Semi-supervised retinal OCT image layer segmentation method combined with generative adversarial network

Abstract

The invention belongs to the computer vision and image processing technology, and relates to a semi-supervised retina OCT image layer segmentation method combined with a generative adversarial network, which comprises the following steps: preparing retina OCT image data, taking labeled pictures of part of patients and unlabeled pictures of all patients as a training set, and taking labeled pictures of other patients as a test set; constructing a generative adversarial network, the generative adversarial network comprising a segmentation network and a discriminator network, and the output end of the segmentation network being connected with the input end of the discriminator network; designing a loss function of the generative adversarial network; setting evaluation indexes; and introducingthe prepared training set by using the designed loss function, and training the generative adversarial network. According to the method, the generative adversarial network is trained by utilizing thelabeled data and the unlabeled data at the same time, so that the robustness of the network is enhanced, and the semantic segmentation accuracy is improved.

Description

technical field [0001] The invention relates to artificial intelligence, computer vision, and image processing technology, in particular to a semi-supervised retinal OCT (Optical Coherence tomography) image layer segmentation method combined with a generated confrontation network. Background technique [0002] In recent years, methods based on convolutional neural networks (CNN), such as Unet, have achieved remarkable results on the task of semantic segmentation of medical images. Although CNN-based methods achieve amazing results, they require a large amount of training data. Different from image classification and object detection, semantic segmentation requires precise pixel-by-pixel annotation of each training image, and medical images require professional doctors to annotate, which will cost a lot of cost and time, thus causing the retinal OCT image layer segmentation direction The training data is less. [0003] The high resolution of retinal OCT images facilitates c...

AI Technical Summary

Problems solved by technology

[0005] 1. Traditional segmentation methods based on feature engineering often require manual design of feature extractors, and the accuracy of segmentation is not high

[0006] 2. Although the method based on convolutional neural network (CNN) has achieved good segmentation results, due to the lack of training data, the generalization ability of the model is low

Images