Variational Radiomics for Active Surfaces

We show how radiomics features can be integrated directly into the data driven terms of a variational active surface using a novel region based strategy in which the segmentation goal will be to match learned values of various radiomics features both inside and outside the region of interest to be localized. We do not merely utilize a one-time, up-front radiomics feature calculation to generate preprocessed inputs for a subsequent variational active surface. Rather, we exploit the variational properties of the radiomics features themselves in the active surface surface formulation. We demonstrate the formulation of the corresponding energy functional as well as its gradient flow PDE for an illustrative selection of common region based radiomics features.