Machine learning model integrating radiomics and clinical features for predicting postoperative bleeding after...

Area of Science:

• Medical imaging informatics
• Artificial intelligence in medicine
• Urology

Background:

• Radiomics and machine learning are crucial for developing clinical prediction models.
• Predicting postoperative bleeding after percutaneous nephrolithotomy (PCNL) is essential for patient management.

Purpose of the Study:

• To integrate radiomic features with clinical variables.
• To develop and evaluate machine learning models for predicting postoperative bleeding risk following PCNL.

Main Methods:

• Retrospective analysis of 151 PCNL patients.
• Identification of clinical variables and radiomic features using univariate analysis and LASSO regression.
• Development of prediction models using Logistic Regression, Random Forest (RF), and Support Vector Machine (SVM); model performance assessed via AUC and calibration; clinical utility evaluated using Decision Curve Analysis (DCA); feature importance elucidated using SHAP analysis.

Main Results:

• The postoperative bleeding rate was 31.1%.
• Logistic Regression and RF models achieved 75.6% accuracy with AUCs of 0.760 and 0.740, respectively.
• Decision curve analysis indicated enhanced clinical benefits for the logistic regression model; SHAP analysis provided feature importance insights.

Conclusions:

• Machine learning models demonstrate strong performance in predicting postoperative bleeding after PCNL.
• These models can serve as a foundation for surgical risk stratification.
• Personalized postoperative monitoring plans can be developed based on these predictive capabilities.