用于湿性AMD辅助诊断的双模态深度学习模型

Dual-modal deep learning model for auxiliary diagnosis of wet age-related macular degeneration

为解决眼科医生阅片工作量过大的问题,将深度学习技术应用于湿性年龄相关性黄斑变性(AMD)辅助诊断领域.针对只考虑单一模态的医学影像且未将湿性AMD进行更细化的分类,构建了适用于深度学习的双模态湿性AMD数据集,并提出一种双模态湿性AMD辅助诊断模型Wet-AMD-Net.针对不同的特征提取模型与不同的特征融合策略进行实验,效果最优的模型受试者工作特征曲线下面积(AUROC)、召回率、精确度分别达到0.9881,0.9792和0.9821,超过了有多年工作经验的4位眼科医生的平均水平,用于临床辅助诊断具有实用性.

In order to solve the problem of excessive workload of ophthalmologists,the deep learning technology was applied to the field of auxiliary diagnosis of wet age-related macular degeneration(AMD).However,most of the work only considered single-modal medical images,and did not classify wet AMD in a more detailed manner,which is inconsistent with the standard procedure for clinical diagnosis of wet AMD,and is not conducive to the application.A dual-modal wet AMD dataset was constructed and a dual-modal wet AMD auxiliary diagnosis model—Wet-AMD-Net was proposed.Experiments were conducted on different feature extraction models and different feature fusion strategies.The model with the best effect reaches 0.9881,0.9792,and 0.9821 in AUROC(area under the receiver operating characteristic curve),recall,and accuracy,respectively,which exceeds the average level of 4 ophthalmologists with many years of experience.It has the practicality of clinical auxiliary diagnosis.