Phantom study identifies stable low-dose CT radiomic features that improve lung cancer screening model consistency

In a phantom study, models for lung nodule malignancy assessment using only stable radiomic features achieved an AUC of 0.995 versus 0.945 with unstable features, and similarly outperformed for growth prediction (AUC 0.799 vs 0.610). Inter-scanner and tube current variability mo…

• A chest phantom with eight simulated nodules was scanned on five CT scanners at varying tube voltages (100-120 kVp) and tube currents (20-60 mA·s) to evaluate inter- and intra-scanner feature stability using intraclass correlation coefficient.
• When tested on two independent lung cancer screening datasets, models built with representative stable features showed smaller performance gaps between training and test sets, suggesting improved generalizability.
• Limitation: Stability thresholds were derived from a single anthropomorphic chest phantom; generalizability to the full range of human nodule heterogeneity remains unexplored.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.