BMC Medical Informatics and Decision Making最新文献

{"title":"A novel network-level fused deep learning architecture with shallow neural network classifier for gastrointestinal cancer classification from wireless capsule endoscopy images.","authors":"Muhammad Attique Khan, Usama Shafiq, Ameer Hamza, Anwar M Mirza, Jamel Baili, Dina Abdulaziz AlHammadi, Hee-Chan Cho, Byoungchol Chang","doi":"10.1186/s12911-025-02966-0","DOIUrl":"10.1186/s12911-025-02966-0","url":null,"abstract":"<p><p>Deep learning has significantly contributed to medical imaging and computer-aided diagnosis (CAD), providing accurate disease classification and diagnosis. However, challenges such as inter- and intra-class similarities, class imbalance, and computational inefficiencies due to numerous hyperparameters persist. This study aims to address these challenges by presenting a novel deep-learning framework for classifying and localizing gastrointestinal (GI) diseases from wireless capsule endoscopy (WCE) images. The proposed framework begins with dataset augmentation to enhance training robustness. Two novel architectures, Sparse Convolutional DenseNet201 with Self-Attention (SC-DSAN) and CNN-GRU, are fused at the network level using a depth concatenation layer, avoiding the computational costs of feature-level fusion. Bayesian Optimization (BO) is employed for dynamic hyperparameter tuning, and an Entropy-controlled Marine Predators Algorithm (EMPA) selects optimal features. These features are classified using a Shallow Wide Neural Network (SWNN) and traditional classifiers. Experimental evaluations on the Kvasir-V1 and Kvasir-V2 datasets demonstrate superior performance, achieving accuracies of 99.60% and 95.10%, respectively. The proposed framework offers improved accuracy, precision, and computational efficiency compared to state-of-the-art models. The proposed framework addresses key challenges in GI disease diagnosis, demonstrating its potential for accurate and efficient clinical applications. Future work will explore its adaptability to additional datasets and optimize its computational complexity for broader deployment.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"150"},"PeriodicalIF":3.3,"publicationDate":"2025-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11956435/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143751214","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Leveraging large language models to mimic domain expert labeling in unstructured text-based electronic healthcare records in non-english languages.","authors":"Izzet Turkalp Akbasli, Ahmet Ziya Birbilen, Ozlem Teksam","doi":"10.1186/s12911-025-02871-6","DOIUrl":"10.1186/s12911-025-02871-6","url":null,"abstract":"<p><strong>Background: </strong>The integration of big data and artificial intelligence (AI) in healthcare, particularly through the analysis of electronic health records (EHR), presents significant opportunities for improving diagnostic accuracy and patient outcomes. However, the challenge of processing and accurately labeling vast amounts of unstructured data remains a critical bottleneck, necessitating efficient and reliable solutions. This study investigates the ability of domain specific, fine-tuned large language models (LLMs) to classify unstructured EHR texts with typographical errors through named entity recognition tasks, aiming to improve the efficiency and reliability of supervised learning AI models in healthcare.</p><p><strong>Methods: </strong>Turkish clinical notes from pediatric emergency room admissions at Hacettepe University İhsan Doğramacı Children's Hospital from 2018 to 2023 were analyzed. The data were preprocessed with open source Python libraries and categorized using a pretrained GPT-3 model, \"text-davinci-003,\" before and after fine-tuning with domain-specific data on respiratory tract infections (RTI). The model's predictions were compared against ground truth labels established by pediatric specialists.</p><p><strong>Results: </strong>Out of 24,229 patient records classified as poorly labeled, 18,879 were identified without typographical errors and confirmed for RTI through filtering methods. The fine-tuned model achieved a 99.88% accuracy, significantly outperforming the pretrained model's 78.54% accuracy in identifying RTI cases among the remaining records. The fine-tuned model demonstrated superior performance metrics across all evaluated aspects compared to the pretrained model.</p><p><strong>Conclusions: </strong>Fine-tuned LLMs can categorize unstructured EHR data with high accuracy, closely approximating the performance of domain experts. This approach significantly reduces the time and costs associated with manual data labeling, demonstrating the potential to streamline the processing of large-scale healthcare data for AI applications.