Radiomics of MRI scans enables differentiation of benign from malignant soft tissue sarcoma.

Abstract

11565 Background: Accounting for >15,000 cases annually, soft tissue sarcomas (STSs) represent a large spectrum of tumors with varying histology and clinical course. However, diagnosis of STSs on MRI is challenging due to its heterogeneous imaging phenotypes, such as surrounding soft tissue edema, vascular lesion characteristics, and specific imaging signs, such as splitting of fat between muscles. The goal of this study was to evaluate the utility of radiomic feature analysis for characterizing subtle heterogeneity patterns on T2-weighted MRI to distinguish benign and malignant STSs. Methods: T2-weighted MRI of N = 176 patients with STS were retrospectively obtained from a single institution. (Table) Tumor regions were manually annotated by a radiologist. The radiomic features included textural features, e.g. gray-level co-occurrence matrix features, which quantify the correlation between textural patterns within annotated regions. Maximum relevance minimum redundancy feature selection was employed to identify top 10 radiomic features that are associated with STS malignancy. Then, the identified features were integrated to generate a radiomic risk score (RRS) via logistic regression on a 10-fold cross validation to differentiate benign and malignant STS. Multivariate analysis and the area under the receiver operating characteristic curve (AUC) were employed to evaluate the association of RRS and clinical parameters, including age, gender, and STS location, with STS malignancy. Results: Multivariate analysis illustrated that age and RRS are independent variables in predicting STS malignancy (p < 0.05). The RRS yielded AUC as 0.76 (confidence interval (CI): 0.69-0.83), age 0.71 (CI: 0.64-0.79) and the combination of the RRS and age 0.84 (CI: 0.78-0.90). Conclusions: The RRS generated by radiomic features was found to be a significant predictor of STS malignancy. Furthermore, RRS could provide complementary information to the clinical parameters in differentiating benign and malignant STSs.[Table: see text]