Radiomics in non-oncologic musculoskeletal diseases: from pixels to practice

Radiomics has emerged as a powerful tool in medical imaging, enabling the extraction of high-dimensional data from radiologic images to uncover quantitative relationships between image features and underlying pathology. Although it is most established in oncology, radiomics is now being widely adopted for non-oncologic musculoskeletal conditions, including osteoporosis, osteoarthritis, rheumatologic diseases, nerve pathologies, and trauma-related injuries. This review provides an overview of the principles and workflows of radiomics, emphasising its applications beyond oncology. It highlights key methodological steps including image acquisition, region-of-interest segmentation, feature extraction, and model building. Special attention is given to segmentation variability and the need for standardisation across imaging protocols and software platforms to ensure reproducibility. The potential of radiomics for early diagnosis and monitoring of disease progression makes it a promising tool in precision medicine. However, challenges persist, particularly in the standardisation and reproducibility of imaging protocols, segmentation, and feature extraction, which limit broader clinical integration. Recent advancements in artificial intelligence (AI) and deep learning have helped address some of these challenges by enhancing the consistency and efficiency of radiomics analysis. Moreover, open-source AI platforms can lower the barriers to conducting radiomics research, encouraging wider adoption. For radiomics to become a reliable tool in clinical practice, it must move beyond complex feature sets to standardised, reproducible methodologies that deliver actionable insights for improved patient care.