[Super-resolution construction of intravascular ultrasound images using generative adversarial networks].

• DOI: 10.12122/j.issn.1673-4254.2019.01.13
• Publication Date: 2019-01-30
• Authors: Yangyang Wu; Feng Yang; Jing Huang; Yaqin Liu
• Keywords: Feature (linguistics); Discriminator; Similarity (geometry); Convolution (computer science)

The low-resolution ultrasound images have poor visual effects. Herein we propose a method for generating clearer intravascular based on super-resolution reconstruction combined with generative adversarial networks. We used the networks to generate by generator and estimate authenticity of discriminator. Specifically, image was passed through sub-pixel convolution layer r2-feature channels maps in same size, followed realignment corresponding pixels each feature map into r ×r sub-blocks, which corresponded sub-block high-resolution image; after amplification, an r2-time resolution generated. can obtain continuous optimization. compared (SRGAN) other methods including Bicubic, convolutional network (SRCNN) efficient (ESPCN), proposed resulted obvious improvements peak signal-to-noise ratio (PSNR) 2.369 dB structural similarity index 1.79% enhance diagnostic effects images.