Quantification of tumor heterogeneity using PET/MRI and machine learning

Abstract

Despite a broad understanding that solid tumors exhibit significant tissue heterogeneity, clinical trials have not seen a remarkable development in techniques that aid in characterizing cancer. Needle biopsies often represent only a partial view of the tumor profile, lacking the ability to comprehensively reflect spatiotemporal phenotypic changes. Recent multimodal multiparametric imaging techniques could provide further valuable insights if the complementary imaging information is sufficiently analyzed. Therefore, in this work I developed and applied machine learning methods on multiparametric positron emission tomography (PET) and magnetic resonance imaging (MRI) datasets, acquired using mice bearing subcutaneous tumors, to obtain a precise spatio-temporal characterization of intratumor heterogeneity.