Physics-informed neural network for parameter identification of a tumor model

The intersection of biomedical applications and advanced control algorithms is a promising direction for therapy optimization. Dynamic models of biological systems, such as tumor growth, enable control algorithms to design personalized and adaptive treatment strategies that improve therapeutic outcomes. Tumor modeling is an evolving field that aims to understand tumor behavior better, predict the precise drug responses, and design optimized dosage regimens. Our approach focuses on incorporating individual patient characteristics into mathematical models, enabling the identification of patient-specific differences. We model tumor progression over time under varying chemotherapy regimens using ordinary differential equations (ODEs). These equations include parameters that describe key biological processes, such as tumor growth and necrosis rates, dead cell washout, and drug efficacy. Parameter estimation is essential for the development of personalized and adaptive treatment strategies. In this work, we use a Physics-informed neural network (PINN) to identify the model parameters. We also extended the method to handle impulsive inputs in the form of instantaneously injected chemotherapeutic drug doses. We evaluated this method on in silico data. The results demonstrated high fitting accuracy and confirmed the applicability of the proposed approach under simplified conditions; however, further refinement is required to handle in vivo data in the future.