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"154"},"PeriodicalIF":3.3,"publicationDate":"2025-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11959812/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143751225","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Developing a high-performance AI model for spontaneous intracerebral hemorrhage mortality prediction using machine learning in ICU settings.","authors":"Xiao-Han Vivian Yap, Kuan-Chi Tu, Nai-Ching Chen, Che-Chuan Wang, Chia-Jung Chen, Chung-Feng Liu, Tee-Tau Eric Nya, Ching-Lung Kuo","doi":"10.1186/s12911-025-02984-y","DOIUrl":"10.1186/s12911-025-02984-y","url":null,"abstract":"<p><strong>Background: </strong>Spontaneous intracerebral hemorrhage (SICH) is a devastating condition that significantly contributes to high mortality rates. This study aims to construct a mortality prediction model for patients with SICH using four various artificial intelligence (AI) machine learning algorithms.</p><p><strong>Method: </strong>A retrospective analysis was conducted on electronic medical records of SICH patients aged 20 and above, admitted to Chi Mei Medical Center's intensive care unit between January 2016 and December 2021. The study utilized 37 features related to mortality. Predictive models were developed using logistic regression, Random forest, LightGBM, XGBoost, and Multi-layer Perceptron (MLP), with assessments of feature importance, and Area under the curve (AUC).</p><p><strong>Results: </strong>A total of 1451 SICH patients were enrolled. Factors associated with mortality included lower initial GCS scores (p < 0.001), pupillary changes (P < 0.001), kidney disease (p < 0.001), and respiratory failure requiring intubation (p < 0.001). Negative correlations were observed between mortality and pupil light reflexes, as well as GCS components E(r=-0.4602), V (r=-0.4132), M(r=-0.4082). Positive correlations were identified with vasopressors (r = 0.4464), FiO2 (r = 0.3901), and sedative-hypnotic drugs (r = 0.1178). XGBoost demonstrated the best predictive performance (AUC = 0.913), outperforming LR (0.899), RF (0.905), LightGBM (0.909), and MLP (0.892). The XGBoost model, utilizing both 18 and 36 features, continues to outperform both the Acute Physiology and Chronic Health Evaluation (APACHE II) (p < 0.001) and Sequential Organ Failure Assessment (SOFA) scoring systems (p < 0.001).</p><p><strong>Conclusion: </strong>This study successfully developed an AI mortality prediction model for SICH patients, with XGBoost exhibiting superior performance. The model, incorporating 18 key features, has been integrated into clinical practice assisting clinicians in treatment decisions and communication with patients' families.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"149"},"PeriodicalIF":3.3,"publicationDate":"2025-03-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11951645/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143742451","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A SuperLearner approach for predicting diabetic kidney disease upon the initial diagnosis of T2DM in hospital.","authors":"Xiaomeng Lin, Chao Liu, Huaiyu Wang, Xiaohui Fan, Linfeng Li, Jiming Xu, Changlin Li, Yao Wang, Xudong Cai, Xin Peng","doi":"10.1186/s12911-025-02977-x","DOIUrl":"10.1186/s12911-025-02977-x","url":null,"abstract":"<p><strong>Background: </strong>Diabetic kidney disease (DKD) is a serious complication of diabetes mellitus (DM), with patients typically remaining asymptomatic until reaching an advanced stage. We aimed to develop and validate a predictive model for DKD in patients with an initial diagnosis of type 2 diabetes mellitus (T2DM) using real-world data.</p><p><strong>Methods: </strong>We retrospectively examined data from 3,291 patients (1740 men, 1551 women) newly diagnosed with T2DM at Ningbo Municipal Hospital of Traditional Chinese Medicine (2011-2023). The dataset was randomly divided into training and validation cohorts. Forty-six readily available medical characteristics at initial diagnosis of T2DM from the electronic medical records were used to develop prediction models based on linear, non-linear, and SuperLearner approaches. Model performance was evaluated using the area under the curve (AUC). SHapley Additive exPlanation (SHAP) was used to interpret the best-performing models.