Imaging-based tumor forecasting to predict brain tumor progression and response to therapy

点击翻译按钮获取中文摘要

The vision for this program is to develop tumor forecasting methods to predict and optimize the response of glioblastoma multiforme to standard-of-care therapies—and do so on a tumor-specific basis. A fundamental challenge in the care of patients with brain tumors is the limitation of standard radiographic methods to accurately evaluate, let alone predict, patient response. We propose to address this shortcoming by developing predictive, biologically-based mathematical models that incorporate the hallmark characteristics of brain tumor growth (e.g., tumor induced angiogenesis, hypoxia, necrosis, proliferation, invasion, and resistance to therapy) that can be initialized using advanced, subject-specific imaging data. This project will address two critical gaps in the care of patients battling brain cancer. First, our imaging-based, mathematical framework accounts for subject-specific characteristics and treatment regimens on model predictions. Second, in most studies, the ground truth used for validation of the predictive model is whether the model can predict future regional contrast enhancement, despite the well-known limitations of this qualitative MRI feature. Thus, while prior human studies have demonstrated the potential of predictive modeling, its translation into a realistic radiologic tool is fundamentally hindered by lack of systematic, pre-clinical validation where critical tumor characteristics (e.g., tumor heterogeneity and whole brain tumor cell distribution) can be precisely known and rigorously controlled. To overcome these limitations, we aim to: 1) establish the accuracy of tumor-specific modeling to predict spatiotemporal progression and 2) establish the accuracy of tumor-specific modeling to predict therapeutic response. Experimentally, we will construct a family of mathematical models that employ quantitative MRI data to capture the fundamental biological features of glioblastoma. These data are longitudinally acquired in patient derived xenografts that are treatment naïve or undergoing radiotherapy and/or chemotherapy. The model family is then calibrated with these data and a novel model selection strategy is employed to choose the most parsimonious model for predicting the spatio-temporal evolution of each tumor which is then compared to MRI data collected at future time points. Model predictions of tumor progression will be validated via registration to 3D fluorescent images of cleared ex vivo tissue, a technique that enables visualization of whole brain tumor burden. We will provide the clinical and scientific community with a validated mathematical description of glioma progression that can reliably predict progression and therapy response across a range of relevant glioma signaling pathways and can be readily applied to the clinical setting.