Explainable PET-Based Habitat and Peritumoral Machine Learning Model for Predicting Progression-free Survival in
Bei-Hui Xue1, Shuang-Li Chen2, Jun-Ping Lan2
1Division of Pulmonary Medicine, the First Affiliated Hospital of Wenzhou Medical University, Key Laboratory of Heart and Lung, Wenzhou, Zhejiang, China (B.H.X., J.P.L.); Department of Radiology, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China (B.H.X., S.L.C., J.G.X., X.W.Z.).
Academic Radiology
|January 5, 2025
Summary
Machine learning models using PET-habitat and peritumoral radiomics predict progression-free survival in early-stage non-small cell lung cancer (NSCLC). These models effectively identify high-risk patients, aiding in personalized treatment strategies.
Area of Science:
- Oncology
- Radiology
- Artificial Intelligence
Background:
- Early-stage non-small cell lung cancer (NSCLC) diagnosis and prognosis remain challenging.
- Accurate prediction of progression-free survival (PFS) is crucial for guiding treatment decisions in clinical stage IA NSCLC.
- Novel biomarkers are needed to improve risk stratification in early-stage NSCLC.
Purpose of the Study:
- To develop and validate machine learning (ML) models for predicting PFS in patients with clinical stage IA pure-solid NSCLC.
- To utilize positron emission tomography (PET)-derived radiomic features from tumor habitats and the peritumoral microenvironment.
- To integrate radiomic signatures with clinical variables for enhanced predictive accuracy.
Main Methods:
- Radiomic features were extracted from intratumoral, peritumoral, and habitat regions on PET scans of 234 NSCLC patients.
- Univariate and multivariate logistic regression identified significant clinical variables.
- A radiomics nomogram was constructed by combining radiomics signatures with clinical variables.
- Kaplan-Meier analysis and Shapley Additive Explanations (SHAP) were used for prognostic evaluation and model interpretation.
Main Results:
- The combined model, incorporating peritumoral (5 mm) and habitat radiomics features with clinical variables, demonstrated strong performance.
- The model achieved an area under the curve (AUC) of 0.905 in the training set and 0.875 in the internal validation set.
- The radiomics signature was significantly associated with PFS, effectively distinguishing between high- and low-risk patient groups (log-rank P < 0.001).
Conclusions:
- Habitat and peritumoral radiomics signatures are independent biomarkers for predicting PFS in early-stage NSCLC.
- The developed ML models effectively stratify survival risk in patients with clinical stage IA pure-solid NSCLC.
- These findings support the clinical utility of radiomics for personalized risk assessment in early-stage lung cancer.