</p><p><strong>Results: </strong>Among 3291 participants, 563 (17.1%) were diagnosed with DKD during median follow-up of 2.53 years. The SuperLearner model exhibited the highest AUC (0.7138, 95% confidence interval: [0.673, 0.7546]) for the holdout internal validation set in predicting any DKD stage. Top-ranked features were WBC_Cnt*, Neut_Cnt, Hct, and Hb. High WBC_Cnt, low Neut_Cnt, high Hct, and low Hb levels were associated with an increased risk of DKD.</p><p><strong>Conclusions: </strong>We developed and validated a DKD risk prediction model for patients with newly diagnosed T2DM. Using routinely available clinical measurements, the SuperLearner model could predict DKD during hospital visits. Prediction accuracy and SHAP-based model interpretability may help improve early detection, targeted interventions, and prognosis of patients with DM.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"148"},"PeriodicalIF":3.3,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11948915/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143718071","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Effectiveness of various general large language models in clinical consensus and case analysis in dental implantology: a comparative study.","authors":"Yuepeng Wu, Yukang Zhang, Mei Xu, Chen Jinzhi, Yican Xue, Yuchen Zheng","doi":"10.1186/s12911-025-02972-2","DOIUrl":"10.1186/s12911-025-02972-2","url":null,"abstract":"<p><strong>Background: </strong>This study evaluates and compares ChatGPT-4.0, Gemini Pro 1.5(0801), Claude 3 Opus, and Qwen 2.0 72B in answering dental implant questions. The aim is to help doctors in underserved areas choose the best LLMs(Large Language Model) for their procedures, improving dental care accessibility and clinical decision-making.</p><p><strong>Methods: </strong>Two dental implant specialists with over twenty years of clinical experience evaluated the models. Questions were categorized into simple true/false, complex short-answer, and real-life case analyses. Performance was measured using precision, recall, and Bayesian inference-based evaluation metrics.</p><p><strong>Results: </strong>ChatGPT-4 exhibited the most stable and consistent performance on both simple and complex questions. Gemini Pro 1.5(0801)performed well on simple questions but was less stable on complex tasks. Qwen 2.0 72B provided high-quality answers for specific cases but showed variability. Claude 3 opus had the lowest performance across various metrics. Statistical analysis indicated significant differences between models in diagnostic performance but not in treatment planning.</p><p><strong>Conclusions: </strong>ChatGPT-4 is the most reliable model for handling medical questions, followed by Gemini Pro 1.5(0801). Qwen 2.0 72B shows potential but lacks consistency, and Claude 3 Opus performs poorly overall. Combining multiple models is recommended for comprehensive medical decision-making.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"147"},"PeriodicalIF":3.3,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11938642/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143718074","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Post-Anesthesia Care Unit (PACU) readiness predictions using machine learning: a comparative study of algorithms.","authors":"Shahnam Sedigh Maroufi, Maryam Soleimani Movahed, Azar Ejmalian, Maryam Sarkhosh, Ali Behmanesh","doi":"10.1186/s12911-025-02982-0","DOIUrl":"10.1186/s12911-025-02982-0","url":null,"abstract":"&lt;p&gt;&lt;strong&gt;Introduction: &lt;/strong&gt;Accurate and timely discharge from the Post-Anesthesia Care Unit (PACU) is essential to prevent postoperative complications and optimize hospital resource utilization. Premature discharge can lead to severe issues such as respiratory or cardiovascular complications, while delays can strain hospital capacity. Machine learning algorithms offer a promising solution by leveraging large amounts of patient data to predict optimal discharge times. Unlike prior studies relying on statistical models or single-algorithm methods, this research assesses multiple ML models to predict discharge readiness, comparing them against staff evaluations and the Aldrete checklist.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Methodology: &lt;/strong&gt;We conducted a cross-sectional study of 830 patients under general anesthesia from December 2023 to April 2024, collecting demographics, surgical details, and Aldrete scores. A power analysis ensured statistical robustness, targeting a 5% accuracy improvement (minimum clinically important difference, derived from Gabriel et al., 2017), with variance (SD ≈ 0.1) from pilot data, using a two-sample t-test (power = 0.8, alpha = 0.05), confirming the sample size's adequacy. Two prediction approaches were tested: discharge timing in 15-minute intervals and binary classification (within 15 min or later). Models included Random Forest (RF), Support Vector Machines (SVM), Logistic Regression (LR), Decision Tree (DT), K-Nearest Neighbors (KNN), Artificial Neural Network (ANN), and XGBoost, assessed via accuracy, precision, recall, F1 score, and AUC. Predictions were benchmarked against staff and Aldrete scores, with 99.5% confidence intervals (CIs) adjusting for multiple comparisons.