ABSTRACT Accurate prediction of the freeze-thaw resistance of fiber-reinforced concrete (FRC) is crucial for optimizing its mix design and evaluating long-term durability. To enhance the accuracy of predicting FRC durability under freeze-thaw cycles, a dataset comprising 1,969 sets of FRC durability data was constructed. A machine learning hyperparameter optimization method based on the improved golden jackal optimization algorithm (PEGJO) was proposed for predicting FRC freeze resistance. The K-means++ algorithm was applied to cluster the data, balancing the training and testing sets. The PEGJO algorithm performs hyperparameter optimization for five machine learning algorithms: Support Vector Regression (SVR), Random Forest (RF), Kernel Ridge Regression (KRR), CatBoost, and XGBoost. Results indicate that the PEGJO-optimized XGBoost model achieved the highest predictive accuracy, with test set R2 values of 0.9581, 0.9585, and 0.9699 for compressive strength, relative dynamic modulus, and mass loss rate, respectively. Additionally, key characteristics influencing the durability of FRC were investigated using interpretable methodologies. To enhance practical applicability, an interactive graphical user interface based on the Tkinter framework was developed. This provides valuable guidance for FRC mix design, thereby advancing the field of performance evaluation for FRC.

To develop and validate a prediction model combining Gd-EOB-DTPA-enhanced magnetic resonance imaging (MRI) diffusion-weighted imaging (DWI) parameters and clinicopathologic features for preoperative prediction of microvascular invasion (MVI) in hepatocellular carcinoma (HCC). This study applied a retrospective method to collect preoperative MRI imaging data of patients with HCC from January 2021 to June 2025. All 279 patients (mean age of 58, M:F=203:76, 195 cases in training set and 84 cases in validation set) underwent MRI by Gd-EOB-DTPA and received DWI imaging scan before surgery. The MRI imaging features were observed, and the apparent diffusion coefficient (ADC) of the tumor solid region and tumor-to-liver parenchyma relative intensity ratio (RIR) were measured. LASSO logistic regression algorithm and multivariate analysis method were conducted to analyze the correlation between preoperative MRI imaging features and MVI. The prediction model was established, and the efficiency evaluation of the model was performed. Among 279 patients, 66 cases (23.66%) were MVI-positive. The study subjects were assigned to a training set (195 cases) and a validation set (84 cases) at a 7:3 ratio. Twelve variables were selected by LASSO regression. Multivariate analysis identified RIR transitional phase (OR=0.21), tumor size (OR=4.52), and ADC (OR=0.63) as independent predictors of MVI (P=0.029, <0.001, 0.032). The model achieved an AUC of 0.85 in the training set and 0.83 in the validation set. The negative predictive value (NPV) was 0.94 and 0.90 in the training and validation sets, respectively, while the positive predictive value (PPV) was limited to 0.49 and 0.40. Calibration curve and decision curve analyses demonstrated good consistency and clinical utility. On the basis of clinicopathologic features, MRI imaging features, and DWI parameters, this study preliminarily constructs a prediction model for positive MVI risk in HCC patients. The model exhibits good discrimination and a high NPV for effectively ruling out MVI, but its limited PPV warrants cautious interpretation of positive predictions due to a high false-positive rate.

Development and validation of a nomogram prediction model for prolonged length of hospital stay in patients with AECOPD.

To identify risk factors, mechanisms, and develop a validated predictive model for platelet transfusion refractoriness (PTR) in pediatric oncology to optimize transfusion outcomes. This retrospective cohort analyzed pediatric oncology patients receiving platelet transfusions, stratified by response (effective vs. refractory). Statistically significant indicators from t-tests underwent multivariable logistic regression to determine independent PTR predictors [platelet count (PLT), prothrombin time (PT), activated partial thromboplastin time (APTT), D-dimer (D-D)]. A combined predictor derived via logistic regression was evaluated using ROC analysis. A Nomogram integrating PLT, hemoglobin (Hb), PT, APTT, and D-D was developed, with calibration (calibration curves) and clinical utility (decision curve analysis, DCA) rigorously validated. Multivariable logistic regression revealed an inverse association between elevated PLT and PTR risk (OR = 1.396), while elevated PT, APTT, and D-D were directly associated with increased PTR risk (OR: 0.794, 0.943, and < 0.001, respectively). The combined predictor demonstrated excellent discriminatory power (AUC = 0.962). The nomogram model exhibited high calibration accuracy in both training and validation sets, as evidenced by well-fitted calibration curves. DCA confirmed superior net benefit compared to alternative strategies. High AUC values (training set: 0.9723; validation set: 0.9639) indicated robust discrimination of PTR risk. PLT, PT, APTT, and D-dimer are key modulators of PTR risk in pediatric oncology. The developed nomogram model demonstrates favorable calibration, strong discriminatory ability, and significant clinical utility. It enables quantitative PTR risk stratification and guides pathway-targeted interventions, providing a precision tool for optimizing platelet transfusion management in this vulnerable population.

