Renal Cortex Segmentation With Convolutional Network And GrowCut

Kidneys are responsible for maintaining body balance and metabolism,are important organs in the human body.The volume and thickness of renal cortex are effective assessment criteria in early clinical diagnosis for renal neoplasms,chronic arteriosclerotic nephropathy,and acute rejection after kidney transplant.Most of the existing algorithms for kidney image segmentation only implement the whole kidney segmentation,and are aimed at only one modality.The application of deep learning in medical image segmentation mostly uses the method based on patch,which is inefficient.To solve these problems and realize the effective segmentation of the renal cortex,this paper proposes an algorithm of renal cortex segmentation based on convolutional network and GrowCut,which can achieve multimodal renal image segmentation and improve the accuracy and efficiency.In this paper,we use the transfer learning strategy to apply the convolutional network in renal image segmentation,and divide the task of renal segmentation into two parts: the region of interest extraction and the renal cortex segmentation.In this paper,the detection algorithm of kidney based on generalized Hough transform and convolutional network is studied to realize the automatic extraction of the region of interest.In the study of detection algorithm based on convolutional network,the parameters of network are need to be modified and the pre-trained model based on target detection network is used for transfer learning.The cortex segmentation algorithm based on the fully convolutional network is studied.A fully convolutional network with fewer parameters and more suitable for kidney segmentation is constructed,the segmentation accuracy of this network is 0.88.The renal cortex segmentation algorithm which combines the fully convolutional network and GrowCut is studied.The seeds are marked automatically based on the feature maps of the last layer of the fully convolutional network.GrowCut can further optimize the segmentation based on this marker,and improve the accuracy to 0.92.The proposed renal cortex segmentation algorithm based on convolutional network and GrowCut proposed in this paper can segment multimodal renal images,including normal and mutated renal images,indicating that the algorithm is better than the mainstream algorithms.The segmentation results have smoother edges and less missegmentation,indicating that the algorithm can provide a reliable basis for clinical diagnosis.