Predictive Digital Twin for Optimizing Patient-Specific Radiotherapy Regimens under Uncertainty in High-Grade Gliomas

TL;DR

The paper develops a patient-specific predictive digital twin for high-grade glioma that accounts for parameter uncertainty in tumor growth and radiotherapy response. It combines an MRI-informed logistic growth model with a Bayesian calibration framework and an $\alpha$-superquantile risk-based multi-objective optimization to generate a suite of personalized RT regimens balancing tumor control and toxicity. In silico experiments with 100 patients show the approach can lengthen time-to-progression for the same total dose and achieve tumor control with substantially reduced dose in many cases, with SOC not universally optimal. The work provides a principled path toward risk-aware, data-driven personalized radiotherapy, while acknowledging limitations and the need for clinical data to translate gains to practice.

Abstract

We develop a methodology to create data-driven predictive digital twins for optimal risk-aware clinical decision-making. We illustrate the methodology as an enabler for an anticipatory personalized treatment that accounts for uncertainties in the underlying tumor biology in high-grade gliomas, where heterogeneity in the response to standard-of-care (SOC) radiotherapy contributes to sub-optimal patient outcomes. The digital twin is initialized through prior distributions derived from population-level clinical data in the literature for a mechanistic model's parameters. Then the digital twin is personalized using Bayesian model calibration for assimilating patient-specific magnetic resonance imaging data and used to propose optimal radiotherapy treatment regimens by solving a multi-objective risk-based optimization under uncertainty problem. The solution leads to a suite of patient-specific optimal radiotherapy treatment regimens exhibiting varying levels of trade-off between the two competing clinical objectives: (i) maximizing tumor control (characterized by minimizing the risk of tumor volume growth) and (ii) minimizing the toxicity from radiotherapy. The proposed digital twin framework is illustrated by generating an in silico cohort of 100 patients with high-grade glioma growth and response properties typically observed in the literature. For the same total radiation dose as the SOC, the personalized treatment regimens lead to median increase in tumor time to progression of around six days. Alternatively, for the same level of tumor control as the SOC, the digital twin provides optimal treatment options that lead to a median reduction in radiation dose by 16.7% (10 Gy) compared to SOC total dose of 60 Gy. The range of optimal solutions also provide options with increased doses for patients with aggressive cancer, where SOC does not lead to sufficient tumor control.

Figures (13)

• Figure 1: Creating and evolving an HGG patient-specific predictive digital twin. Our digital twin methodology is illustrated for the case of HGG growth and response to RT from the post-surgery imaging visit to post-RT monitoring for disease progression. The digital twin is personalized through calibration using MRI data and then used to design an optimal risk-aware treatment regimen under uncertainty. The digital twin also allows for monitoring the disease progression throughout the patient's treatment and recovery.
• Figure 2: Overview of the components comprising a predictive digital twin in the oncology setting.
• Figure 3: Illustration of TTP and use of negative TTP as QoI. (a) Visual representation of TTP for various trajectories obtained from $n_\text{MC}$ Monte Carlo samples of $\theta$, (b) histogram of samples of $T_\text{TTP}(\boldsymbol{u}, \theta)$ and (b) histogram of samples of QoI $-T_\text{TTP}(\boldsymbol{u}, \theta)$ along with the estimated risk with $\alpha=0.95$.
• Figure 4: Predictive digital twin timeline for HGG patients. The predictive digital twin features a personalized risk-based strategy to optimize the adaptive RT regimen under uncertainty. This strategy is constructed assuming a standard collection of MRI data during the course of the clinical management of HGG after surgery, whereby MRI scans are prescribed after the surgical intervention (day 0), before the onset of RT (day 20), and during the second week of the RT regimen (day 27). On day 27, the digital twin is calibrated using the MRI data and then the risk-aware treatment plan is solved using the calibrated model and deployed for the remaining five weeks of RT.
• Figure 5: Posterior parameter distributions of the calibrated digital twins after assimilating observed data at the three imaging visits are shown for three patients: (a) Patient 1, (b) Patient 2, and (c) Patient 3. The dashed gray line indicates true parameter values for reference and not seen by the digital twin. Note that the same prior distribution is used for initializing digital twins of all the patients. Posterior distributions concentrate around the unseen true parameters and show reduction in uncertainty compared to prior distributions.