[18F]FDG PET/CT radiomics to predict local failure of stereotactic body radiation therapy of liver metastases

Purpose To develop machine learning models for prediction of local failure for liver metastases in patients with various primary cancer treated by stereotactic body radiation therapy (SBRT) using [ 18 F]FDG PET/CT radiomics combined with clinical parameters includind dosimetric data. Methods We retrospectively collected 90 patients with 117 liver metastases (60% for training and 40% for testing) from University Hospital of Liège treated with SBRT between 2010 and 2020 and who had undergone pretreatment [ 18 F]FDG PET/CT. Radiomics features were extracted using the Pyradiomics toolbox®. Clinical, radiomics and combined models were trained and tested using a neural network approach to predict local failure. Results In the training ( n = 68) and testing sets ( n = 49), the clinical model achieved moderate performances to predict local failure with AUCs of 0.62 (95% CI: 0.49–0.73) and 0.55 (95% CI: 0.40–0.69), respectively. The best radiomics (original_glcm_InverseVariance, original_gldm_DependenceEntropy and original_glcm_Correlation) model achieved higher predictive ability in the training and the testing sets, with AUCs of 0.83 (95% CI: 0.75–0.89) and 0.87 (95% CI: 0.74–0.95), respectively, while the combined model is slightly less efficient with AUCs of 0.75 (95% CI: 0.63–0.84) and 0.73 (95% CI: 0.59–0.85). Conclusion Radiomics features extracted from pre-SBRT [ 18 F]FDG PET/CT outperform clinical parameters in the prediction of local failure of liver metastases and could help select patients eligible for alternative treatment or dose escalation. Multicentric and prospective validation of our models with very good results should now be carried out.