School of Computer Science and Engineering, Central South University, Changsha, 410083, China.
The Second Xiangya Hospital of Central South University, Changsha, 410011, China.
J Med Syst. 2019 May 1;43(6):163. doi: 10.1007/s10916-019-1303-8.
Glaucoma is an eye disease that damages the optic nerve and can lead to irreversible loss of peripheral vision gradually and even blindness without treatment. Thus, diagnosing glaucoma in the early stage is essential for treatment. In this paper, an automatic method for early glaucoma screening is proposed. The proposed method combines structural parameters and textural features extracted from enhanced depth imaging optical coherence tomography (EDI-OCT) images and fundus images. The method first segments anterior the lamina cribrosa surface (ALCS) based on region-aware strategy and residual U-Net and then extracts structural features of the lamina cribrosa, such as lamina cribrosa depth and deformation of lamina cribrosa. In fundus images, scanning lines based on disc center and brightness reduction are used for optic disc segmentation and brightness compensation is utilized for segmenting the optic cup. Afterward, the cup-to-disc ratio (CDR) and textural features are extracted from fundus images. Hybrid features are used for training and classification to screen glaucoma by gcForest in the early stage. The proposed method has given exceptional results with 96.88% accuracy and 91.67% sensitivity.
青光眼是一种损害视神经的眼部疾病，如果不治疗，会逐渐导致周边视力不可逆转地丧失，甚至失明。因此，早期诊断青光眼对于治疗至关重要。本文提出了一种自动早期青光眼筛查方法。该方法结合了增强深度成像光相干断层扫描（EDI-OCT）图像和眼底图像提取的结构参数和纹理特征。该方法首先基于区域感知策略和残差 U-Net 对前梁表面（ALCS）进行分割，然后提取梁的结构特征，如梁的深度和梁的变形。在眼底图像中，基于视盘中心和亮度降低的扫描线用于视盘分割，并且亮度补偿用于分割视杯。之后，从眼底图像中提取杯盘比（CDR）和纹理特征。混合特征用于 gcForest 的早期训练和分类，以筛查青光眼。该方法的准确率为 96.88%，灵敏度为 91.67%，结果非常出色。
