Researches On Retinal Optical Coherence Tomography Images Classification By Deep Convolutional Neural Network

Optical coherence tomography(OCT)is currently the most commonly used inspection method in ophthalmology clinical examinations.It can quickly obtain crosssectional images of its retina without damaging the patient’s body.This is of great importance in the research and analysis of retinal related diseases.Very important role.The classification of retinal OCT images is a necessary process for doctors to diagnose and study retinal-related diseases,but currently it is mainly achieved by manual classification by doctors,which is prone to subjective misjudgment by doctors and is very time-consuming.Therefore,the research on automatic classification of retinal OCT images is of great significance for achieving efficient diagnosis and treatment and reducing labor costs.In recent years,the application effect of the deep convolutional neural network in the field of image processing is very significant,and some researchers have also begun to apply the deep convolutional neural network model to the classification of retinal OCT images.Traditional classification methods based on deep convolutional neural networks only use the last layer of convolutional feature information,and do not use the previous layer of convolutional feature information in the final classification task.In addition,each frame of B-scan images in 3D tomographic retina OCT images has a contextual connection,but the existing deep convolution-based neural network methods are mostly based on the classification study of a single B-scan image,which is easy to ignore The relationship between B-scan images of adjacent frames is lost.In view of the above problems,this paper proposes two classification algorithms.The main contents of this article are summarized as follows:1)Aiming at the problem that the traditional convolutional neural network does not use the feature information of different convolutional layers for classification,this paper proposes an iterative fusion convolutional neural network method for automatic classification of retinal OCT images.This method uses iterative fusion decision to effectively combine the feature information extracted by the current convolutional layer with the feature information extracted by other convolutional layers,and finally obtain more accurate retina OCT image classification results.The experimental results show that the feature information of different convolutional layers can be fully utilized by using iterative fusion decision-making,thereby effectively improving the classification effect.2)Regarding the classification of retinal OCT images,the relationship between adjacent B-scan images in 3D tomographic retinal OCT images is not taken into consideration.In this paper,a classification method of retinal OCT images based on long-short-term memory convolution network.This method first extracts visually representative feature information from the B-scan of each frame in the 3D retina OCT image through the deep residual network,and then uses the long-term and short-term memory network’s long-term dependence between the obtained convolution feature information By modeling,the feature information between adjacent B-scan images has a long-term dependence,and finally a more accurate retina OCT image classification result is obtained.The experimental results show that the long-term and short-term memory convolutional network model can solve the long-term dependence problem between adjacent B-scan images in the 3D tomographic retina OCT image,thereby improving the classification accuracy.