基于平稳小波与条件随机场的视网膜血管分割

Retinal Vascular Segmentation Based on Stationary Wavelet and Conditional Random Field

视网膜血管是眼科疾病与心血管疾病诊断中极为可靠的生物标志物,视网膜血管自动分割对于诊断和监测多种疾病都至关重要。本文提出了一种新的方法,将平稳小波变换(Stationary Wavelet Transform,SWT)提供的多尺度分析与多尺度条件随机场(Multi-scale Conditional Random Field,MCRF)相结合,检测视网膜血管结构的宽度和方向的变化。我们使用旋转不变性作为联合检测的基础,用于数据扩充和预测,从而得到在训练中学习到的信息,实现细化分割。该方法在DRIVE数据库上进行了评估,ROC曲线面积为0.9796,准确率为0.9569,相比于目前其他算法各性能指标有所提高。

Retinal blood vessels are very reliable biomarkers in the diagnosis of ophthalmic and cardiovascular diseases.Automatic segmentation of retinal vessels is very important for the diagnosis and monitoring of many diseases.A new method is proposed in this paper.The multi-scale analysis provided by stationary wavelet transforms(SWT)and multi-scale conditional random fields(MCRF)are combined to detect the width and direction of retinal vascular structure.We use rotation invariance as the basis of joint detection,and use it to expand and forecast data,so as to get the information learned in training and realize the segmentation.The method is evaluated on drive database.ROC curve area is 0.9796 and accuracy is 0.9569.Compared with other algorithms,the performance index has been improved.