Answers are grounded in this paper and cite their sources.
Demonstrates a Physics-Informed Neural Network effectively estimates tumor growth parameters, indicating its potential for therapeutic applications.
Mechanistic tumor growth models are widely used to describe disease progression and treatment response, but their utility depends on accurate estimation of parameters governing the underlying biological processes. In this study, we employ a Physics-Informed Neural Network (PINN) to estimate the parameters of a tumor growth model that captures both tumor dynamics and drug effects. We introduce a piecewise PINN that splits the time domain at dosing events to handle non-smooth dose-driven dynamics, and we incorporate drug injection by representing the pharmacokinetic subsystem analytically via an impulse-response function. The approach is evaluated on synthetic tumor-volume trajectories generated from known parameter sets and dosing schedules from an experimental cohort of 54 mice. Across the cohort, the PINN accurately reconstructs total tumor volume and robustly estimates the tumor proliferation rate a, with inferred values closely aligned with the true values (R2=0.841). The framework was also able to estimate the drug killing effect parameter b. This consistency is further supported by forward ODE simulations using the PINN-estimated parameters. Within the evaluated setting, performance depended on the model structure, parameter identifiability, and training configuration, underscoring the need for careful loss weighting and further validation. Overall, the results demonstrate the feasibility of piecewise PINNs for parameter inference in tumor growth models and support their further study in realistic therapeutic settings.
No takes yet. Share an insight, caveat, or question.
Kisbenedek et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: