深度学习在糖尿病视网膜病变诊疗中的应用

Research progress of deep learning in diagnosis and treatment of diabetic retinopathy

基于深度学习的影像组学算法在计算机辅助诊断领域展现出了高效、准确等诸多优点.归纳了卷积神经网络的基本模型并对其在糖尿病视网膜病变辅助诊断的相关研究进行了总结.基于深度学习的影像组学算法可以快速、高效、准确地实现糖尿病视网膜病变的诊断、分类及预后预测,但图像质量不可控、图像数量少等问题影响了算法的进一步优化.提出了规范公开数据集,建立图像质量标准以及利用迁移学习、生成对抗网络对图像进行增强等解决方法,为糖尿病视网膜病变眼底图像研究人员提供建议与参考,以促进深度学习在糖尿病视网膜病变眼底图像中更进一步的应用.

Imaging omics algorithm based on deep learning shows many advantages in the field of computer-aided diagnosis,such as high efficiency,accuracy and so on.The basic model of convolution neural network is summarized,and the related research on assistant diagnosis of diabetic retinopathy is summarized.Deep learning based image grouping algorithm can rapidly,efficiently and aceurately diagnose,classify and predict the prognosis of diabetic retinopathy.However,the problems such as uncontrollable image quality and small number of images affect the further optimization of the algorithm.This paper proposes the methods of regulating opendata sets,establishing image quality standards,and using migration learning,generating confrontation network to enhance image and so on.It will provide suggestions and references for researchers in diabetic retinopathy fundus image,so as to promote deep learning in diabetic retinopathy.