•Radiation pneumonitis (RP), a main complication of thoracic radiotherapy(RT), is a severe and potentially fatal to many cancer patients especially lung cancer patients.
•Statistical differences between RP/nonRP groups were observed in several individual image features.
•The accuracy rate derived from SVM classification using multiple radiomics features as input can be significantly increased to 85%
•Radiomics analysis predicts RP using the pre-treatment PET/CT images.
•SVM classification significantly improves the accuracy of RP prediction and provides a potential prognostic biomarker to assist the personalized treatment planning.
The aim of this study:
•Aim 1: To investigate radiomics features derived from pre-RT FDG-PET/CT as potential prognostic biomarkers of symptomatic RP in lung cancer patients.
•Aim 2: Support Vector Machine (SVM) was applied to improve the accuracy of prognosis with multiple feature inputs