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Results: &lt;/strong&gt;he RF algorithm showed high performance in both prediction approaches. In the first approach, RF achieved an AUC of 0.75 (99.5% CI: 0.70-0.80) and accuracy of 0.87 (99.5% CI: 0.83-0.91) per staff evaluations, and an AUC of 0.87 (99.5% CI: 0.83-0.91) and accuracy of 0.71 (99.5% CI: 0.66-0.76) per Aldrete scores. In the second approach, RF recorded an AUC of 0.85 (99.5% CI: 0.81-0.89) and accuracy of 0.86 (99.5% CI: 0.82-0.90) per staff evaluations, with ANN also showing strong results (AUC = 0.88, 99.5% CI: 0.84-0.92; accuracy = 0.78, 99.5% CI: 0.74-0.82). Due to overlapping CIs, differences between models were not statistically significant (P &gt;.005). According to the Aldrete checklist, RF, SVM, and ANN exhibited competitive predictive capability, with AUCs ranging from 0.80 to 0.86.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Conclusion: &lt;/strong&gt;The strong performance of Random Forest (RF) and Artificial Neural Network (ANN) models in predicting PACU discharge timing upon admission highlights their potential as effective tools for evaluating discharge readiness, as compared to staff assessments and the Aldrete checklist. This study focused on assessing these models, showing their ability to produce consistent predictions, though differences between to","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"146"},"PeriodicalIF":3.3,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11934757/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143708628","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"COVID-19 vaccination and test management for healthcare workers-development, implementation and feasibility of a custom human resources information platform at a university hospital.","authors":"Matthias Bonigut, Ana Zhelyazkova, Mathias Weber, Stefanie Geiser-Metz, Markus Geis, Bernhard Heindl, Stephan Prückner","doi":"10.1186/s12911-025-02974-0","DOIUrl":"10.1186/s12911-025-02974-0","url":null,"abstract":"<p><strong>Background: </strong>The continuously evolving legislative and reporting requirements during the COVID-19 pandemic posed the demand for establishing an efficient real-time human resources management system at the LMU University Hospital, one of the largest university hospitals in Germany. Developing a system allowing for agile real-time analysis as well as for reporting employees' COVID-19 vaccination and testing status while ensuring the security of personnel data presented several technical and managerial challenges.</p><p><strong>Methods: </strong>We designed and implemented a custom COVID-19 human resources information platform in order to fulfill the LMU University Hospital's legal requirement to report employees' vaccination and testing status. We designed the platform as an all-in-one solution for all relevant COVID-19 data, merged from five individual sources. The development process was guided by the principles of findability, accessibility, interoperability and reusability (FAIR) with particular focus on interoperability. Here, we present the platform's design, cumulative user data and discuss the feasibility of the approach including its intended and unintended outcomes.</p><p><strong>Results: </strong>The COVID-19 human resources management platform was the first solution of its kind at the LMU University Hospital, emerging from the specific need for an efficient exterior and interior mandate fulfillment. It served both for operational management purposes as well as for strategic pandemic and hospital management. The immediate dependency on data privacy and regulatory adaptations due to the evolving pandemic situation posed the necessity for regular adaptations to the platform's structure.</p><p><strong>Conclusions: </strong>The presented case reveals how data utilization requires the concurrent and proactive consideration of data security and interoperability against the background of a scalable architecture. Simultaneously, the development of such platforms needs to be open to new cases, functions and sources, thus requiring a dynamic and agile environment.