский

Association between neutrophil-platelet ratio and 28-day mortality in patients with sepsis: a retrospective analysis based on MIMIC-IV database DOI Creative Commons

Jin Zhu,

Chaorong Zhang,

Zhexuan Deng

et al.

BMC Infectious Diseases, Journal Year: 2025, Volume and Issue: 25(1)

Published: May 9, 2025

The immune system and inflammation are intimately linked to the pathophysiology of sepsis. neutrophil‒platelet ratio (NPR), associated with immunology, may be useful in predicting sepsis outcomes. According earlier research, NPR is prognosis several diseases. This study aimed investigate connection between unfavorable outcomes patients We retrieved patient clinical data from Medical Information Mart for Intensive Care IV database (MIMIC-IV 2.2) based on inclusion exclusion criteria. quartile was used divide population into four groups. 28-day mortality main result, whereas 90-day secondary result. Cox regression model, Kaplan‒Meier survival curve, limited cubic spline were examine associations negative Subgroup analysis also conducted. At same time, we Latent Class Trajectory Model (LCTM) assess trajectory within six days ICU admission, relationship at 28 90 days. included 3339 patients. Quartile 4 had greatest rates, according model curve. A J-shaped found restricted investigations. means higher lower NPRs mortality, = 3.81 as tipping point. total 434 analysis, three patterns identified. Patients an increased rate slow-decline group compared stable development group. has prognostic value sepsis, there a two variables. who have slowly declining rate. Not applicable.

Language: Английский

Citations

0
Associations between triglyceride-glucose body mass index and all-cause mortality in ICU patients with sepsis and acute heart failure DOI Creative Commons
Heping Xu,

Jinyuan Xie,

Huan Niu

et al.

BMC Cardiovascular Disorders, Journal Year: 2025, Volume and Issue: 25(1)

Published: May 9, 2025

Language: Английский

Citations

0
The predictive significance of the triglyceride-glucose index in forecasting adverse cardiovascular events among type 2 diabetes mellitus patients with co-existing hyperuricemia: a retrospective cohort study DOI Creative Commons

Jianyong Zhao,

Na Li, Shiqi Li

et al.

Cardiovascular Diabetology, Journal Year: 2025, Volume and Issue: 24(1)

Published: May 21, 2025

The triglyceride-glucose (TyG) index serves as a crucial indicator for evaluating insulin resistance (IR) and cardiovascular risk among patients with type 2 diabetes mellitus (T2DM). Concurrently, hyperuricemia (HUA) strongly correlates adverse outcomes. However, the prognostic value of TyG index, particularly in exhibiting both conditions, remains inadequately defined. This study assessed association between measurements incidence major events (MACEs) simultaneously diagnosed T2DM HUA. retrospective, single-center cohort included 628 HUA at Chaohu Hospital (Anhui Medical University) 2019 2024. Participants were stratified into tertiles based on their values. Kaplan-Meier survival curves log-rank tests estimated MACEs, Cox regression analyses calculated hazard ratios. additional predictive contribution was evaluated using C statistics, net reclassification improvement (NRI), integrated discrimination (IDI) metrics. During 38.00 ± 8.78 months follow-up period, 74 MACEs recorded. A significant proportional relationship emerged events-patients highest tertile demonstrated markedly increased compared those lowest (HR = 2.45, 95% CI 1.23-4.95). pivotal threshold identified > 8.40, beyond which each standard deviation increase corresponded to 66% higher probability 1.66, 1.36-2.36, P 0.014). Integrating traditional models significantly improved performance (C statistic increase: 0.64 → 0.67, 0.029; NRI 0.14, IDI 0.02, < 0.05). constitutes an autonomous MACE predictor specifically within distinctive manifesting is first validate 8.40 identify synergistic interaction serum uric acid (SUA) TyG, providing novel stratification tool managing dual metabolic disorders.

Language: Английский

Citations

0
Construction and clinical visualization application of a predictive model for mortality risk in sepsis patients based on an improved machine learning model DOI Creative Commons
Ting Chen, Xuefeng Zhang, Qifeng Yu

et al.

Frontiers in Physiology, Journal Year: 2025, Volume and Issue: 16

Published: May 21, 2025

Objective To explore the construction and clinical visualization application of a mortality risk prediction model for sepsis patients based on an improved machine learning model. Methods This retrospective study analyzed 1,050 admitted to Longyou County People’s Hospital between January 2010 August 2023. Patients were divided into survival group (n = 877) death 173) their 30-day status. Clinical laboratory data collected used as feature variables. A Self-Weighted Self-Evolutionary Learning Model (SWSELM) was developed identify independent factors create system application. Results The algorithm significantly outperformed other algorithms 23 standard test functions. SWSELM achieved ROC-AUC PR-AUC values 0.9760 0.9624, respectively, training set, 0.9387 0.9390, both higher than those three models. identified 10 important features, with multivariate logistic regression retaining five variables: B-type Natriuretic Peptide Precursor (NT-proBNP), Lactate, Albumin, Oxygenation Index, Mean Arterial Pressure (MAP) (OR 4.889, 3.770, 3.083, 1.872, 1.297), consistent top features selected by Conclusion NT-proBNP, are in patients. successfully created self-evolutionary using methods, demonstrating significant potential value broader implementation.

Language: Английский

Citations

0
Enhancing gastroenterology with multimodal learning: the role of large language model chatbots in digestive endoscopy DOI Creative Commons
Yuanyuan Qin, Julie Chang, Li Li

et al.

Frontiers in Medicine, Journal Year: 2025, Volume and Issue: 12

Published: May 21, 2025

Introduction Advancements in artificial intelligence (AI) and large language models (LLMs) have the potential to revolutionize digestive endoscopy by enhancing diagnostic accuracy, improving procedural efficiency, supporting clinical decision-making. Traditional AI-assisted endoscopic systems often rely on single-modal image analysis, which lacks contextual understanding adaptability complex gastrointestinal (GI) conditions. Moreover, existing methods struggle with domain shifts, data heterogeneity, interpretability, limiting their applicability. Methods To address these challenges, we propose a multimodal learning framework that integrates LLM-powered chatbots imaging patient-specific medical data. Our approach employs self-supervised extract clinically relevant patterns from heterogeneous sources, enabling real-time guidance report generation. We introduce domain-adaptive strategy enhance model generalization across diverse patient populations Results discussion Experimental results multiple GI datasets demonstrate our method significantly improves lesion detection, reduces variability, enhances physician-AI collaboration. This study highlights of LLM-based advancing gastroenterology providing interpretable, context-aware, adaptable AI support endoscopy.

Language: Английский

Citations

0