基于U-net的眼底视网膜血管语义分割研究

Research on Semantic Segmentation of Fundus Retinal Vessels based on U-net

眼底血管的健康状态对于研究各类眼科疾病具有重要的参考意义。为了帮助临床医疗人员对眼底微血管形态结构图像的分析来诊断疾病,文中提出了一种基于编码-解码(Encoder-Decoder)结构的U-net的眼底血管分割方法。首先,在模型训练之前对图像进行预处理,然后使用Leaky ReLU激活函数替换U-net ReLU,避免了神经元的死亡问题,同时使用Adam(Adaptive Moment Estimate)优化器代替梯度下降法优化学习策略,最后对血管分割的平均交并比进行计算评估。实验表明,优化后的模型的平均精度可达到93.29%,相比原算法提升了3.26%。

The health status of fundus vessels is of great significance to the study of various ophthalmic diseases.In order to help clinical medical personnel diagnose diseases by analyzing the morphological structure images of fundus microvessels,this paper proposes a method of fundus vessels segmentation based on Encoder-Decoder structure of U-net.First,images are preprocessed before model training,and the Leaky ReLU(Leaky Rectified Linear Unit)activation function is used to replace U-net ReLU during model training,avoiding neuron death.Meanwhile,Adam optimizer is used instead of gradient descent method,to optimize the learning strategy.Finally,the Mean Intersection over Union of the vessel segmentation is calculated and evaluated.Experimental results show that the average accuracy of the optimized model can reach 93.29%,which is 3.26%higher than the original algorithm.