Prediction of local recurrence using clinical and radiomic features in lung oligometastases treated with stereotactic body radiotherapy

The purpose of this study was to create machine learning-based models by combining clinical and radiomic features in lung oligometastases treated with stereotactic body radiotherapy (SBRT) and to assess the potential of these models in predicting local recurrence. A total of 65 patients and 80 lesions who underwent SBRT for lung oligometastases were evaluated retrospectively. Lesions were categorized into two groups based on the Response Evaluation Criteria in Solid Tumors (RECIST 1.1): those with local recurrence (n=12) and those without local recurrence (n=68). Radiomic features were obtained from simulation computed tomographies in four different groups: first-order statistics, morphological, textural, and wavelet transform-based features. The potential of clinical and radiomic features in predicting local recurrence was investigated using machine learning-based models and assessed with performance metrics. Among a total of 90 different models, it was observed that the combination of the "SVMAttributeEval" feature selection method and the "Logistic" machine learning model was the most successful in predicting local recurrence in lung oligometastases treated with SBRT. The accuracy of this model was found to be 93.75%, precision 0.76, sensitivity 0.83, specificity 0.95, F1 score 0.80 and AUC value 0.84. This machine learning model was created using features such as soft tissue sarcoma pathology, age, Wavelet-LLL-FirstOrder-Kurtosis, Original-FirstOrder-Kurtosis, and Wavelet-HHH-GLCM-ClusterShade. The results indicate the potential of machine learning models created by combining clinical and radiomic features in predicting local recurrence in lung oligometastases treated with SBRT. Validation of these results in a multicenter, large, and independent dataset is needed.