Low-Dose CT-Based Radiomic Model to Identify Malignant Lung Nodules

Lung cancer screening using low-dose computed tomography (LDCT) faces challenges in nodule classification and radiologist burden. This study introduces a novel approach leveraging degraded standard-dose CT (SDCT) images and develops an explainable radiomics-based model to improve lung nodule classification. Using data from LIDC-IDRI public dataset, our pipeline extracts and preprocesses 851 radiomic features per nodule, ensuring stability, discriminant characteristics, and non-redundancy. To address dataset imbalance, we applied data augmentation and random undersampling techniques. Varying the number of features selected by gradient-boosted decision tree across multiple classifiers, we identified an optimal combination of 15 features with random forest classifier, achieving balanced accuracy, sensitivity, specificity and AUC-ROC scores of 0.82 ± 0.04, 0.81 ± 0.08, 0.83 ± 0.05, 0.88 ± 0.05, respectively. Explainability analysis using SHAP highlights features influencing classification, including major axis length, sphericity, and wavelet-LHH-ngtdm-Busyness. In conclusion, our study demonstrates the potential of integrating degraded SDCT and radiomics to improve lung nodule classification.