Design And Implementation Of Fundus Vascular Segmentation Auxiliary Diagnosis System Based On Encoder-decoder

Studies have shown that the early symptoms of many eye diseases and systemic diseases can be observed in retinal blood vessels.The distribution and morphology of retinal blood vessels can be analyzed and studied by medical image segmentation,which can provide an important basis for the diagnosis of these diseases.The segmentation of retinal blood vessels by computer can improve the efficiency of segmentation tasks and greatly reduce the workload of doctors,which has far-reaching significance for modern medical auxiliary diagnosis.U-Net is one of the most representative encoder-decoder architecture networks in the field of medical image segmentation in recent years,which has had good performance in retinal vessel segmentation tasks.However,the unreasonable structure of U-Net affects the performance of the network,such as the semantic gap between upper and lower layers when the network only implements the feature channel connection of the same layer,the single feature extraction method of the encoder cannot obtain multi-scale feature information,and the contextual information is not fully utilized in the feature fusion process.To address the above problems,this thesis improves based on U-Net and does related experiments,and the main research work is as follows:(1)An encoder-decoder structure segmentation model based on multiple attention mechanisms is proposed and applied to study the segmentation of retinal vascular images.By making full use of attention mechanisms such as space,channel,and scale,the improved model assigns weights to effective feature channels,highlights the salient regions related to the segmentation task target,suppresses the background information or unrelated parts of the image,and makes full use of the feature information in the relevant feature channels to achieve feature recalibration.Comparative experiments show that the improved model can enhance the representation of features and improve the segmentation performance.(2)A PAFE-based multi-scale feature extraction segmentation model is proposed and applied to study the segmentation of retinal vascular images.By adding the Multi Res module to the encoder part of the network,the normal convolution operation in the original convolution layer is replaced to achieve feature extraction from different scales,and the PAFE module is added to the bridge part of the encoder and decoder to expand the perceptual field of the feature map and obtain richer contextual information to achieve optimization of the segmentation performance of the model.The comparison experiments show that the proposed model enhances the ability of multi-scale feature extraction and verifies that the model segmentation performance is effectively improved.(3)An auxiliary diagnosis system based on the encoder-decoder fundus vessel segmentation model is designed and implemented.The main functions of the system are to manage case information and automatically segment fundus vascular images.According to the input retinopathy image,the vascular image is automatically processed and segmented.By comparing different processing results,the diagnostic efficiency of doctors is effectively improved and the workload is reduced.