Radiomics is defined as a quantitative high-throughput analysis of imaging data combined with model development aiming to predict biological correlates or clinical endpoints. The investigators of this study hypothesize that radiomic features may correlate with pathology-defined tumor grading in soft tissue sarcoma patients. The aim of this study is to develop a predictive radiomics model for tumor grading determination.

Soft tissue sarcomas (STS) constitute an overall rare malignant entity comprising 1% of all cancers with a yearly incidence rate of 3.8 per 100.000 inhabitants. Therapy decisions are made using clinical and pathological determinants defined by the American Joint Committee on Cancer (AJCC). It involves the TNM staging system that classifies STS by their tumor size (measured as maximal diameter), pathological tumor grading defined by the French Fédération Nationale des Centres de Lutte Contre le Cancer (FNCLCC) and the occurrence of nodal or distant metastases.

For the guidance of therapy, the most important factor constitutes tumor grading. In "low-grade" sarcomas (G1), surgical resection is often sufficient for durable tumor control. In "high risk" STS, however, resection of the tumor is combined with radiotherapy improving locoregional control and eventually survival.

Currently, invasive biopsies followed by pathological work-up are necessary to determine tumor grading. However, bioptic specimens are always restricted to small tumor subvolume.

Medical imaging-based analyses constitutes an alternative tool to characterize tissue. Recent developments in quantitative image analysis and data science have led to the evolvement of "Radiomics". It is defined as an algorithm-based large-scale quantitative analysis of imaging features. It should be considered as a two-step process with (1) extraction of relevant imaging features, and (2) incorporating these features into a mathematical model to ultimately predict patient or tumor-specific outcomes. In previous scientific studies, radiomic models have been associated with survival, tumor progression, and molecular changes including genetic mutations or expression profiles as shown in multiple malignant entities. In addition, radiomic models were able to predict tumor grading e.g. for gliomas, meningiomas, hepatocellular carcinoma or pancreatic neuroendocrine tumors. In contrast to pathology, quantitative image analysis (radiomics) has the principal advantage of analyzing the whole tumor.

In this study, the investigators are aiming to correlate radiomic features with tumor grading of STS. The ultimate goal is to develop a prediction model to non-invasively classify tumor grading. In a first step, the focus will be laid on differentiating "low-grade" and "high-grade" STS. In a second step, "high-grade" STS will be divided into G2 and G3 tumors.

To this end, the investigators will retrospectively analyze a patient cohort of 138 patients (139 tumors) with known tumor grading and available pre-therapeutic MRI-scans. As secondary endpoint overall survival will be determined for all patients. An independent patient cohort from the University of Washington (139 patients) will be used for external validation of the developed models.

Inclusion Criteria:

• Histologically proven soft tissue sarcoma
• Available pre-therapeutic MRI with a contrast-enhanced T1 weight fat saturated sequence +/- fat saturated T2 sequences (e.g. STIR)

Exclusion Criteria:

• Indeterminate tumor grading
• Osteosarcoma
• Ewing Sarcoma
• Endoprothesis-dependent MRI artifacts
• Previous radiotherapy or chemotherapy
• Lack of a contrast-enhanced T1 weight fat saturated MRI sequence

Patients with histologically proven soft tissue sarcomas with known FNCLCC tumor grading determined by biopsy prior to therapy.