Fundus blood vessel image segmentation method based on self-supervised learning

Abstract

The invention provides a fundus blood vessel image segmentation method based on self-supervised learning. The method comprises the following steps of S1, setting a data set which comprises a labeled image sample D and label-free left and right eyeball image information XT belonging to (L, R) of a same patient; S2, constructing a self-supervised network which is composed of two identical U-Net networks based on an auxiliary task, wherein the two parallel U-Net networks are respectively marked as a baseline (a) and a baseline (b); S3, constructing an image segmentation loss model, respectively inputting the labeled image sample D and the label-free left and right eyeball image information XT belonging to (L, R) of the same patient into the baseline (a) and the self-supervision module composed of the baseline (a) and the baseline (b), optimizing the parameters of a baseline (a) network through a cross entropy loss function and a local consistency comparison loss function, wherein the network parameters of the baseline (b) in the training process are obtained through the momentum transmission of the baseline (a); and S4, inputting the to-be-detected left and right eye image information to any network in the optimized self-supervised network for image segmentation processing to obtain a final segmentation result. Through the above method, the accurate segmentation can be performed on the fundus blood vessel image.

Description

technical field [0001] The invention relates to an image segmentation method, in particular to a fundus blood vessel image segmentation method based on self-supervised learning. Background technique [0002] The image segmentation results of fundus blood vessels (that is, retinal blood vessels) can be used as the basis for the diagnosis of patients with fundus diseases, and can also provide pathological references for complications of other diseases, such as diabetes and hypertension. [0003] Among the existing fundus image segmentation algorithms, most retinal blood vessel image segmentation methods based on deep learning mostly rely on annotated images manually segmented by clinical experts. However, due to the complexity of the vascular structure and the low contrast of eyeball images, this makes labeling samples a tedious and expensive task, and at the same time, due to the subjective factors of experts in the labeling process, it will eventually lead to inaccurate labe...

AI Technical Summary

Problems solved by technology

However, due to the complexity of the vascular structure and the low contrast of eyeball images, this makes labeling samples a tedious and expensive task, and at the same time, due to the subjective factors of experts in the labeling process, it will eventually lead to inaccurate labeling samples.

Images