Research On Algorithms For Medical Image Segmentation Based On Deep Learning

Medical imaging is an essential part of modern healthcare,as it helps doctors to gain a more intuitive understanding of a patient’s condition and make accurate diagnoses and treatment plans.With advances in technology,medical imaging equipment is constantly evolving,and the application of medical imaging is becoming increasingly widespread.However,medical segmentation remains a challenging task in image processing due to the diversity of imaging device types and the small difference between the background and foreground of medical images.The development of computer technology has made the analysis and processing of medical imaging more convenient,and medical segmentation has become a key step in medical image processing.Through the analysis of medical segmentation results,doctors can better understand the patient’s condition and propose accurate diagnoses and treatment plans.Therefore,medical segmentation technology has become an indispensable step in the field of healthcare.In recent years,the sudden advancement of deep learning has led researchers to apply deep learning to the field of medical segmentation,enabling neural networks to automatically complete the learning of image features in segmentation tasks without relying on human hands.Therefore,the medical image segmentation technology based on deep learning has gradually become the mainstream choice for segmentation at present.However,with the progress of science and technology,the existing medical image segmentation algorithms based on deep learning have gradually failed to meet the current needs of medical image segmentation and there are a number of problems.For example,semantic differences between codecs,learning of low-level features and poor generalisation in small-scale datasets are problems that need to be addressed.To solve these problems,new techniques such as the introduction of attention mechanisms and cross-fusion modules can be used to reduce the semantic differences between codecs.For the learning of low-level features,iterative sampling methods and pyramid pooling modules can be used to improve the accuracy of segmentation.In addition,a gated position-sensitive axial attention mechanism can be introduced to improve the performance performance on small data sets.The main research in this paper is as follows:(1)A medical image segmentation model based on a U-Net variant network has been proposed,which includes two branches: Se U-Net and Trans-Net.Se U-Net improves the skip connections of U-Net by introducing channel-crossing fusion Transformer and channelcrossing attention to address the semantic gap between the encoder and decoder.In the TransNet network,the encoder and decoder of this branch are redesigned to learn low-level features such as edge details.SE attention is also introduced at the skip connections to assign higher weights to important features.Furthermore,the two networks are fused through a crossresidual fusion module to fully leverage the strengths of both networks.Finally,through qualitative and quantitative comparisons,the proposed method achieves good segmentation results on both large and small datasets.(2)A medical image segmentation model based on iterative sampling is proposed,which divides the segmentation process into two parts: global structure and local structure.The global structure utilizes pyramid pooling modules and stripe convolution modules to enhance the network’s feature extraction ability,while the local structure employs a progressive sampling module to focus on segmentation of the region of interest.Additionally,the model incorporates a gated position-sensitive axial attention mechanism,enabling the network to perform exceptionally well even on small-scale datasets.Experimental results demonstrate that the proposed model performs well in the field of medical image segmentation,outperforming existing models in terms of both accuracy and robustness.