Discriminant Radiomic Feature Selection for PD-L1 Prediction in Clear Cell Renal Cell Carcinoma

Clear cell renal cell carcinoma (ccRCC) is the most common type of kidney cancer. Previously, researchers explored the concept of utilizing radiomic features from multiphase computer tomography (CT) imaging to predict the tumor immune microenvironment (TIME) measurements, such as the PD-L1 levels, from multiplex immunofluorescence (mIF) images. PD-L1 is an important prognostic marker for ccRCC and CT radiomics can be a non-invasive modality to predict PD-L1. However, prediction performance in the prior work is still limited. This could be due to the limitation of either the underlying nature of these associations, the richness of imaging features, or the capabilities of supervised learning algorithms. This work implements two Green Learning (GL) methods discriminant feature test (DFT) and least-square normal transform (LNT) to improve predictive performance while minimizing model complexity. We observed a significant improvement in the radiomics prediction performance of tumor epithelium PD-L1 >1%, >5% and PD-L1>10%. Compared to prior research, the AUROC values improved from 0.61 to 0.76, 0.75 to 0.85 and 0.85 to 0.88, respectively.