Radiomics-driven tumor prognosis prediction across imaging modalities : advances in sampling, feature selection, and multi-omics integration

Abstract

Radiomics has shown remarkable potential in predicting cancer prognosis by noninvasive and quantitative analysis of tumors through medical imaging. This review summarizes recent advances in the use of radiomics across various cancer types and imaging modalities, including computed tomography (CT), magnetic resonance imaging (MRI), ultrasound, positron emission tomography (PET), and interventional radiology. Innovative sampling methods, including deep learning-based segmentation, multiregional analysis, and adaptive region of interest (ROI) methods, have contributed to improved model performance. The review examines various feature selection approaches, including least absolute shrinkage and selection operator (LASSO), minimum redundancy maximum relevance (mRMR), and ensemble methods, highlighting their roles in enhancing model robustness. The integration of radiomics with multi-omics data has further boosted predictive accuracy and enriched biological interpretability. Despite these advancements, challenges remain in terms of reproducibility, workflow standardization, clinical validation and acceptance. Future research should prioritize multicenter collaborations, methodological coordination, and clinical translation to fully unlock the prognostic potential of radiomics in oncology.