Chronic Kidney Disease Instance Segmentation Model Based On Deep Learnin

Chronic Kidney Disease(CKD)is a public health disease with a high incidence worldwide.In recent years,the prevalence rate of chronic kidney disease in Chinese adults has reached 10.8%,and the burden of CKD disease is constantly rising.Deep learning is an artificial intelligence technology that automatically extracts high-dimensional feature information.The accuracy of medical image instance segmentation technology based on deep learning has exceeded the traditional image processing method,and can greatly reduce the complexity of manual measurement and improve the objectivity and accuracy of diagnosis.There are mainly the following problems in deep learning-based case segmentation of chronic kidney disease: For multi-objective case detection with a small gap,the model is difficult to distinguish the difference between objects,resulting in misjudgment;There are a lot of artifacts and noise in ultrasonic images,which makes the edge of the segmentation result not accurate enough and the segmentation result is rough.Few data sets can not meet the high precision requirements of complex tasks;In ultrasonic images of kidney,the shape of kidney is variable,and the size and position of kidney in cross section and longitudinal section images are very different,which leads to the expansion of the scope of model learning,and it is difficult to accurately learn the effective features for the task.Due to the existence of these problems,the case segmentation results of baseline model are difficult to meet the actual demand,the detection of chronic kidney disease has a high misjudgment rate,and the segmentation of renal boundary is not accurate enough.In view of the current chronic kidney disease diagnosis is time-consuming and laborintensive,artificial diagnosis is difficult,and the recognition and segmentation accuracy is low,this paper proposes a case segmentation model that can combine the glomerular filtration rate index and image texture features for comprehensive analysis,which improves the Mask RCNN framework,and improves the accuracy of chronic kidney disease stage detection and kidney segmentation.At the same time,a data set of chronic kidney disease is established and compared with other advanced case segmentation models to verify the effectiveness of the proposed model.The main research work of this paper is as follows:(1)An instance segmentation model of chronic kidney disease based on SCAFF-Mask was proposed.In view of the low detection accuracy and inaccurate segmentation of multiscale chronic kidney disease images by baseline network,the feature pyramid network structure was improved and features of different scales were adaptively fused,so as to obtain effective information from different semantic levels and provide a more accurate scheme for regional proposals.With the addition of attention module,the effective information of features is retained from both channel and space,and the overall accuracy of the model is improved.Compared with baseline networks and other Specificity studies,the model demonstrated varying degrees of improvement in m AP,IOU,Dice,Precision and specificity.(2)A Boundary-Aware SCAFF-Mask model for boundary optimization is proposed.In order to further improve the segmentation quality of the boundary region,a mask boundary learning branch is proposed,which adopts multi-magnification hollow convolution to extract features and fully integrate semantic information and detail information.The Loss function combining Dice Loss and Mean Square Error(MSE)was used to monitor the target boundary,and the boundary points that were difficult to be classified were further processed.Compared with the SCAFF-Mask based instance segmentation model of chronic kidney disease and other advanced instance segmentation models,the parameters of IOU,Dice,Precision and Specificity of the model in this chapter were further improved.