Comparison of Semantic Segmentation Methods on Renal Ultrasounds Images

There are many people with chronic kidney disease in China, so manual segmentation can not meet the huge social needs. Due to the ability to accurately segment the images, deep learning methods can be used for the detection of kidney diseases. In this paper, a total of 881 renal ultrasound images were collected and labelled. Four semantic segmentation networks, including FCN, U-Net, SegNet and Deeplab were used to segment renal ultrasound images. In order to measure the segmentation effect of different networks, two common indicators, PA and IoU, were used to evaluate the results. The results showed that all the four semantic segmentation networks achieved good results in renal ultrasound image segmentation, among which Deeplab had the best effect on the test set, with PA reaching 99.14% and IoU reaching 0.8219.