Risk stratification for early-stage NSCLC progression: a federated learning framework with large-small model synergy
Zijun Huang1, Bao Feng2,3, Yehang Chen2
1School of Electronic Engineering and Automation, Guilin University of Electronic Technology, Guilin, China.
Frontiers in Oncology
|January 1, 2026
Summary
A new deep learning model, FedCPI, accurately predicts non-small cell lung cancer (NSCLC) progression. This tool aids clinical decisions and reduces overtreatment for patients with early-stage lung cancer.
Area of Science:
- Oncology
- Medical Informatics
- Artificial Intelligence
Background:
- Accurate non-small cell lung cancer (NSCLC) progression prediction is vital for patient management and prognosis.
- Overtreatment and undertreatment pose significant risks, highlighting the need for precise risk stratification.
- Current methods may lack the accuracy required for optimal clinical decision-making in early-stage NSCLC.
Purpose of the Study:
- To develop a precise risk stratification system for non-small cell lung cancer (NSCLC).
- To improve prediction accuracy and enable stratified management for NSCLC patients.
- To validate the proposed framework's versatility and robustness across multiple tasks.
Main Methods:
- A retrospective study of 926 patients with resected stage I-IIA NSCLC from four centers.
- Development of a multi-center intelligent risk stratification model: Federated cross-scale Common-Personal-Interactive learning (FedCPI).
- Evaluation using AUC, accuracy, sensitivity, specificity, PPV, NPV, five-fold cross-validation, and decision curve analysis (DCA); cross-task validation for gastric and endometrial cancers.
Main Results:
- FedCPI outperformed clinical stratification and federated learning baselines (DeLong test, p < 0.05).
- Achieved AUCs up to 0.9255 and ACCs up to 0.8909 in early-stage NSCLC, with significant gains over competing models.
- Demonstrated outstanding performance in predicting gastric cancer recurrence and endometrial cancer infiltration, validating the methodology's effectiveness.
Conclusions:
- The deep learning-based FedCPI framework offers a non-invasive, accurate, and reliable tool for early-stage lung cancer risk stratification.
- The methodology showed excellent performance in independent validation for gastric and endometrial cancer clinical tasks.
- Improved diagnostic precision through FedCPI has the potential to optimize clinical decision-making and reduce overtreatment burdens.
