Murine in vivo tumor model to explain the interpretability of radiomic features.

Abstract

Medical imaging plays a crucial role in the management of oncological patients, where it is routinely used for diagnosis, staging and follow-up of tumors. Additionally, in the context of radiotherapy (RT), medical physicists utilize medical images for dose planning and accurate dose delivery. These images contain valuable information that reflects the underlying phenotype. This information can be accessed through the extraction of quantitative features that are subsequently used to design machine learning prediction models. This process is referred to as radiomics. However, the inherent non-biological-interpretability of radiomic features strongly hinders their clinical application. Therefore, we aim to unravel the biological meaning of radiomic features by performing dedicated preclinical in vivo experiments. In this study, we aimed to advance and optimize our original setup. Radiomic features from computed tomography (CT) scans of an in vivo murine tumor model were extracted. Mice were scanned, afterwards treated with RT, an oxygen-inducing drug or a combination hereof and re-scanned. Features were analyzed and compared using an exploratory factor analysis (EFA). The results revealed that some features are able to differentiate between the treatment groups. Furthermore, the features exhibited a high level of repeatability upon rescanning.