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"142"},"PeriodicalIF":3.3,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11934466/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143699688","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Understanding EMS response times: a machine learning-based analysis.","authors":"Peter Hill, Jakob Lederman, Daniel Jonsson, Peter Bolin, Veronica Vicente","doi":"10.1186/s12911-025-02975-z","DOIUrl":"10.1186/s12911-025-02975-z","url":null,"abstract":"<p><strong>Background: </strong>Emergency Medical Services (EMS) response times are critical for optimizing patient outcomes, particularly in time-sensitive emergencies. This study explores the multifaceted determinants of EMS response times, leveraging machine learning (ML) techniques to identify key factors such as urgency levels, environmental conditions, and geographic variables. The findings aim to inform strategies for enhancing resource allocation and operational efficiency in EMS systems.</p><p><strong>Methods: </strong>A retrospective analysis was conducted using over one million EMS missions from Stockholm, Sweden, between 2017 and 2022. Advanced ML techniques, including Gradient Boosting models, were applied to evaluate the influence of diverse variables such as call handling times, travel times, weather patterns, and resource availability. Feature engineering was employed to extract meaningful insights, and statistical models were used to validate the relationships between key predictors and response times.</p><p><strong>Results: </strong>The study revealed a complex interplay of factors influencing EMS response times, aligning with the study's aim to deepen the understanding of these determinants. Key drivers of response time variability included weather conditions, call priority, and resource constraints. ML models, particularly Gradient Boosting, proved effective in quantifying these impacts and provided robust predictions of response times across scenarios. By providing a comprehensive evaluation of these influences, the results support the development of adaptive resource allocation models and evidence-based policies aimed at enhancing EMS efficiency and equity across all call priorities.</p><p><strong>Conclusions: </strong>This study underscores the potential of ML-driven insights to revolutionize EMS resource allocation strategies. By integrating real-time data on weather, call types, and workload, EMS systems can transition to adaptive deployment models, reducing response times and enhancing equity across priority levels. The research provides a blueprint for implementing predictive analytics in EMS operations, paving the way for evidence-based policies that improve emergency care efficiency and outcomes.</p><p><strong>Clinical trial number: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"143"},"PeriodicalIF":3.3,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11934472/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143699765","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Constructing an artificial intelligence-assisted system for the assessment of gastroesophageal valve function based on the hill classification (with video).","authors":"Jian Chen, Ganhong Wang, Kaijian Xia, Zhenni Wang, Luojie Liu, Xiaodan Xu","doi":"10.1186/s12911-025-02973-1","DOIUrl":"10.1186/s12911-025-02973-1","url":null,"abstract":"<p><strong>Objective: </strong>In the functional assessment of the esophagogastric junction (EGJ), the endoscopic Hill classification plays a pivotal role in classifying the morphology of the gastroesophageal flap valve (GEFV). This study aims to develop an artificial intelligence model for Hill classification to assist endoscopists in diagnosis, covering the entire process from model development, testing, interpretability analysis, to multi-terminal deployment.</p><p><strong>Method: </strong>The study collected four datasets, comprising a total of 1143 GEFV images and 17 gastroscopic videos, covering Hill grades I, II, III, and IV. The images were preprocessed and enhanced, followed by transfer learning using a pretrained model based on CNN and Transformer architectures. The model training utilized a cross-entropy loss function, combined with the Adam optimizer, and implemented a learning rate scheduling strategy. When assessing model performance, metrics such as accuracy, precision, recall, and F1 score were considered, and the diagnostic accuracy of the AI model was compared with that of endoscopists using McNemar's test, with a p-value < 0.05 indicating statistical significance. To enhance model transparency, various interpretability analysis techniques were used, including t-SNE, Grad-CAM, and SHAP. Finally, the model was converted into ONNX format and deployed on multiple device terminals.