PD-L1 Prediction of Lung Cancer by a Radiomics Method Using Rough Tumor Mask in CT Images

Programmed Death Ligand-1 (PD-L1) plays a key role in the success of immunotherapy for cancer treatment. However, immunohistochemistry(IHC) based PD-L1 detection requires biopsy of patients ’ tumor through surgery. In this paper, we discussed the possibility of predicting PD-L1 expression based on radiomics and rough tumor mask on CT images. We used various methods to obtain different rough tumor masks, and explored the radiomics feature based on these masks. Proved the feasibility of PD-L1 prediction based on rough tumor mask. The results of 5-fold cross validation experiment on our PD-L1 dataset shows that, the average AUC results of our method reached 0.8284 which outperformed the method based on precise delineation, provides a new paradigm for PD-L1 identification.