Soft Tissue Sarcoma: Preoperative MRI-Based Radiomics and Machine Learning May Be Accurate Predictors of Histopathologic Grade

OBJECTIVE. The purpose of this study was to assess the value of radiomics features for differentiating soft tissue sarcomas (STSs) of different histopathologic grades. MATERIALS AND METHODS. The T1-weighted and fat-suppressed T2-weighted MR images of 70 STSs of varying grades (35 low-grade [grades 1 and 2], 35 high-grade [grade 3]) formed the primary dataset used to train multiple machine learning algorithms for the construction of models for assigning STS grade. The models were tested with a separate validation dataset. RESULTS. Different machine learning algorithms had different strengths and weaknesses. The best classification algorithm for the prediction of STS grade had a combination of the least absolute shrinkage and selection operator feature selection method and the random forest classification algorithm (AUC, 0.9216; 95% CI, 0.8437-0.9995) in the validation set. The accuracy of the combined methods applied to the validation set was 91.43%; sensitivity, 88.24% and specificity, 9444%. CONCLUSION. Because of tumor heterogeneity, initial biopsy grade may be an underestimate of the final grade identified in extensive histopathologic analysis of surgical specimens. This creates an urgent need to construct an accurate preoperative approach to grading STS. This radiomics study revealed the optimal machine learning approaches for differentiating STS grades. This capability can enhance the precision of preoperative diagnosis.