The use of immunotherapy to treat cancer continues to generate hope and excitement among those involved in cancer care and research. However, our inability to explain why some patients do not respond to immunotherapy, combined with our inability to identify early response or predict the responders, poses serious challenges in this field. Currently, biopsies serve as the most informative way to assess the immunological activity within a cancerous area, but we are spatially and temporally limited in the number of biopsies we can obtain from patients, especially in cases of brain cancer. Clear evidence of tumor-immune environment heterogeneity across patients suggests that we will have to use an individualized approach in order to accurately assess patient tumor?s specific immune environment and the evolution of these complex systems. We propose to use computational modeling and artificial intelligence to bridge the spatial scales of the cellular content comprising each MRI at the voxel level, but also to bridge the temporal scales. We will focus on the most cellular immune population in glioblastoma, microglia/macrophages, that constitute as much as 50% of the cellular content of tumor specimens. By fusing MRI with the biological heterogeneity found in image- localized biopsies through such radiomics approaches provides an opportunity to individualize our understanding of the the tumor-immune environment, broadly benefiting scientists across the fields of oncology and immunology. In addition to providing a deeper understanding of the tumor at every imaging time point, the radiomics maps can also be used to parameterize dynamic mechanistic models of tumor growth to allow for prediction of future dynamics. These spatio-temporal models allow us to test hypotheses about causal relationships between different cell types and microenvironmental factors, as well as to verify whether the radiomics maps provide early dynamic insights into tumor response that can impact clinical decision making.
Our inability to explain why some patients do not respond to immunotherapy, combined with our inability to predict the responders, poses a serious challenge to advancing cancer care and research. Currently, biopsies serve as the most informative way to assess the immunological activity within a cancerous area, but we are spatially and temporally limited in the number of biopsies we can obtain from patients, especially in cases of brain cancer. We propose here a novel strategy merging artificial intelligence and mechanistic modeling to reveal the spatial and temporal tumor-immune landscape within patients by connecting a unique resource of image-localized biopsies and clinical imaging for cohorts treated with standard and immunotherapies.
