SoC-DT: Standard-of-Care Aligned Digital Twins for Patient-Specific Tumor Dynamics

TL;DR

This paper tackles the challenge of predicting tumor trajectories under standard-of-care therapies by building SoC-DT, a differentiable, physics-based digital twin that couples reaction-diffusion tumor growth with discrete interventions such as surgery, chemotherapy, and radiotherapy, modulated by genomic and demographic covariates. The core advances are an IMEX–ETD solver that preserves stability and positivity while handling discontinuous treatment events, and an event-aware adjoint enabling gradient-based calibration from multimodal data. Evaluations on three synthetic datasets and a real glioma cohort show SoC-DT outperforms classical PDE models and mobile neural baselines in post-treatment mask forecasting and time-to-progression prediction, with robust performance across hyperparameters. The framework offers a principled, interpretable path toward patient-specific digital twins for oncology, with potential to support counterfactual treatment planning and dynamic updating as new data arrive.

Abstract

Accurate prediction of tumor trajectories under standard-of-care (SoC) therapies remains a major unmet need in oncology. This capability is essential for optimizing treatment planning and anticipating disease progression. Conventional reaction-diffusion models are limited in scope, as they fail to capture tumor dynamics under heterogeneous therapeutic paradigms. There is hence a critical need for computational frameworks that can realistically simulate SoC interventions while accounting for inter-patient variability in genomics, demographics, and treatment regimens. We introduce Standard-of-Care Digital Twin (SoC-DT), a differentiable framework that unifies reaction-diffusion tumor growth models, discrete SoC interventions (surgery, chemotherapy, radiotherapy) along with genomic and demographic personalization to predict post-treatment tumor structure on imaging. An implicit-explicit exponential time-differencing solver, IMEX-SoC, is also proposed, which ensures stability, positivity, and scalability in SoC treatment situations. Evaluated on both synthetic data and real world glioma data, SoC-DT consistently outperforms classical PDE baselines and purely data-driven neural models in predicting tumor dynamics. By bridging mechanistic interpretability with modern differentiable solvers, SoC-DT establishes a principled foundation for patient-specific digital twins in oncology, enabling biologically consistent tumor dynamics estimation. Code will be made available upon acceptance.

Figures (10)

• Figure 1: Architecture. A. A plug-and-play framework for generating synthetic datasets for different cancer types, B. An adaptation of a timeline for standard-of-care cancer treatment, C. Proposed Standard-of-Care tumor growth modeling framework. Post-treatment tumor structure is predicted from pre-treatment scans using a PDE framework that incorporates basic diffusion and proliferation terms, along with modules simulating surgery, chemotherapy, and radiotherapy.
• Figure 2: (a) For our experiments, we use 3 synthetic datasets and 1 real clinical dataset. Initial and final timepoint images along with the different treatment methods are shown. AG: Adult Glioma, HCC: Hepatocellular Carcinoma, and NAC: Neoadjuvant Chemotherapy for Breast cancer. We perform quantitave analysis on both synthetic and real datasets. Synthetic datasets are primarily used for stress testing and real datasets are used for clinical downstream tasks (b) Comparison of our method against baseline models for regression of time-to-progression (days), evaluated using MAE and RMSE ($\mu\pm\sigma$ is reported).
• Figure 3: Standard-of-Care Digital Twins. Examples of SoC-DT projections comparing predicted post-treatment tumor masks with the ground-truth follow-up tumor masks, given pre-treatment and post-operative tumor masks overlaid on the corresponding FLAIR sequences. Different treatment options (surgery, radiotherapy, chemotherapy) are illustrated, with the elapsed time from post-operative to post-treatment scan indicated. Quantitative changes in SNFH volume are also reported.
• Figure 4: A. Qualitative comparisons. We compare the SoC-DT generated post-treatment tumor masks with baselines like PINN and Hybrid PDE. We also report the DSC scores, B. Sensitivity analysis. Box plots are shown for DSC scores across different genomic marker types, SNFH change and chemotherapy/+immunotherapy.
• Figure 5: AG (Brain Cancer). An example case from the generated AG data where we show different phantom MR sequences (T1C, T2F, and T2W) along with the tumor masks for multiple timepoint visits.

Theorems & Definitions (10)

• Theorem 1: Well-posedness and bounds
• proof
• Lemma 1: Positivity of the implicit diffusion step
• proof
• Theorem 2: Stability and convergence of IMEX--ETD
• proof
• Lemma 2: Nudging preserves bounds
• proof
• Lemma 3: Event maps: invariance and Lipschitz
• proof