Abstract
INTRODUCTION: Personalized cancer treatment requires precise scheduling of multiple therapeutic agents under biological constraints. Optimizing such regimens is especially challenging due to the nonlinear dynamics of tumor-immune interactions and strict feasibility boundaries. This study aims to develop an intelligent optimization approach capable of handling these complexities within a mathematical tumor treatment model. METHODS: We propose an Adaptive Dynamic ϵ-Simulated Annealing (ADϵSA) algorithm that integrates a multi-population search framework, dynamic ϵ-constraint control, and boundary-aware mutation mechanisms. The algorithm is applied to an improved tumor immunotherapy model (ITIT), formulated using ordinary differential equations (ODEs) based on established experimental and clinical studies. The model incorporates tumor cells, immune effector cells, and three types of anti-tumor drugs: chemotherapy, immunotherapy, and anti-angiogenic agents. RESULTS: Simulation experiments were conducted on twelve classical benchmark functions to evaluate the convergence performance and robustness of the algorithm. ADϵSA demonstrated strong global search capability, fast convergence, and solution stability. When applied to the ITIT model, the algorithm successfully identified optimal drug dosing schedules that significantly reduced simulated tumor burden-from ~1500 to below 500 cells-while maintaining treatment within physiologically acceptable limits. DISCUSSION: Unlike traditional metaheuristics such as PSO or GA, which are less suited for constraint-rich, dynamic ODE-based systems, ADϵSA offers structural advantages in trajectory feasibility and adaptive convergence. This study highlights the potential of biologically informed optimization algorithms in personalized oncology and provides a computational basis for future closed-loop, patient-specific treatment strategies.