The term “radiomics” derives from a combination of “radio”, which means medical images, and “omics”, which indicates the analysis of large amounts of data representing an entire set of some kind like genome (genomics) and proteome (proteomics). Radiomics includes extraction, analysis, and interpretation of large numbers of quantitative data, known as radiomic or texture features, from medical images. Radiomics has recently gained much attention in oncologic imaging, as it can potentially quantify tumor heterogeneity.

Bone and soft-tissue tumors, and particularly primary bone tumors or sarcomas, are rare entities with several histological subtypes, each with an incidence < 1/100,000/year. Among them, osteosarcoma is the most common sarcoma of the bone and, together with Ewing sarcoma, it has a higher incidence in the second decade of life. Chondrosarcoma is the most common bone sarcoma of adulthood. The most frequent soft tissue sarcomas are liposarcoma and leiomyosarcoma.

The impact of radiomics-based machine learning for the characterization of bone and soft-tissue tumors is expected to be high. In clinical practice, both biopsy and imaging findings integrate clinical data before any treatment is started. However, preoperative biopsy may be inaccurate in large heterogenous lesions and may erroneously lead to tumor downgrading, as only small tumor areas are sampled. In turn, inaccurate preoperative diagnosis may result in an inadequate treatment and subsequent need for further interventions, with increased morbidity. Imaging plays a crucial role in the assessment of bone and soft tissue tumors. Particularly, magnetic resonance imaging (MRI) is the method of choice for local staging, whereas computed tomography (CT) and positron emission tomography (PET)-CT are employed for general staging. However, benign and low-grade tumors as well as low-grade and high-grade tumors may show overlapping features. Hence, current imaging techniques may be further implemented to better characterize bone and soft tissue tumors non-invasively. In this regard, the creation of radiomics-based machine-learning models has the ambitious purpose to develop systems able to automatically analyze radiological images and provide a correct diagnosis in a very high percentage of cases. This in turn may be immediately transferred to clinical practice, reducing the need of unnecessary biopsies and maximizing the diagnostic performance of currently available methods.

The primary purpose of radiomics is to extract as much and meaningful quantitative information as possible to be used in decision support. This is also known as precision medicine, where patients that belong to different subtypes can be identified to improve their outcomes. Quantitative information about heterogeneity reflects, indeed, adverse tumor biology which is a parameter hard to capture using conventional qualitative imaging assessment or sampling biopsies. Particularly, to date, radiomic studies focused on discriminating tumor grades and types before treatment, monitoring response to therapy, and predicting patients’ outcome.

The rarity of these entities has limited the number of radiomic studies to a few very preliminary investigations. From a recent systematic literature review on radiomics of bone and soft-tissue sarcomas, it emerges that encouraging results have been achieved for discrimination of certain tumor entities, such as benign vs. malignant or low-grade vs. high-grade lesions, and outcome prediction. However, the reproducibility of the radiomics-based method has not been evaluated in more than one-half of the papers included in this review, thus resulting in very variable approaches. Furthermore, almost all proposed radiomics-based methods have not been validated in independent clinical datasets (e.g. a series of cases from different institutions), thus limiting generalizability of the results and their transferability to clinical practice. The issues of variability and clinical validation of these approaches need to be addressed in order to better understand how the radiomic profile changes from one tumor subtype to another, help establishing the correct diagnosis and direct the treatment properly.