Abdominal Medical Image Segmentation Based On A U-Shaped Network Model

The abdominal organs are the golden point of human organs,undertaking the functions of digestion,absorption,secretion,metabolism and detoxification.However,with the changes in people’s living environment and the formation of unhealthy dietary habits,the extent of pathological changes in abdominal organs has become an important factor threatening human health and safety.The rapid development of science and technology has made computer tomography the most commonly used medical imaging method in the diagnosis and treatment of abdominal organ diseases.With the rise of computer vision technology,the fusion of computer-aided technology and medical imaging technology for abdominal medical image segmentation has become a long-term international research hotspot.It accurately segments abdominal medical images through machine learning,pattern recognition,and other methods,which not only helps doctors accurately diagnose the location of the patient’s lesion and tumor size.Abdominal medical image segmentation can assist doctors in preoperative planning and intraoperative guidance,providing them with more accurate diagnosis,treatment,and surgical options.It plays an important role in improving the accuracy and efficiency of clinical diagnosis,promoting medical research and education,and promoting automatic processing of medical images.Due to significant differences in the morphology and distribution of different tissues and organs in medical images.In medical organs,the boundaries between tissues are fuzzy and the morphology is complex.Medical image segmentation methods based on deep learning have great challenges.It is more difficult to segment some small lesions.At present,although researchers have explored a series of medical image segmentation methods based on U-shaped networks,due to the need to consider factors such as data complexity,noise,and image artifact interference,resulting in a large amount of network parameters and computational complexity,and the need to use a large amount of data for training to obtain better segmentation results.Therefore,this paper studies the segmentation method of abdominal medical images based on U-shaped networks,mainly to solve the problems of the current mainstream U-shaped networks in medical image segmentation,such as low accuracy,large amount of model parameters,and difficulty in model training for liver,kidney,and tumor.The main research contents include:(1)Due to the current mainstream of U-shaped convolutional neural networks,which segment abdominal medical images through convolution checks with fixed geometric shapes,the shape information of organs is missing,and the ability to extract information from liver and liver tumors with complex shapes is greatly limited.At the same time,these network parameters are redundant and difficult to train.This paper proposes a lightweight deformable codec network(LDNet).By using lightweight deformable convolution,the network can adaptively change the shape of the convolution core according to the self structure of the liver and liver tumors,match irregular organ shapes,further enhance the network’s modeling and geometric transformation capabilities,and ultimately improve the segmentation accuracy of abdominal medical images.Secondly,by introducing deep separable convolutions into the offset layer,the computational complexity of the network is reduced and the flexibility of network deployment is enhanced.In addition,in order to better capture context features,a collaborative coupling attention mechanism is proposed,which uses two one-dimensional codes to learn the correlation between medical image channels and space,tightly coupling channel information and spatial information,thereby enabling the acquisition of more detailed information in medical images.The final experiment shows that LDNet exhibits a more accurate and efficient segmentation effect on liver and liver tumor organs.(2)Because convolutional neural networks lack the ability to capture longdistance dependencies,they cannot meet the demand for accurate segmentation of small target organs such as liver,kidney,and kidney tumors in medical applications.This paper proposes a dual branch abdominal medical image segmentation network based on sparse dynamic transformer(SDU-TransNet).The network utilizes a dual scale parallel attention fusion module to interactively fuse the dual branch network information of the convolutional neural network and the Transformer network,and then uses dual channel attention to weight and sum the acquired dual branch feature information,thereby achieving complementarity between the shallow global information and deep detail information of medical images.SDU-TransNet can not only capture the global relationship modeling of medical images,but also capture richer feature information.In addition,in order to reduce the computational burden brought about by the Transformer structure,a sparse dynamic token learning model is used to weight the input data through the spatial domain,enabling the network to allocate limited token resources to effective regions,thereby suppressing the number of tokens in regions with invalid features.Thereby generating the number and location of tokens in a sparse dynamic learning Transformer structure,while reducing storage and computing overhead,and improving the operational efficiency of the model.Experiments show that the proposed dual branch abdominal medical image segmentation network based on sparse dynamic transformer can bring more accurate segmentation accuracy for abdominal medical images of small organs such as kidneys,kidney tumors,and liver tumors.