</p><p><strong>Results: </strong>Compared through performance metrics, the EfficientNet-Hill model surpassed other CNN and Transformer models, achieving an accuracy of 83.32% on the external test set, slightly lower than senior endoscopists (86.51%) but higher than junior endoscopists (75.82%). McNemar's test showed a significant difference in classification performance between the model and junior endoscopists (p < 0.05), but no significant difference between the model and senior endoscopists (p ≥ 0.05). Additionally, the model reached precision, recall, and F1 scores of 84.81%, 83.32%, and 83.95%, respectively. Despite its overall excellent performance, there were still misclassifications. Through interpretability analysis, key areas of model decision-making and reasons for misclassification were identified. Finally, the model achieved real-time automatic Hill classification at over 50fps on multiple platforms.</p><p><strong>Conclusion: </strong>By employing deep learning to construct the EfficientNet-Hill AI model, automated Hill classification of GEFV morphology was achieved, aiding endoscopists in improving diagnostic efficiency and accuracy in endoscopic grading, and facilitating the integration of Hill classification into routine endoscopic reports and GERD assessments.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"144"},"PeriodicalIF":3.3,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11934607/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143699677","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Development and validation of a machine learning model to predict hemostatic intervention in patients with acute upper gastrointestinal bleeding.","authors":"Kajornvit Raghareutai, Watcharaporn Tanchotsrinon, Onuma Sattayalertyanyong, Uayporn Kaosombatwattana","doi":"10.1186/s12911-025-02969-x","DOIUrl":"10.1186/s12911-025-02969-x","url":null,"abstract":"<p><strong>Background: </strong>Acute upper gastrointestinal bleeding (UGIB) is common in clinical practice and has a wide range of severity. Along with medical therapy, endoscopic intervention is the mainstay treatment for hemostasis in high-risk rebleeding lesions. Predicting the need for endoscopic intervention would be beneficial in resource-limited areas for selective referral to an endoscopic center. The proposed risk stratification scores had limited accuracy. We developed a machine learning model to predict the need for endoscopic intervention in patients with acute UGIB.</p><p><strong>Methods: </strong>A prospectively collected database of UGIB patients from 2011 to 2020 was retrospectively reviewed. Patients older than 18 years diagnosed with UGIB who underwent endoscopy were included. Data comprised demographic characteristics, clinical presentation, and laboratory parameters. The cleaned data was used for model development and validation in Python. We conducted 80%-20% split sample training and test sets. The training set was used for supervised learning of 15 models using a stratified 5-fold cross-validation process. The model with the highest AUROC was then internally validated with the test set to evaluate performance.</p><p><strong>Results: </strong>Of 1389 patients, 615 (44.3%) of the cohorts received the endoscopic intervention (293 variceal- and 336 nonvariceal-bleeding interventions). Eighteen features, including demographic characteristics, clinical presentation, and laboratory parameters, were selected as input for 15 machine learning models. The result revealed that the linear discriminant analysis model could achieve the highest AUROC of 0.74 to predict endoscopic intervention. The model was validated with the test set, in which the AUROC was increased from 0.74 to 0.81. Finally, the model was deployed as a web application by Streamlit.</p><p><strong>Conclusions: </strong>Our machine learning model can identify patients with acute UGIB who need endoscopic intervention with good performance. This may help primary care physicians prioritize patients who need referrals and optimize resource allocation in resource-limited areas. Further development and identification of more specific features might improve prediction performance.</p><p><strong>Trial registration: </strong>None (Retrospective cohort study) PATIENT & PUBLIC INVOLVEMENT: None.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"25 1","pages":"145"},"PeriodicalIF":3.3,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11934503/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143699690","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}