“The image quality of low-dose computed tomography (LDCT) is affected by noise and artifacts, and existing deep learning methods have shortcomings. Experts propose a high-resolution reconstruction framework HRDRL Net based on deep reinforcement learning, which models the LDCT denoising task as a sequential decision-making process. An asynchronous dominant action evaluation algorithm is used to train the agent, and a dual path multi branch collaborative architecture and low-dose noise suppression module are constructed. A composite reward function is designed to guide the agent to learn adaptive denoising strategies. Experiments have shown that HRDRL Net outperforms mainstream baseline methods in quantitatively reconstructing images on Mayo and Piglet datasets, providing a new solution for improving LDCT image quality.”