AI- and Deep Learning-Based Radiomics: A State-of-the-Art Review.
Authors
Abstract
Radiomics has emerged over the past decade as a powerful image-based omics approach in oncology. Handcrafted radiomic features can reveal subtle imaging characteristics associated with tumor biology and clinical outcomes. In parallel, deep learning (DL) has transformed medical image analysis by learning complex feature representations directly from imaging data. Recent research has increasingly focused on hybrid models that integrate radiomic and DL-derived features to combine the interpretability of radiomics with the representational capacity of DL. Methodological advances have also been made across the radiomics pipeline, including preprocessing, segmentation, feature selection, and machine learning algorithms. This review summarizes current trends in radiomics and hybrid radiomics-DL models applied to CT and MRI, with emphasis on methodological developments in radiomics. It then discusses key challenges, including feature reproducibility and model generalizability, and outlines future directions for developing robust, clinically translatable solutions.