Radiomic Analysis of CT Predicts Tumor Response in Human Lung Cancer with Radiotherapy

Abstract

Abstract Purpose Radiomics features can be positioned to monitor changes throughout treatment. In this study, we evaluated machine learning for predicting tumor response by analyzing CT images of lung cancer patients treated with radiotherapy. Experimental Design For this retrospective study, screening or standard diagnostic CT images were collected for 100 patients (mean age, 67 years; range, 55–82 years; 64 men [mean age, 68 years; range, 55–82 years] and 36 women [mean age, 65 years; range, 60–72 years]) from two institutions between 2013 and 2017. Radiomics analysis was available for each patient. Features were pruned to train machine learning classifiers with 50 patients, then trained in the test dataset. Result A support vector machine classifier with 2 radiomic features (flatness and coefficient of variation) achieved an area under the receiver operating characteristic curve (AUC) of 0.91 on the test set. Conclusion The 2 radiomic features, flatness, and coefficient of variation, from the volume of interest of lung tumor, can be the biomarkers for predicting tumor response at CT.