Researches On Classification Method Of Retinal Disease In Optical Coherence Tomography Images

Optical coherence tomography(OCT)is a novel imaging technique that can produces non-contact,non-invasive and high-resolution cross-sectional images of microstructures in biological tissues.Nowadays,OCT technique has been used as a routine examination tool in ophthalmology clinical diagnosis.For many blind eye disease,early dectection and timely diagnosis is crucial to prevent vision loss.In order to make the accurate diagnosis of patients,opthalmologies are required to qualitatively and quantitatively analyze the retinal lesions in each image.However,manual diagnosis is a tedious and time-consuming work,and subjective misjudgment is also easily caused in this process.Therefore,the development of computer-aided diagnostic(CAD)tools for reducing the burden on doctors,accelerating the flow of consultation,and improving the diagnostic accuracy are of great value and significance.In this context,this article focuses on the research of retinal lesions detection and diseases classification in OCT images.Clinical retinal disease is very complex,and usually has multiple lesions.However,the current OCT image based retinal diseases automatic diagnosis techniques are still in the early stages.Most methods are only designed for one single lesion,which is hard to diagnose the complex diseases.To address this issue,we propose a multiple retinal lesions classification method in optical coherence tomography images.With the combination of medical knowledge and machine learning algorithm,the problem of complex multiple lesions classification can be solved by degenerating it into multiple single lesion classification problems.Convolutional Neural Network(CNN)is the most widely used and most accurate algorithm in the field of image classification and recognition.However,given the limited labeled data in medical imaging,it is difficult for the CNN model to achieve satisfactory results in OCT image classification tasks.In order to solve this problem,this paper proposes an automatic retinal diseases classifica tion method based on convolutional neural network and transfer learning,By fine-tuning the pre-trained CNN model(VGG/ResNet)on the ImageNet database,the algorithm can achieve excellent performance in the task of retinal disease classification with limi ted OCT image training data.