Research On Kidney Segmentation Of CT Images Based On Deep Learning

Since kidney cancer is not sensitive to radiotherapy,surgery is the first choice for kidney cancer treatment.With the progress of science and technology,in addition to the method of radical nephrectomy,nephron sparing surgery(NSS)has been increasingly adopted.This surgical method retains the normal nephron as much as possible while removing the diseased tissue to improve the patient's postoperative quality of life.During the implementation of this operation,it is necessary to accurately locate the kidney and tumors.Block renal arteries which offer bloods to renal tumors,obtain a clear surgical field of vision,and then perform renal tumor cutting.At present,the location of kidney and kidney tumors has been manually calibrated by experienced experts.Due to the large amount of CT images of the kidneys,manual segmentation has brought tremendous workload to doctors.In addition,the incidence of kidney cancer has gradually emerged in recent years.With the rising trend of incidence rate of renal tumor,manual segmentation cannot meet the huge social needs.There is an urgent need for an automated segmentation algorithm to achieve automatic segmentation of kidneys and tumors.For different patients,renal and renal tumors have large morphological differences,and there are many types and different pathological characteristics of renal tumors,which all make the traditional methods of segmentation have greater limitations.This paper presents an automated segmentation method for kidneys and tumors based on convolutional neural networks.The method uses a three-dimensional full convolutional neural network,and incorporates a pyramid pooling and an global feature enhancement module.The pyramid pooling module can quickly enhance the receptive field of the network on the one hand,and can be better adapt to objects of different sizes on the other hand.The global feature enhancement module,by enhancing different features,has a better ability to differentiate between kidneys and kidney tumors.Because this method makes full use of the three-dimensional spatial information of the kidney data,compared with the two-dimensional convolutional neural network,it further enhances the recognition ability of kidneys and tumors and achieves better segmentation results.Experiments show that this method has greatly improved compared with the traditional methods.The dice coefficients of renal and renal tumors were 0.933 and 0.854,respectively.In addition,our method can also be used for the detection of renal tumors,and has also achieved good performance,reaching 97.5% in specificity and sensitivity.