Research On Kidney And Tumor Segmentation Algorithm Of CT Image Based On Deep Learning

Kidney cancer is one of the most common malignant tumors,which brings great damage to people’s health.The commonly used method in current clinical treatment is surgical resection,mainly radical nephrectomy and nephron-sparing surgery.The choice of surgical method can be based on the results of the evaluation of renal function,which requires accurate tumor location and calculation of tumor removal after renal function.The location of renal tumors can be located in the enhanced CT images of the abdomen.However,SPECT renal dynamic imaging is required for the evaluation of renal function.The images of these two modalities can be established by plain CT.The segmentation of scanned CT images was studied.At present,the segmentation of CT images is performed manually by experienced experts.This manual segmentation method will consume a great many of time and vigour,and the results will be influenced by subjective factors.If there is a method for automatic segmentation of kidney and tumor,this will greatly promote the development of clinical diagnosis.With the continuous development of deep learning in the field of medical image segmentation,computer-aided diagnosis systems have become a trend.Compared with traditional segmentation methods,deep learning-based medical image segmentation methods can automatically learn image semantic features without too much manual intervention.This paper builds a CT image segmentation model based on deep learning.The main work contents are as follows:（1）A three-stage method based on convolutional neural network is proposed for the segmentation of kidney and tumor from enhanced CT images.First,the Mask R-CNN network is used to summarize the slices containing the kidney to narrow the target range;then,the kidney and the tumor are segmented and the slices containing the tumor are aggregated.Based on U-Net,downsampling increases dense connections and upsampling Using the network of bicubic interpolation,more accurate global location features and local detail features were obtained;then,the tumor segmentation was continued,and the results were fused with the previous stage;finally,the segmentation results were further optimized using a three-dimensional connected domain-based method.（2）A two-stage method based on convolutional network is proposed for the segmentation of kidney（including tumor）in plain CT images.First,collect plain CT images,make data sets,and complete data annotation,and then perform preprocessing operations according to image features.This method is consistent with the enhanced CT segmentation method.First,the Mask R-CNN network is used to summarize the slices containing the kidney,and then the U-Net network is used to increase the dense connection to segment the kidney.Due to the low contrast between the kidney and other surrounding tissues in the image,the characteristics of It is not obvious enough.Design a multi-scale feature extraction module and introduce this module before upsampling,so as to facilitate the acquisition of image features under different receptive fields,fully combine global and local semantic information,and finally optimize the segmentation results through post-processing operations.In this article,a multi-stage segmentation method is proposed for different CT images.It is proved by experiments that the proposed method can improve the segmentation accuracy.