Research On Image Segmentation Of Abdominal Organs Based On Deep Learning

Image segmentation is a key step in computer-aided diagnosis and surgical planning of abdominal organs.However,in the traditional method,radiologists usually use manual drawing of organs,which is time-consuming and laborious,reducing the efficiency of diagnosis and treatment.In recent years,with the development of medical image segmentation technology,computers are used to automatically segment patients' CT,MR and other abdominal images to assist doctors in diagnosing and treating,reducing the workload of radiologists.Because the image of abdominal organs is affected by uneven intensity,weak borders,noise and similar objects close to each other,it brings certain difficulties to the image segmentation of abdominal organs.How to extract deep features more accurately and achieve more accurate segmentation effects on the images of abdominal organs is still an area worth exploring.Therefore,this article studies the image segmentation technology of abdominal organs.The main work of the thesis is as follows:(1)Design and implement an image segmentation algorithm based on dual context aggregation and attention-guided cross deconvolution.In the convolutional neural network,each scale processes different information.Generally,shallow scales have more location information,and deep scales have richer semantic features.In order to better mine the deep feature information of abdominal organ images,an image segmentation algorithm based on dual context aggregation and attention-guided cross deconvolution is proposed.The dual context aggregation method is used to capture more fine-grained deep features,extract context features from each different scale,and merge the context feature information extracted from these different scales to provide richer prior information for the decoding process.The attention mechanism is introduced to obtain channel and spatial feature information of medical images,and cross deconvolution is used for decoding.Attraction-guided cross deconvolution is used to enhance the representation of feature semantics and improve the segmentation accuracy.Experimental verification and evaluation is performed on two public data sets.The proposed algorithm is used to segment the liver,left kidney,right kidney,and spleen on the abdominal image.The DSCs are 94.05%,90.65%,90.89%,and 90.99%respectively.(2)Propose and implement an image segmentation algorithm based on boundary enhancement guided packet rotation dual attention decoder.The automatic segmentation of abdominal organ images is an important task in clinical applications.However,due to the complexity of the organ background and blurred edges,some features are lost,which makes the segmentation effect poor.In order to solve these problems,the group rotation fusion module is used to initially extract the image feature information of the abdominal organs.Introduce the boundary enhancement guide group rotation dual attention decoder algorithm to enhance the boundary of the organ feature and effectively fuse the prior information.Through multi-resolution fusion,the high-resolution abdominal organ segmentation map is output,and the experimental verification and evaluation are carried out on two public data sets.The proposed algorithm is aimed at the segmentation of the liver,left kidney,right kidney and the spleen on the abdominal image.The DSCs are 94.23%,90.31%,91.28% and 91.10% respectively,which are better than other advanced algorithms.The results show that the algorithm has better segmentation performance.