Radiomics Feature-level Heterogeneity Improves Bone Metastasis Risk Stratification Over Habitat Analysis in Breast Cancer

Feature-level intratumoral heterogeneity (ITH1) radiomics outperformed habitat-based modeling (ITH2) for bone metastasis-free survival in breast cancer. An integrated model combining ITH1, whole-tumor radiomics, and vision transformer deep features achieved C-index 0.878-0.908 a…

• ITH1 (feature-level radiomics complexity) showed greater robustness and generalizability than ITH2 (habitat-based supervoxel) for bone metastasis prediction.
• Integration of ITH1 with global tumor radiomics and ViT-derived deep learning features significantly improved BMFS prediction.
• SHAP analysis revealed the ITH1 score as the most prominent contributor to risk prediction.

Automated summary

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