Capturing "source" and "evolution": A 5-year recurrence risk prediction model for nasopharyngeal carcinoma based on integrated radiomic spatial heterogeneity of primary tumor and lymph nodes.

Frailty is increasingly recognized as a predictor of poor surgical outcomes, yet its preoperative assessment in patients with non-small-cell lung cancer (NSCLC) remains challenging. This study aimed to develop and validate a high-performance predictive model for assessing frailty risk using routinely available clinical parameters and machine learning (ML) techniques. This single-center, cross-sectional study enrolled 489 preoperative patients with NSCLC hospitalized from April to October 2024. Participants were randomly divided into training (n=342) and validation (n=147) sets. Frailty was assessed using the FRAIL scale. We developed a logistic regression-based nomogram and compared it with six ML models. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. Frailty/pre-frailty prevalence was 36.1%. Logistic regression identified age, body mass index, comorbidity grade, fatigue, walking difficulty, single-breath diffusing capacity of the lung for carbon monoxide and the triglyceride-glucose (TyG) index as independent predictors. Although the nomogram achieved an AUC of 0.77 (95% confidence interval 0.69-0.85) in the validation set, the Light Gradient Boosting Machine (LGBM) model demonstrated superior discrimination, with an AUC of 0.965 in the training set and 0.807 in the validation set. Feature importance analysis highlighted the TyG index, comorbidity grade, and maximal voluntary ventilation as top predictors. This study developed a robust frailty risk prediction framework. The integration of ML algorithms with objective physiological markers (TyG index and respiratory reserve) significantly enhanced predictive accuracy over traditional methods, providing a reliable tool for preoperative risk stratification in patients with NSCLC.

Development and evaluation of a predictive biomarker model based on the niacin flush pathway for differentiating schizophrenia and bipolar disorder.

Reflux esophagitis (RE) is common among individuals with obesity and candidates for Metabolic Bariatric Surgery (MBS). Therefore, this study aimed to develop and internally evaluate a clinical nomogram to estimate the risk of RE in candidates for MBS. A total of 694 patients scheduled for MBS between July 2020 and December 2023 were retrospectively enrolled and randomly split into a training set and an internal test set at a 7:3 ratio. All patients underwent preoperative upper gastrointestinal endoscopy to assess RE status. The least absolute shrinkage and selection operator (LASSO) regression was applied to identify candidate predictors. Variables with non-zero coefficients were subsequently entered into multivariable logistic regression analysis, and independent factors were used to construct the nomogram. Model discrimination, calibration, and clinical utility were evaluated using the concordance index (C-index), calibration curves, and decision curve analysis (DCA), respectively. The overall prevalence of RE was 20.03%. Multivariable logistic regression analysis identified age, BMI, HDL-C, sex, and Helicobacter pylori (H. pylori) infection status as independent factors associated with RE. The nomogram constructed based on these variables demonstrated acceptable predictive performance in both sets. The area under the curve (AUC) was 0.796 in the training set and 0.79 in the internal test set. After bootstrap internal validation with 1,000 resamples in the training set, the bias-corrected C-index was 0.786. Calibration curves showed acceptable agreement between predicted probabilities and the observed incidence of RE. We developed a concise nomogram incorporating preoperative clinical variables to estimate the risk of RE in candidates for MBS. The model demonstrated acceptable discrimination and calibration based on internal validation; however, its clinical applicability requires further external validation.

Predictive value of early blood glucose trajectory for poor prognosis in ARDS patients.

ARDSMLpred: A machine learning model for predicting SCAP-associated ARDS based on MIMIC-IV database.

Identification of potential biomarkers for Polygonum multiflorum-induced liver injury in a clinical cohort: integrating machine learning of metabolomics with transcriptomic profiling.

Integrating machine learning and SHAP analysis to identify the risk of cognitive frailty in older adults with chronic heart failure.

3D deep-learning radiomics from MR-T2WI for predicting placenta accreta spectrum disorders: A multicenter study.

The application of a nomogram model integrating clinical factors and multimodal MRI radiomics features for predicting cervical lymph node metastasis for patients with oral tongue squamous cell carcinoma in different tumor stages.

MUSIOMICS: A multi-region radiomics framework that outperforms single-region analysis in classifying malignant pulmonary nodules.

