Abstract 2742: A biology-based, mathematical model to predict the response of recurrent glioblastoma to treatment with 186Re-labeled nanoliposomes

Abstract

Abstract Introduction: 186Re-nanoliposomes (RNL) are a theranostic that emits a therapeutic payload of ionizing beta radiation, and a gamma photon to be measured with SPECT. The RNL is delivered via convection-enhanced delivery, resulting in a highly localized distribution around the glioma that produces up to a 30-fold increase in maximum tolerable dose. RNL provides a continuous source of low dose rate irradiation, until the particles are cleared biologically or decay. The goal of this study is to evaluate the accuracy of a patient calibrated reaction-diffusion equation for predicting the growth and response of recurrent glioblastoma multiforme (GBM) following treatment with RNL. Methods: Multi-parametric images were collected from patients (n=10) receiving RNL treatment, consisting of pre-treatment and follow-up MRIs (Day 0, 28*, 56, 112*) and SPECT/CTs acquired at the middle and end of infusion, and 24-, 112-, and 192-hours post-infusion. For each time point, tumor segmentations and cell count maps are computed using contrast enhanced and diffusion weighted MRI, respectively. The spatio-temporal response to RNL is modeled using a biology-informed reaction-diffusion model describing tumor cell proliferation, invasion, and radiation induced death. Key model parameters related to the RNL activity are populated through a quantification of the SPECT time course. The remaining model parameters related to diffusivity, proliferation, and death rate are calibrated via the Levenberg-Marquardt algorithm for each individual patient, and then used to forecast growth. Calibrations are performed in two different scenarios, first to all imaging time points to assess the model capabilities (Scenario 1), and then without the last acquired MRI, which is set aside to evaluate prediction accuracy (Scenario 2). Error will be assessed at the global (Dice similarity coefficient and percent error in total cell number) and local (concordance correlation coefficient or CCC) levels for both scenarios. *Patients are scanned on day 28 or day 112 at a minimum, potentially both (n=5) Results: Scenario 1 calibrations produced on average, Dice=0.92, CCC=0.69, and total cell percent error = 10.2%, validating usage of the current model formulation. Scenario 2 calibrations show high prediction success on a global scale, mean Dice=0.78, mean total cell percent error =23%, but resulted in poor local accuracy, mean CCC=0.21. Discussion & conclusion: The mathematical model and processing framework can predict the spatiotemporal evolution of recurrent GBM after treatment with RNL. Ongoing efforts include validating the methodology on a larger cohort and further model selection to improve predictive capabilities. Citation Format: Chase Christenson, Chengyue Wu, David A. Hormuth, Shiliang Huang, Andrew Brenner, Thomas E. Yankeelov. A biology-based, mathematical model to predict the response of recurrent glioblastoma to treatment with 186Re-labeled nanoliposomes [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 2742.