基于深度学习的白内障自动诊断方法研究

Research on Automatic Diagnosis Method of Cataract Based on Deep Learning

针对目前白内障自动筛查以人工寻找特征为主,特征选取主观性强的问题,根据深度学习理论,建立了基于卷积神经网络的白内障特征自动学习模型。为克服当前数据集缺乏的问题,构建一个具有16239张图片,包含多种类型白内障并按病情严重程度分为三类的MSLPP数据集;利用卷积神经网络自动提取深度特征,并采用迁移学习的思想来训练网络,降低模型过拟合。实验结果表明模型识别准确率达94.84%。该模型对于核性白内障,皮质性白内障和后囊性白内障普遍适用,满足实用化要求。

According to the theory of deep learning,an automatic cataract feature learning model based on convolution neural network is established to solve the problems of manual feature searching and subjective feature selection.In order to overcome the shortage of current data sets,an MSLPP data set with 16,239 pictures,including various types of cataracts and divided into three categories according to the severity of the disease is constructed.The convolution neural network is used to automatically extract depth features,and the idea of transfer learning is used to train the network to reduce over-fitting of the model.The experimental results show that the model recognition accuracy rate is 94.84%.The model is generally applicable to nuclear cataract,cortical cataract and posterior cystic cataract,and meets the practical requirements.