Medical image target detection method and system

• Main Authors: WANG SHUO, PAN XIYAN, TIAN ZHIQIANG
• Format: Patent
• Language: Chinese; English
• Published: 20.12.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention discloses a medical image target detection method and system, and the method comprises the steps: building an interaction environment of a deep reinforcement learning agent according to a target detection task target in a medical image, and positioning a region where a specified organ is located through detecting a region of interest in a two-dimensional medical image; according to the Markov decision process suitable for the target detection task, a single-agent competition double-depth Q network structure is constructed, and a network model is subjected to competition structure improvement and action value function calculation optimization on the basis of a depth Q network, so that the stability and learning efficiency of the network model are enhanced; according to the invention, the training data of the deep neural network is obtained through interaction between the intelligent agent in deep reinforcement learning and the two-dimensional medical image environment, and the stability and effic