目的 视网膜血管健康状况的自动化分析,对糖尿病,心脑血管疾病及各种眼科疾病的快速无创诊断具有重要参考意义。但由于视网膜图像中血管网络结构复杂,光照不均,使得视网膜血管区域的准确自动提取极具挑战性。本文通过使用全连接深度神经网络,实现视网膜血管的高精度分割。方法 利用深度学习技术,结合U-net网络中端到端的图像特征提取思想和Dense-net网络中稠密连接的思想,提出了一种基于监督学习的视网膜血管分割方法。首先对使用白化预处理对图像数据进行降噪,并通过对每一幅图片进行旋转、Gamma变换等操作实现数据增广,然后将每一幅图片分割成若干较小的图块,用于模型的训练。结果 本文使用灵敏度(0.7409)、特异性(0.9929)、准确率(0.9707)和AUC(0.9834)等性能指标对训练之后的模型进行了综合评定,并和目前主流的算法进行了比较,对比结果显示该方法的各项性能均表现良好。结论 结果表明本文提出的方法在视网膜血管分割上能达到较好的效果,具有较好的应用价值。
Objective Automated analysis of retinal vascular health status has important reference significance for rapid non-invasive diagnosis of diabetes, cardiovascular and cerebrovascular diseases and various ophthalmic diseases. However, due to the complex structure of the vascular network in the retinal image and uneven illumination, accurate and automatic extraction of the retinal vascular region is extremely challenging. This paper achieves high-precision segmentation of retinal vessels by using a fully connected deep neural network. Method Using deep learning technology, combined with the idea of end-to-end image feature extraction in U-net network and the dense connection in Dense-net network, a retinal vessel segmentation method based on supervised learning is proposed. Firstly, the image data is denoised by using whitening preprocessing, and the data is augmented by rotating and gamma transforming each picture, and then each picture is divided into several smaller pieces for the model. Training. Result In this paper, the model after training is comprehensively evaluated using performance indicators such as sensitivity (0.7409), specificity (0.9929), accuracy (0.9707) and AUC (0.9834), and compared with the current mainstream algorithms. The comparison results show that The performance of the method performed well. Conclusion The results show that the proposed method can achieve better results in retinal vascular segmentation and has good application value.