基于随机森林的频谱域光学相干层析技术的图像视网膜神经纤维层分割

Retinal Nerve Fiber Layer Segmentation of Spectral Domain Optical Coherence Tomography Images Based on Random Forest

频谱域光学相干层析技术是一种广泛应用于眼科疾病诊断的成像技术,而视网膜层分割对青光眼的诊断有很好的参考价值。该文利用随机森林分类器寻找视网膜层间单像素宽的边界,随机森林分类器由12个特征训练产生,其中相对灰度特征和邻域特征较好地解决灰度不均匀的分割误差大问题。对10组带有青光眼病变的视网膜图像进行分割,并与传统算法和Iowa软件进行比较,平均边界绝对误差为9.20±2.57μm,11.33±2.99μm和10.27±3.01μm。实验结果表明,改进算法可以较好地分割视网膜神经纤维层。

Spectral Domain Optical Coherence Tomography （SD-OCT） imaging technique is widely used in the diagnosis of ophthalmology diseases. The segmentation of retinal layers plays a very important role in the diagnosis of glaucoma. In this paper, a random forest classifier is used which is trained by twelve different features to find the boundaries between layers. What＇s more, the relative gray feature and the neighbor features are used to solve the problem of large errors under the condition of uneven illumination. In the last, the segmentation results of the proposed algorithm, a traditional algorithm and Iowa segmentation software on ten sets of retinal images are compared with manual segmentation, and the average absolute boundary errors are 9.20±2.57 μm, 11.33±2.99μm 10.27±3.01 μm, respectively. The experiments show that the proposed algorithm can segment the Retinal Never Fiber Layer （RNFL） better.