| dc.description.abstract | Clinical imaging and biopsies are major tools for cancer diagnosis. We explore whether serial imaging coupled with a mathematical model could test and refine patient specific forecasts for therapy outcomes.
We first examine whether parameters for a basic mathematical tumor model can be extracted from diffusion-weighted images (DW-MRI) and used for early detection of drug resistance to targeted therapy. We find that incorporating the spatial information available in clinical images improved the parameter extraction accuracy by an order of magnitude. Using statistical methods to compare the predictions of models with and without resistance, we find our method can identify tumor resistance well before current standards.
As the mixing of phenotypes within a tumor is not known, we proposed two possibilities for the spatial mixing of drug resistant and sensitive cell types, namely, a free diffusion and a constrained diffusion model. From experiments, we find that the resistant cell line spreads more rapidly than the sensitive cell line; that the diffusion constant describing the spread of resistant cells decreases with cell density; and that the sensitive cells reach higher packing density. Subsequent experiments find that asymptotic cell packing varies across tumor subtypes. As both candidate models assume constant diffusion rates and uniform packing, neither model agrees quantitatively with the mixing experiments. For the experimental parameter ranges, both models are equally effective at identifying resistance from serial DW-MRI.
We find that both protein expression, measurable from biopsies, and the microenvironment impact tumor resistance. For the non-small cell lung cancer line investigated, the concentration of the receptor c-Met increases the likelihood of positive growth under therapy (resistance). Here, we find that the growth rate varies with the logarithm of the microenvironment stiffness, indicating that not all tumors positive for pro-proliferation markers will be resistant. Taken together, our work presents a promising method to incorporate clinical imaging and tumor biopsies towards refining tumor models and forecasting patient specific treatment response. | |
