.png){kind=icon}
.png){kind=icon}
Return
Double deep Q-network-based multi-drug chemotherapy scheduling optimization
.png){kind=icon}
PRE
AI
DOI：10.1007/s13534-026-00599-4.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Designing optimal chemotherapy schedules remains a fundamental challenge in oncology, requiring a careful balance between effective tumor suppression and the prevention of excessive systemic toxicity. In this study, a deep reinforcement learning-based control framework is proposed for optimizing multi-drug chemotherapy dosing strategies under physiological uncertainty and clinical safety constraints. Specifically, a Double Deep Q-Network (DDQN) algorithm is employed to learn adaptive dosing policies for a three-drug chemotherapy regimen within a dynamic tumor–toxicity modeling environment. The proposed framework optimizes the scheduling of three chemotherapeutic agents represented within a mechanistic PK/PD model using a DDQN approach. A multi-objective reward function is formulated to simultaneously promote tumor reduction while penalizing excessive toxicity and violations of dose constraints. Through simulations over 5000 training episodes and robustness analyses under parameter perturbations up to $$\pm 50\%$$ , the DDQN-based controller demonstrates the ability to reduce the tumor population from $$4.60517 \times 10^{11}$$ cells to approximately 42 residual cells while maintaining aggregate toxicity below the imposed safety limit of 300 units, with a mean value of 274. Robustness analyses are conducted under significant physiological parameter variations (up to $$\pm 50\%$$ ) and abrupt tumor growth disturbances, where the proposed approach consistently maintains bounded state trajectories and avoids constraint violations. These simulation-based results suggest that Double Deep Q-Network reinforcement learning may provide a useful framework for adaptive chemotherapy scheduling under uncertainty. However, further validation using experimentally derived datasets, retrospective clinical cohorts, and prospective studies is required before clinical applicability can be established.
Keywords:
Deep learning
Reinforcement learning
Cancer chemotherapy optimization
Multi-drug scheduling
DDQN
Journal
IF:
2.8
Papers:
515
Citations:
1.5K