Cross-Software Radiomic Feature Robustness Assessed by Hierarchical Clustering and Composite Index Analysis: A Multi-Cancer Study on Colorectal and Liver Lesions.

Background: Radiomic feature robustness is a key prerequisite for the reproducibility and clinical translation of imaging biomarkers. Variability across software platforms can significantly affect feature consistency, compromising predictive modeling reliability. This study aimed to develop and validate a hierarchical clustering-based workflow for evaluating radiomic feature robustness within and across software platforms, identifying stable and reproducible features suitable for clinical applications. Methods: A multi-cancer CT dataset including 97 lesions from 71 patients, comprising primary colorectal cancer (CRC), colorectal liver metastases, and hepatocellular carcinoma (HCC), was analyzed. Radiomic features were extracted using two IBSI-compliant platforms (MM Radiomics of syngo.via Frontier and 3D Slicer with PyRadiomics). Intra-software reliability was assessed through the intraclass correlation coefficient ICC(A,1), while cross-software stability was evaluated using hierarchical clustering validated by the Adjusted Rand Index (ARI). A Composite Index (CI) integrating correlation, distributional similarity, and mean fractional ratio quantified inter-platform feature robustness. Results: Over 95% of radiomic features demonstrated good-to-excellent intra-software reliability. Several clustering configurations achieved ARI = 1.0, confirming strong cross-platform concordance. The most robust and recurrent features were predominantly wavelet-derived descriptors and first-order statistics, particularly cluster shade (GLCM-based) and mean intensity-related features. Conclusions: The proposed multi-stage framework effectively identifies stable, non-redundant, and transferable radiomic features across IBSI-compliant software platforms. These findings provide a methodological foundation for cross-platform harmonization and enhance the reproducibility of radiomic biomarkers in oncologic imaging.