Relating Macroscopic PET Radiomics Features to Microscopic Tumor Phenotypes Using a Stochastic Mathematical Model of Cellular Metabolism and Proliferation

1. Introduction

Cancer is classified as a set of diseases related to uncontrolled cell proliferation and is a leading cause of death globally [1]. Various hallmarks of cancer, including resistance to cell death, genetic diversity, vascular network reconstruction, and dynamic tumor tissue microenvironment, result in distinct tumor phenotypes with significant heterogeneity between and within tumors. An increasing amount of evidence is showing that more heterogeneous tumors tend to exhibit more aggressive progressions and are more resistant to treatment with conventional therapy types. Moreover, tumors with similar heterogeneous properties have shown similar progression patterns and sensitivity to treatment despite manifesting at different locations [2,3,4]. Despite the inherent variability in tumor phenotypes, even for the same type of cancer, the de facto standard of care follows the “one-size-fits-all” approach wherein a standard dose is delivered to most patients. However, it has been shown that such an approach only works well for 25% of patients [5]. The growing field of precision medicine aims to tackle this problem and make a shift toward personalized treatments, including a larger role of imaging and radiomics [6].

Given a sufficiently large tumor size, positron emission tomography (PET) imaging with tracer ¹⁸F-fluorodeoxyglucose (FDG), a glucose analog, can be used to assess the metabolic heterogeneity of tumors in vivo. Accurate and standardized quantification of radiomics features in PET images that describe tumor shape, texture, and morphology can provide meaningful information relating to tumor heterogeneity. For instance, in a tumor texture analysis study by Orlhac et al., several texture indices were highly correlated with the molecular volume across three tumor types [4]. Another study by Hatt et al. [3] showed that heterogeneity quantification in five different tumor types had prognostic value for clinical decisions. An increasing amount of evidence is showing that imaging-derived radiomics signatures can facilitate better cancer diagnosis, prognosis, and treatment planning that is not just specific to each patient, but to each distinct tumor [7].

Despite numerous research studies on leveraging PET radiomics in different cancer types, there is a lack of understanding of how the observable tumor heterogeneity in PET images, quantified via radiomics feature extraction, is linked to the properties of the tumor microenvironment. No models exist that establish a link between the tissue microenvironmental parameters and functional heterogeneity...