MRI-Based radiomic signature for predicting pathologic treatment response to neoadjuvant chemoradiotherapy and radioimmunotherapy in soft tissue sarcoma.

BACKGROUND

Prediction of treatment outcomes is essential for improving clinical management, particularly in patients with soft tissue sarcoma, where treatment options remain suboptimal. Given the limitations of current therapies, there is increasing interest in combining neoadjuvant radiotherapy with immunotherapy, referred to as neoadjuvant radioimmunotherapy (NRIT). We aim to develop a predictive model for assessing pathologic treatment response in patients undergoing NRIT or chemoradiotherapy, integrating radiomic features with radiologist assessments, clinical data, and pathology findings.

MATERIALS AND

METHODS

Radiomic and semantic features were extracted from pre- and post-treatment MRI scans. The XGBoost algorithm was used for feature selection and model development. Models included a model based on clinical variables and semantic features, a model based on radiomic features and clinical features and a model using all available features.

RESULTS

Study cohort included 213 patients (mean age of 54 years, male/female of 1.6). There were 17 patients in the prospective arm. The best model used all radiomic, clinical, and semantics features. It achieved an area under the receiver operating characteristic curve (AUC) of 0.72 (95% CI = 0.51-0.89) on the hold-out testing set.

CONCLUSION

Multi-modal radiomic-based models are effective in identifying patients at higher risk of non-response to neoadjuvant therapy. Furthermore, the performance of multi-modal radiomics-based models exceeded those based solely on radiologist evaluations. Our findings underscore the potential of radiomics in enhancing precision medicine by enabling identification of treatment response in STS patients undergoing NRIT before surgical excision of the tumor.