Purpose/Objective(s): Preoperative radiotherapy (RT) followed by surgery is an effective treatment for localized high-risk Soft Tissue Sarcoma (STS). However, metastases develop in approximately 30-50% of patients. This study aims to develop a predictive model for metastatic relapse in STS patients treated with preoperative RT and surgery, utilizing radiomic features extracted from T1 gadolinium-enhanced MRI.

Materials/Methods: We retrospectively analyzed data from a consecutive cohort of STS patients who underwent preoperative RT and surgery between 2011 and 2020. All patients received 50 Gy of preoperative radiotherapy in 25 fractions, followed by surgery after four weeks. A contrast-enhanced MRI was performed before surgery to evaluate treatment response. The cohort included 45 patients, of whom 27 experienced metastatic relapse, with a median follow-up of 42 months (range: 3-96 months). T1 gadolinium-enhanced MRI scans were analyzed for all patients. Radiomic features were extracted from both normalized and non-normalized images using the Python package PyRadiomics, yielding 107 features. These features were then used to train and test a Random Forest-based classifier designed to predict metastatic onset. The model was trained, optimized, and tested using a 5-fold Nested Cross-Validation approach, ensuring realistic performance estimates for the AI-based algorithms. The implementation was carried out using the Python package Scikit-Learn.

Results: The AUC values for non-normalized and normalized images were 0.65 and 0.67, respectively, showing a slight improvement with normalization. These promising results suggest that such algorithms could be useful in clinical practice once trained on larger databases.