Non-reference CT image quality evaluation method based on multi-scale features

• Main Authors: FENG RUI, SHEN QUANLI, ZHANG XIAOBO, CHEN ZHAOHAO, HE WEN, FU WEIJIA, SU LING
• Format: Patent
• Language: Chinese; English
• Published: 26.04.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention provides a non-reference CT image quality evaluation method based on multi-scale features. The method comprises the following steps: S1, generating CT image training sets with different qualities from a Mayo Clinic data set; and S2, based on the convolutional neural network, constructing a non-reference CT quality evaluation network. And S3, inputting the training set in the step S1 into the CT quality evaluation network, and training the network until the network converges. And S4, inputting a to-be-tested CT image into the trained CT quality evaluation network, and obtaining a quality score of the CT image. According to the method, a CT quality evaluation network combining multi-scale feature extraction, a structure evaluation branch, a feature evaluation branch and a quality coefficient branch can generate an accurate quality evaluation score for details and an overall structure in the CT image, so that quantitative evaluation of the quality of the CT image is realized. 本发明提供一种基于多尺度特征的无参考CT图像