Multiregional radiomics feature fusion for predicting lymph node metastasis in nonsmall cell lung cancer

This study developed radiomics-based models to predict lymph node metastasis in non-small cell lung cancer using data from 186 patients (150 non-metastatic, 36 metastatic). To capture spatial heterogeneity, we employed a multi-regional fusion radiomics approach, integrating features from intratumoral, habitat, and peritumoral regions. Our results showed that combining features from all three habitat regions with those from the 2mm peritumoral zone provided the highest predictive power (AUC=0.896, ACC=0.893) using an SVM classifier. Incorporating clinical data led to slight performance enhancements (AUC=0.907, ACC=0.893), reinforcing the robustness of the multi-regional fusion approach. These findings underscore the importance of multi-regional feature integration in radiomics analysis, demonstrating its potential to enhance predictive accuracy and support clinical decision-making.