To assess the stability of radiomic features derived from lung nodules under low-dose CT lung cancer screening conditions and to evaluate the influence of feature stability on model performance.
Approach:
Phantom Study: A chest phantom with eight simulated lung nodules was scanned on five CT scanners using varying tube voltages and currents to simulate inter-scanner and intra-scanner variability.
Feature Extraction: Nodule radiomic features were extracted, and their stability was evaluated using the intraclass correlation coefficient.
Model Comparison: Models using stable features were compared to those using unstable features in two independent lung cancer screening datasets for nodule malignancy assessment and growth prediction.
Key Findings:
Inter-scanner variability had a greater impact on feature stability than intra-scanner variability.
Tube current exerted more influence on feature stability than tube voltage.
Models based on representative stable features achieved better performance in malignancy assessment (AUC 0.995 vs 0.945) and growth prediction (AUC 0.799 vs 0.610).
Models using stable features showed smaller performance differences between training and test sets.
Interpretation:
Radiomic features with high stability under lung cancer screening conditions were associated with improved performance consistency in independent screening datasets.
Limitations:
The study utilized a phantom model, which may not fully replicate real patient conditions.
The analysis was limited to specific CT acquisition parameters and may not generalize to all clinical settings.
Conclusion:
Stability-informed feature selection may contribute to more reliable radiomics applications in lung cancer screening.
