A retina OCT image classification method based on a three-dimensional convolutional neural network

Abstract

The invention discloses a retina OCT image classification method based on a three-dimensional convolutional neural network, and the method comprises the following steps: (a), collecting images, dividing the collected images into a training set and a test set, and carrying out the preprocessing of the images; (b) constructing a Vinception C3D network structure, and constructing a Vinception C3D network structure, wherein the Vinception C3D network is an improvement based on a C3D convolutional neural network. On the basis of a C3D network, a Vinception module fused with multi-channel characteristics is added, the batch standardization method is applied to the original C3D network; and (c) training and testing the model: using the C3D pre-training model as a Vinception C3D pre-training model, training the network loaded with the pre-training model by using data in the training set to obtain the trained Vinception C3D model, and testing the model by using the test set after the model training is finished. According to the method, the three-dimensional retina OCT images can be integrally classified, and a foundation is laid for improving the efficiency of subsequent retina OCT image segmentation and analysis.

Description

technical field [0001] The invention belongs to a retinal image classification method, in particular to a retinal OCT image classification method based on a three-dimensional convolutional neural network. Background technique [0002] The retina is a visual nerve ending tissue that extends outward from the brain. Its structure is complex, delicate, fragile and metabolically vigorous. Its blood vessels belong to the terminal vascular system, and any pathological damage and tissue hypoxia caused by vascular obstruction can lead to tissue necrosis and loss of its ability to sense and transmit light stimulation. [0003] Spectral-domain optical coherence tomography (OCT) detects light that reaches the retina through the refractive interstitium of the eye, and obtains the thickness and distance information provided by the reflection of different tissue interfaces in the eye and restores them to images and data. An indispensable clinical examination technology in ophthalmology, w...

AI Technical Summary

Problems solved by technology

[0004]Currently, most of the existing retinal OCT image classification algorithms are based on two-dimensional images, such as the HOG operator and support vector machine proposed by Srinivasan et al. The OCT image classification method can be used for the classification of age-related macular degeneration and diabetic macular edema; Venhuizen et al. proposed an automatic grading system for age-related macular degeneration based on machine learning; Chakraborty et al. proposed a migration learning-based diabetes Automatic classification of retinal edema and dry age-related macular degeneration, but these algorithms cannot make full use of three-dimensional spatial feature information, and a single image is easily affected by noise, resulting in classification errors

The existing 3D convolutional image classification network applied to medical images needs to be improved in classification accuracy, lacks parallel supplementation between feature maps, and the classification category is relatively single, especially for the simultaneous classification of macular center, optic nerve head center and large field of view. Three scan modes for normal / abnormal classification have not been reported

Images