Estimation of pharmaceutical solubility values under various conditions has been a subject of great interest and is useful for supercritical processing in enhancing solubility values. In this study, we have accurately estimated the solubility of drugs in supercritical solvent (carbon dioxide) under various pressure and temperature levels. Artificial intelligence models (AIMs) including XGBoost, Gradient Boosting, and Random Forest were developed to calculate the solubility of drugs in supercritical CO2. A comprehensive dataset (1619 points, 58 drugs, with a solubility range of 10⁻7-10⁻3 mol fraction) comprising operational conditions (temperature: 308-348.2 K, pressure: 80-400 bar) and physicochemical properties of drugs was collected and considered in model development and optimization. Logarithmic transformation of solubility values was employed to enhance predictive robustness. Comparative analysis of the developed AIMs revealed that the XGBoost and Gradient Boosting models exhibited the most reliable performance. Solubility values were log‑transformed before modeling, and performance metrics (RMSE, MAE, AARD%) were computed on the back‑transformed solubility values. For the optimized XGBoost, the values of R2, RMSE, MAE, and AARD% were determined as 0.998, 0.0259, 0.0134, and 4.12% for training; 0.9832, 0.0210, 0.0114, and 3.60% for validation; and 0.9864, 0.0352, 0.0171, and 9.51% for testing, respectively. Furthermore, the obtained R2 (training set) for Gradient Boosting and Random Forest are equal to 0.997 and 0.983, respectively. In addition, sensitivity analysis indicated that pressure and temperature were the most influential variables on model performance in solubility prediction.

Open-set deep learning enabled LIBS sensor for label-free and on-site identification of unknown pathogen powders.

This study aimed to develop and validate a non-invasive, multimodal radiomics model based on preoperative 1⁸F-FDG PET/CT to predict CLDN18.2 expression in gastric adenocarcinoma (GAC), addressing the limitations of intratumoral heterogeneity and invasiveness associated with endoscopic biopsies. This retrospective study enrolled 291 patients with pathologically confirmed GAC who underwent preoperative 1⁸F-FDG PET/CT. The cohort was randomly divided into a training set (n = 204) and an independent validation set (n = 87). High-dimensional radiomic features were extracted from PET and CT images. Feature selection was performed using the minimum redundancy maximum relevance (mRMR) algorithm and LASSO regression. A radiomics signature (Rad-score) was constructed using XGBoost and integrated with clinical variables. Among five machine learning algorithms evaluated, AdaBoost was identified as the optimal model. Performance was assessed via receiver operating characteristic (ROC) analysis, and interpretability was visualized using SHapley Additive exPlanations (SHAP). CLDN18.2-positive tumors exhibited a distinct hypometabolic phenotype, characterized by significantly lower SUVmax (P = 0.012) and SUVmean (P < 0.001) compared to negative tumors. The combined multimodal model demonstrated superior discrimination, achieving an AUC of 0.926 (95% CI: 0.891-0.961) in the training cohort and 0.765 (95% CI: 0.665-0.865) in the validation cohort, consistently outperforming single-modality models. SHAP analysis highlighted PET-derived radiomic features and serum CEA as the most influential predictors, confirming the inverse correlation between metabolic activity and CLDN18.2 expression. The developed 1⁸F-FDG PET/CT-based multimodal radiomics model serves as an effective, non-invasive tool for predicting CLDN18.2 expression in GAC. The study validates the hypometabolic nature of CLDN18.2-positive tumors, suggesting this radiomics approach can effectively complement traditional biopsy methods for patient stratification.

Carbohydrate antigen 19-9 (CA19-9) has limited utility in the ∼10% of patients with pancreatic ductal adenocarcinoma (PDAC) who are Lewis-negative CA19-9 "nonproducers" (FUT3-null). This study aims to define the prognosis of FUT3-null PDAC and establish a CA19-9 cutoff that can help identify this subgroup in clinical practice. Germline whole-exome sequencing was performed in a multicenter cohort of 615 patients with PDAC to determine FUT2/FUT3 genotypes. Receiver operating characteristic (ROC) analysis identified the optimal CA19-9 cutoff for FUT3-null status. Maximally selected rank statistics were used to derive CA19-9 cutoffs that best stratified overall survival (OS) in a training set (n = 307) and were tested in a validation set (n = 308). FUT3-null patients had similar demographics except for a lower median baseline CA19-9 level (2.4 vs. 496 U/mL in FUT3-intact patients) and a comparable median OS (13.5 months) to FUT3-intact patients with CA19-9 >200 U/mL (12.9 months). ROC analysis identified 7 U/mL as the optimal CA19-9 cutoff for FUT3-null (positive predictive value 95.1%; accuracy 87.9%). Using CA19-9 cutoffs of 7 and 200 U/mL (without genotyping), patients were stratified into four prognostic groups: >7 to 37 U/mL (median OS, 23.2 months), >37 to 200 U/mL (median OS, 22 months), >200 U/mL (median OS, 12.8 months), and ≤7 U/mL (likely FUT3-null; median OS, 13.5 months; P < 0.001). Findings were consistent in multivariable Cox models. CA19-9 nonproducers whose outcomes are comparable with those with CA19-9 >200 U/mL represent a high-risk subgroup. CA19-9 ≤7 U/mL is a practical surrogate for identifying these patients. When genotyping is unavailable, a dual-threshold model (≤7 and >200 U/mL) improves risk stratification.

Radiomics-Based Classification of Pathological Patterns in Common Carotid Artery Wall.