Closing the Loop in Precision Oncology: A Digital Twin‐Driven Paradigm for Dynamic Decision‐Making

Precision oncology is evolving from a predictive paradigm towards adaptive decision‐making with dynamic data. This paper systematically presents a forward‐looking conceptual framework for a Closed‐Loop Intelligent Oncology System (CIOS), which aims to support iterative evaluation of candidate treatment strategies and inform timely adjustments in care by updating a patient‐specific digital twin that approximates key aspects of the patient's pathophysiological state. The core idea of CIOS is that it does not only predict tumor evolution but frames decision‐making as a closed‐loop process that evaluates candidate policies within a computational environment under explicit constraints. A five‐layer reference architecture is described to elucidate the operational mechanism of CIOS: (1) Perception Layer: Multimodal data fusion. (2) Simulation Layer: PINN‐based digital twins. (3) Decision Layer: Safe RL for adaptive therapy. (4) Evolution Layer: Meta learning & federated learning. (5) Trust Layer: XAI & uncertainty reporting. This paper systematically reviews the preliminary research progress and application explorations of the key technologies at each of these layers within the field of oncology. It also provides an in‐depth discussion of the technical challenges, clinical translation barriers, and future research directions, aiming to provide a theoretical reference and potential guidance for further research and eventual realization of this conceptual framework.