Renal Ultrasound Image Segmentation Based on U-Net and Generative Adversarial Nets
Weite Feng, Jinyu Liu, Zixu Guan, Zihao He, Gongping Chen, Yu Dai · 2022
Ultrasound images are quite useful for doctors to diagnose the kidney diseases nowadays. Since the U-Net network was proposed, it has been applied in the field of ultrasonic image segmentation. We propose a method that combine the U-Net with the generic advantageous networks, creating a more efficient approach of the segmentation of renal ultrasound images. By introducing GAN's supervision mechanism, the segmented network has a better performance. We constructed a training set of 200 renal ultrasound images and a test set composed of 40 images, and carried out experiments on them. Experiments show that compared with the traditional U-Net, the method proposed in this paper improves the accuracy on the same test set. The accuracy of our network reaches 97.83%, which is higher than 97.00% of U-Net.