基于深度学习的眼底图像辅助诊断系统设计

Design of Fundus Image Aided Diagnosis System Based on Depth Learning

目前多数眼底图像分类方法主要针对单一类别疾病不同级别进行分类,且网络模型存在参数量大、计算复杂等问题。基于轻量化的SqueezeNet深度学习模型,设计了一种实现多标签眼底图像辅助诊断系统。在公开数据集ODIR上进行训练和验证,实验结果表明该系统具备强大的特征提取能力,可以在非大幅降低模型精度的情况下最大程度地提高系统的运行速度、降低系统内存消耗。该基于深度学习的眼底图像辅助诊断系统适合于硬件部署,可以为人工智能应用于计算机辅助医疗设备中提供一定的参考价值。

At present,most fundus image classification methods are mainly used to classify a single category of diseases at different levels,and the network model has many problems such as large parameters and complex calculation.Based on the lightweight SqueezeNet depth learning model,a multi label fundus image aided diagnosis system is designed.The training and verification are carried out on the public data set ODIR.The experimental results show that the system has a strong feature extraction ability,which can maximize the system’s running speed and reduce the system’s memory consumption without significantly reducing the model’s accuracy.The fundus image aided diagnosis system based on deep learning is suitable for hardware deployment and can provide certain reference value for the application of artificial intelligence in computer-aided medical devices.