Radiomics in application to diseases of the musculoskeletal system: a review

Radiomics is a technique used to extract numerous quantitative features from digital medical images. A decade ago, this method was applied in oncology, but now it has expanded to non-oncological diseases, particularly those affecting the musculoskeletal system and connective tissues. This article provides an overview of the current advances in radiomics for diagnosing diseases of the musculoskeletal system. In this review, we assessed 37 original research papers published in English between 2020 and 2023. The most commonly used imaging modalities were magnetic resonance imaging (54%) and computed tomography (32%), while dual-energy X-ray absorptiometry (14%), ultrasound (5%), and radiographs (5%) were less frequently used. The majority of the studies apply manual segmentation to identify the regions of interest. Various classification models have been developed that incorporate clinical, radiomics, and deep features, with combined clinical-radiomics models being the most prevalent one. The most commonly affected areas in diseases of the musculoskeletal system were the spine and large joints. The prevalence of the multi-source input models (primarily clinical-radiomics) compared to that of single-source input models (clinical only, radiomics only) for diagnosing diseases of the musculoskeletal system can be explained by the higher classification performance, likely due to the inclusion of a larger number of independent information sources. Although the development of models or deep-learning features for automatic segmentation and classification holds promise, it requires significant efforts in creating image databases for deep model training. Thus, radiomics may be particularly beneficial for the early detection of diseases of the musculoskeletal system that cause pathological changes in the soft tissues, which may not be visible to the naked eye.