Neural network model for ocular fundus image blood vessel segmentation

Abstract

The invention relates to a neural network model for blood vessel segmentation of fundus oculi images. The blood vessel feature processing layer of the highest layer is connected with the blood vesselfeature optimization layer of the lowest layer through a backward short connection. Each vessel feature optimization layer is connected with a vessel feature optimization layer of a higher layer through a forward short connection. Each vessel feature optimization layer for acquiring an upsampled image of the connected vessel feature processing layer; The lowest vessel feature optimization layer isalso used for acquiring the feature image of the lowest vessel feature extraction layer; Each vessel feature optimization layer is also used for sequentially performing vessel feature extraction andnonlinearization processing on each acquired image to obtain the corresponding nonlinearized image of each image, and sending each acquired image to the vessel feature optimization layer of a higher layer through a forward short connection. The invention transmits the high-level information to the low-level through the backward short connection, and transmits the low-level information to the high-level through the forward short connection, fully fuses the characteristics of each level, and makes the blood vessel segmentation more accurate.

Description

technical field [0001] The embodiment of the present invention relates to the field of computer technology, in particular to a neural network model for blood vessel segmentation in fundus images. Background technique [0002] Retinal fundus image analysis helps ophthalmologists to deal with the diagnosis, screening and treatment of cardiovascular and ophthalmic diseases, such as macular degeneration, diabetic retinopathy, glaucoma, hypertension, etc. These diseases can lead to blindness if left untreated. Vessel segmentation is a fundamental step in retinal image analysis and helps localize diabetic retinopathy and foveal regions. However, in clinical practice, manual labeling of blood vessels in retinal images is time-consuming and requires a lot of experience. Therefore, automatic retinal vessel segmentation is necessary to reduce labeling time. [0003] The automatic segmentation schemes of retinal image vessels in recent decades can be divided into two categories: uns...

AI Technical Summary

Problems solved by technology

In the supervised scheme, the neural network model needs to extract image features layer by layer, and a lot of useful information is lost, resulting in the parameters learned by the neural network model not being able to fully describe the characteristics of blood vessels.

Images