A Dual-Path Machine Learning Strategy for Pulmonary Nodule Risk Stratification

This archive contains de-identified analysis datasets, locked fitted machine learning models, and fully reproducible analysis code . All model design choices (feature preprocessing, algorithm selection, and hyperparameter tuning) were made exclusively using the training cohort via nested cross-validation. The external validation cohort was evaluated only once after the entire pipeline was locked to ensure no data leakage. The archive includes: (1) training, internal validation, and external validation datasets with 10 pre-specified predictors; (2) four final fitted model pipelines; (3) core Python scripts to reproduce all primary analyses and numerical results; (4) raw result files from the nested cross-validation and external evaluation. Raw clinical source records are not publicly available to protect patient privacy.