Ultrasound Image Super-Resolution Reconstruction Based On Generative Adversarial Network

Ultrasound imaging is currently the most frequently used medical imaging method in clinical medical treatment.It has real-time and portability and occupies the most important position in imaging technology.Ultrasound imaging technology,especially B-mode ultrasound,has brought great help to doctors in clinical medical diagnosis.The clinically commonly used desktop B-ultrasound diagnostic equipment is large in size and expensive,and is only suitable for clinical diagnosis scenarios,and cannot perform diagnosis and treatment operations at the scene of the accident or at the patient's home.The rapid development of embedded technology in recent years has given birth to a series of portable ultrasound diagnosis and treatment equipment that can be flexibly used in various complex diagnosis and treatment scenarios.Meanwhile,there has been a significant reduction in the volume of ultrasound diagnostic equipment,resulting in a lack of quality of ultrasound images generated by portable ultrasound diagnosis equipment,and a lot of noise,including certain electromagnetic interference noise and background noise.Based on this background,this article has launched a super-resolution reconstruction study based on a generative confrontation network on low-resolution ultrasound images to help doctors use portable ultrasound equipment for more accurate disease diagnosis.The main research of thie paper is the super-resolution reconstruction of ultrasound images based on generative confrontation networks.The main contributions of this article are as follows:1)Most of the existing super-resolution reconstruction models are suitable for natural images,and there is a lack of research on ultrasound images.This paper design a super-resolution reconstruction model for ultrasound images based on generative confrontation networks that can be applied to ultrasound images,which effectively improves The image quality of low-resolution ultrasound images generated by portable ultrasound equipment.2)For the purpose of simulate the complex degradation process more closer to ultrasound images,this paper proposes a high-order degradation model for ultrasound images,and uses a sinc filter to simulate the common ringing and overshoot artifacts in ultrasound images,and constructs An ultrasound image data set containing matching pairs of high-resolution and low-resolution ultrasound images is used to train the model.3)The research on ultrasound images needs to solve the image degradation space that is more complicated than natural images.The existing discriminator design is no longer suitable for such complex situations.This paper uses the U-Net discriminator with spectrum normalization function to improve The discriminative ability of the discriminator and at the same time ensure the dynamic stability of training.