Segmentation Of Retinal Vessel Based On Deep Learning Technology

In recent years,with the growth of datasets and the improvement of computer computing power,deep learning technology has developed rapidly,and its application in the field of medical image analysis has become more and more extensive.Among them,the semantic segmentation technology plays a huge role in various application scenarios such as treatment planning,disease diagnosis,pathological research and so on.For medical images,in order to accurately identify the class of each object in the image,not only a medical expert with knowledge in the professional field is required,but also a certain amount of time for the professional authority.Through the study of semantic segmentation technology,it is possible to automatically and accurately segment the input medical images,so that doctors can make more accurate judgments and design better treatment plans.In this paper,focusing on the problem of retinal vessel segmentation,a segmentation model based on residual recursive convolution and pyramid pooling and a segmentation model based on pyramid dilated convolution are proposed,the main research contents are as follows:(1)The data preprocessing and block extraction methods of retinal images are proposed.The retinal image datasets used in this paper are obtained by color fundus photography technology,they have the disadvantages of low contrast and large influence by equipment,so a data preprocessing methods for retinal images are proposed to improve image quality.At the same time,the public datasets of retinal images are generally relatively small,it is difficult to meet the requirements of network model training on the size of the dataset.Therefore,in this paper,the block extraction method is adopted to achieve data enhancement.(2)A segmentation model based on residual recursive convolution and pyramid pooling is proposed.By using recursive convolution instead of ordinary convolution,it is beneficial to extract richer semantic information.By using a residual network,it can avoid the risk of gradient disappearance when the parameter is updated,and accelerate the convergence rate of the network.And by integrating the pyramid pool module,it is beneficial to multi-scale extraction of context information and deeper semantic information in retinal images to further improve the segmentation results.(3)A segmentation model based on pyramid dilated convolution is proposed.Because the retinal image has the characteristics of low contrast and contain many fine blood vessels;And the traditional pooling operating will cause the image resolution to decrease,resulting in theloss of detailed information.This paper has designed a pyramid dilated convolution module.It realizes the extraction of richer deep semantic information while reducing the loss of shallow detailed information.Through an experimental comparison on the two public retinal datasets of DRIVE and STARE,it was found that compared with many existing retinal vessel segmentation methods,the segmentation model based on residual recursive convolution and pyramid pooling and the segmentation model based on pyramid dilated convolution proposed in this paper all have better performance.