白内障是一种严重影响人类视觉功能的眼科疾病。为准确评估白内障患者的视力等级,提出了一种基于高效通道注意力的白内障视力分级算法(efficient channel attention deep residual netowrk,ECRN) 。该算法首先使用限制对比度自适应直方图均衡化算法(contrast limited adaptive histogram equalization,CLAHE) 对眼底图像进行预处理,增强图像中的血管、视盘和黄斑的关键特征。然后,将高效通道注意力机制(efficient channel attention,ECA) 和深度残差网络相融合,关注与视力等级相关的眼底组织和病变区域。为解决眼底图像数据集不平衡的问题,引入焦点损失(focal loss,FL) 函数为优化目标,使模型偏向于视力等级严重的患者。该算法在临床数据上进行了实验,正常、中等视力白内障和低视力白内障3个类别的准确率分别为98.3%、90.5%和92.1%,实验结果表明,该算法在白内障视力分级上表现出良好的性能。

Cataract is a severe ophthalmic disease that significantly affects human visual function.To precisely assess the grading of visual acuity afflicted by cataracts,we propose a cataract visual acuity grading algorithm based on the efficient channel attention technique.This algorithm first employs the contrast limited adaptive histogram equalization (CLAHE) technique to preprocess fundus images,enhancing critical features such as blood vessels,optic disc,and macula.Subsequently,the efficient channel attention (ECA) mechanism is fused with a deep residual network to focus on fundus tissues and lesion areas relevant to visual acuity grading.To address the challenge of imbalanced dataset of fundus images,a focal loss (FL) function is introduced as the guiding objective for optimization,biasing the model towards patients with severe visual acuity.The algorithm is experimented with clinical data,achieving accuracies of 98.3%,90.5%,and 92.1% for normal vision,moderate vision cataract,and low vision cataract,respectively.The experimental results demonstrate that the proposed algorithm exhibits excellent performance in cataract visual acuity grading.

TP391.4

国家自然科学基金(62276210,61775180,82201148)、 陕西省自然科学基础研究计划(2022JM-380)和西安邮电大 学研究生创新基金(CXJJYL2022032) 资助项目