一种眼病发展趋势的自动预测方法

Method for automatic prediction of the development trend of an ophthalmic disease

由于目前对眼病辅助诊断的研究都是基于当前阶段拍摄的影像进行的自动分类或分级诊断,对眼病的自动预测还非常稀少,故提出了一种基于代价敏感的时间序列预测模型,用于分析和预测眼病发展的趋势.该方法首先使用坎尼边缘检测算子和霍夫变换对裂隙灯图像进行预处理,获取晶状体病灶区域;然后使用残差卷积神经网络提取晶状体区域的高层特征,再把提取的高层特征按照患者复查的时间顺序输入到长短时记忆网络中以挖掘时间序列数据之间的内在规律;最后使用带有代价敏感的Softmax分类器预测眼病的发展趋势.实验结果表明,该方法具有较高的预测准确率和敏感度,同时可对长度为3～5的序列图像进行预测.

The current researches on the computer aided diagnosis of an ophthalmic disease focus mainly on the automatic classification or grading based on the currently available images, the method for the prediction of an ophthalmic disease is scarce, and therefore, a cost sensitive temporal sequence method is proposed to analyze and predict the development trend of an ophthalmic disease. First, the Canny edge detector operator and Hough transform are used to preprocess the slit lamp image and obtain the lens area. Second, the residual convolutional neural network is employed to extract the high level features from the lens area, which are then inputted into the long short term memory network to mine the inherent laws between the temporal sequence data. Finally, the cost sensitive Softmax classifier is used to predict the development trend of an ophthalmic disease. Experimental results prove that this method has higher accuracy and sensitivity for prediction, and can simultaneously predict different sequence data with a length of 3-5.