Developing of risk models for small solid and subsolid pulmonary nodules based on clinical and quantitative radiomics...

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Accurate evaluation of pulmonary nodule malignancy is crucial for effective lung cancer management.
• Small solid and subsolid pulmonary nodules pose diagnostic challenges.

Purpose of the Study:

• Develop and evaluate risk models for pulmonary nodule malignancy.
• Incorporate clinical and quantitative radiomics features for improved prediction.
• Assess model performance for different nodule types and sizes.

Main Methods:

• Enrolled patients with 5-20 mm pulmonary nodules confirmed pathologically.
• Extracted 1218 radiomics features from thoracic high-resolution computed tomography (HRCT) images.
• Utilized LASSO for feature selection and compared Random Forest, XGBOOST, SVM, and logistic models against the Mayo model.

Main Results:

• Developed risk models for all nodules, nodules ≤10 mm, solid nodules, and subsolid nodules (SSNs).
• Achieved high performance metrics, e.g., accuracy up to 0.86 and AUC up to 0.91 for all nodules (XGBOOST).
• Clinical-radiomics models outperformed the Mayo model for most nodule groups, except for SSNs.

Conclusions:

• Clinical and radiomics-based models effectively assess malignancy risk in small pulmonary nodules.
• These models are valuable for managing both solid and subsolid nodules, including those ≤10 mm.
• The developed models offer a promising tool for lung cancer diagnosis and management.