Machine learning-based prognostic model for bloodstream infections in hematological malignancies using Th1/Th2 cytokines

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

Published: March 26, 2025

Bloodstream infection (BSI) is a significant cause of mortality in patients with hematologic malignancies(HMs), particularly amid rising antibiotic resistance. This study aimed to analyze pathogen distribution, drug-resistance patterns and develop novel predictive model for 30-day HM BSIs. A retrospective analysis 231 positive blood cultures was conducted. Logistic regression identified risk factors mortality. Th1/Th2 cytokines were collected at BSI onset, LASSO restricted cubic spline used refine predictors. Seven machine learning(ML) algorithm (XGBoost, Regression, LightGBM, RandomForest, AdaBoost, GBDT GNB) trained using 10-fold cross-validation performance evaluated the ROC, calibration plots, decision learning curves Shapley Additive Explanations (SHAP) analysis. The developed by integrating clinical features, aiming enhance accuracy prediction. Among cohort, acute myeloid leukemia (38%) most common HM, while gram negative bacteria (64%) predominant pathogens causing BSI. Age, polymicrobial BSI, IL-4, IL-6 AST levels predictors Regression achieved AUCs 0.802, 0.792, 0.822 training, validation, test cohorts, respectively, strong benefit shown curves. SHAP highlighted IL-4 as key introduces ML-based features predict BSIs, demonstrating applicability.

Language: Английский

---

From pixels to prognosis: leveraging radiomics and machine learning to predict IDH1 genotype in gliomas

Neurosurgical Review, Journal Year: 2025, Volume and Issue: 48(1)

Published: April 29, 2025

Language: Английский

---

Predicting the efficacy of bevacizumab on peritumoral edema based on imaging features and machine learning

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 8, 2025

This study proposes a novel approach to predict the efficacy of bevacizumab (BEV) in treating peritumoral edema metastatic brain tumor patients by integrating advanced machine learning (ML) techniques with comprehensive imaging and clinical data. A retrospective analysis was performed on 300 who received BEV treatment from September 2013 January 2024. The dataset incorporated 13 predictive features: 8 variables 5 radiological variables. divided into training set (70%) test (30%) using stratified sampling. Data preprocessing carried out through methods such as handling missing values MICE method, detecting adjusting outliers, feature scaling. Four algorithms, namely Random Forest (RF), Logistic Regression, Gradient Boosting Tree, Naive Bayes, were selected construct binary classification models. tenfold cross-validation strategy implemented during training, like regularization, hyperparameter optimization, oversampling used mitigate overfitting. RF model demonstrated superior performance, achieving an accuracy 0.89, precision 0.94, F1-score 0.92, both AUC-ROC AUC-PR reaching 0.91. Feature importance consistently identified volume most significant predictor, followed index, patient age, volume. Traditional multivariate logistic regression corroborated these findings, confirming that index independent predictors (p < 0.01). Our results highlight potential ML-driven models optimizing selection, reducing unnecessary risks, improving decision-making neuro-oncology.

Language: Английский

---

Deep learning-driven modality imputation and subregion segmentation to enhance high-grade glioma grading

BMC Medical Informatics and Decision Making, Journal Year: 2025, Volume and Issue: 25(1)

Published: May 30, 2025

Language: Английский

---