Integrating peritumor and tumor CT radiomics features in predicting local control after SBRT in patients with pulmonary oligometastases.

Abstract

Purpose: Local control prediction for patients with pulmonary oligometastases underwent stereotactic body radiotherapy (SBRT) is crucial for optimizing therapeutic strategies. This study aims to develop and validate a predictive radiomics model integrating both tumor-intrinsic and peritumoral features along with clinical factors to enhance local control prediction using a multi-center dataset.

Materials and methods: We analyzed 223 tumors from 146 patients, which was divided into a training set (n = 165) and an external validation set (n = 58). Radiomic features from the gross tumor volume (GTV) and peritumoral regions (pGTV) representing the tumor microenvironment (TME) in CT images were extracted and combined with clinical factors to build a clinical outcome prediction model. Tumor response was classified into Favorable Response Group (FRG) and Unfavorable Response Group (URG) according to the 3-month and 1-year follow-up. Models were built using a Multilayer Perceptron (MLP) approach with SHAP analysis.

Results: Model-G (with GTV features) and Model-P (with pGTV features) achieved a validation area under curve (AUC) of 0.806 and 0.708, respectively. Meanwhile, Model-GP (with GTV and pGTV features) demonstrated an improved performance with a validation AUC of 0.851, reflecting the added value of peritumoral features. The Model-GPC, which incorporated GTV, pGTV, and clinical features, achieved a best validation AUC of 0.902, demonstrating the model's ability to robustly integrate clinical and radiomic data for accurate local control prediction.

Conclusion: The Model-GPC, integrating clinical and radiomic features, accurately predicts post-SBRT local control in pulmonary oligometastases. Incorporating peritumoral features and SHAP analysis enhances prediction accuracy, offering insights to optimize SBRT strategies.