Design And Implementation Of Automatic Kidney Segmentation System In CT Images

Computer Tomography(CT)has become one of the most commonly used methods in clinical renal imaging.The accurate and reliable results of kidney segmentation in CT images are important in clinical diagnosis and treatment of renal diseases.However,there are some challenges in kidney segmentation in CT images.Firstly,mainstream manual segmentation methods and semi-automatic segmentation methods need not only high manual operation workload,but also support from experience of doctors at the professional level.The segmentation results can hardly be reproduced due to subjective factors in the operating process.Secondly,being influenced by contrast media,the CT value of renal arteriovenous vessels appears to be highly similar to that of the kidney,while the latter is no longer fixed in a stable interval.Facing the above challenges,this thesis studies automatic segmentation methods on the basis of kidney morphological characteristics of CT images.Firstly,in order to obtain kidney location,the YOLOv3 algorithm is used to detect the approximate location of the kidney in CT images.And then the improved region growth algorithm is used to obtain a more accurate kidney location with YOLOv3 algorithm’s results.Secondly,this thesis proposes a SDRLSE algorithm,which is an improved distance regularized level set evolution algorithm in the consideration of structural similarity and target area contrast.This algorithm obtains the kidney segmentation results layer by layer using the contextual connection of CT sequences.Finally,a three-dimensional distance regularized level set evolution algorithm is used to optimize kidney segmentation results.Based on the above methods,this thesis designs and implements an automatic kidney segmentation system.The system is then applied to the segmentation of clinical contrast-enhanced CT images.Experimental results show that the system obtains an average dice coefficient of 0.967 and an average intersection-over-union of 0.936,which are obviously better than the results of other segmentation algorithms.The system’s segmentation efficiency is improved by GPU,with which the speed is 3 times higher than CPU mode.As a conclusion,the automatic kidney segmentation system realizes automatic segmentation,obtains accurate and robust segmentation results of kidney,cortex and medulla,and verifies the stability and reliability of its results through the test on clinical CT images.