Segmentation Algorithm And Auxiliary Analysis System For Retinal Image Lesions

As one of the most important organs of the human body,the eye is responsible for information acquisition.About 80% of the knowledge and memory in the brain are acquired through the eye.With the increasing life expectancy of the population and the irregular habit of using eyes,a variety of eye diseases have appeared,including Central Serous Chorioretinopathy(CSC).CSC has a high incidence and recurrence rate in middle-aged men,and can cause irreversible damage after repeated recurrence.With the development of retinal imaging technology,Optical Coherence tomography(OCT)has become a common imaging method for the diagnosis of retinal diseases.In medical image analysis,the main form is to look at a number of two-dimensional image sections to find and map out the lesion area,which usually depends on the professional experience of the doctor to make judgments.However,the structure of retina is complex,especially when lesions occur,the image will have a large topological change.At the same time,manual outline is time-consuming and laborious,and the use of manual outline is very vulnerable to the influence of subjective elements,and the segmentation results cannot be guaranteed.The incidence of domestic eye diseases is also increasing year by year,but the number of doctors is limited.Therefore,many researchers began to focus on the development of automatic segmentation system of medical images,in order to provide better results for clinical diagnosis,through the segmentation and visualization of organs,tissues and pathological areas can effectively help doctors to carry out quantitative analysis of pathological areas.This paper focuses on the development of medical aided analysis and visualization system,and studies the segmentation algorithm and 3D modeling technology of retinopathy.The main work of this paper is summarized as follows:(1)A retinal image lesion region segmentation algorithm based on U-NET neural network with Attention mechanism was proposed.Based on U-NET,the Attention module was added to the residual learning branch of the residual network model to replace the original backbone network of U-NET.As a result,this model not only retains the advantage of fast convergence speed of network,but also solves the problem of weakened learning performance of deep network.By acquiring the relevance of each channel,the attention between channels is obtained,which enhances useful features and supresses noise features.(2)A computer aided analysis and visualization system is designed,and the segmentation algorithm and 3D model reconstruction are embedded into the platform.The main functions of the system include: image opening,image viewing,image saving,pathological region segmentation,retinal layer segmentation,3D model reconstruction,etc.Three-dimensional modeling of segmentation results was carried out using cross-platform and cross-language API.The pixel information of the 3D image is reconstructed into the 3D model,and the model is colored in a variety of ways.The model has a good effect and provides a good interactive experience.The segmentation results can be viewed from multiple angles by dragging the window.