A choroidal segmentation method for OCT images based on improved U-net network

Abstract

The invention discloses an OCT image choroid segmentation method based on an improved U-net network. The main improvement points of the U-net include: (1) extracting more by increasing the number of encoders and decoders in the network. Feature information; (2) Adding an exquisite residual block after the encoder to enhance each layer's recognition capability; (3) Adding a attention module behindthe decoder to let the high-level semantic information guide the underlying details; (4) Using the loss function The traditional L2 loss and Dice loss are combined to constrain the network model. Theimproved U-net network of the present invention can automatically segment the upper and lower boundaries of the choroid of the human eye or the pathological myopia, and the segmentation result is highly accurate.

Description

technical field [0001] The invention relates to a method for segmenting the choroid of an OCT image based on an improved U-net network, and belongs to the technical field of fundus image segmentation. Background technique [0002] Swept optical coherence tomography (SS-OCT) with a center wavelength of 1050 nanometers is the latest three-dimensional fundus scanning technology with the advantages of short imaging time, real-time performance, biopsy performance, and high resolution. The acquired wide-field image scan range includes the center of the macula and the optic nerve head (ONH) area. The image can show the complete choroidal tissue and part of the scleral structure. [0003] The choroid is a layer of vascular network tissue between the retina and the sclera. It is composed of rich blood vessels and pigments. Its main function is to provide oxygen and nutrients for the entire eyeball, and it has the effect of light isolation, making the reflected image clearer. At the...

AI Technical Summary

Problems solved by technology

[0004] At present, most algorithms for automatic segmentation of the choroid in OCT images are traditional algorithms, which have the following defects and deficiencies: (1) Most of the algorithms are two-dimensional algorithms, which only perform independent segmentation on each slice image, which is easy to achieve. Affected by image noise and artifacts, and does not take into account the spatial correlation of 3D images, resulting in relatively large errors in segmentation results

(2) Many algorithms can only adapt to the choroid segmentation of normal OCT images. When choroidal lesions occur, these algorithms will fail, and many algorithms are not universal and robust

(3) Due to the particularity of the neural head (ONH) area, some algorithms are not suitable for large-field scanning imaging, and some algorithms require preprocessing, resulting in complex algorithms but low actual accuracy

Images