Method for detecting branch points and cross points of retinal blood vessels in OCTA image

Abstract

The invention provides a method for detecting a retinal vessel bifurcation point and a cross point in an OCTA image. The method comprises the following steps: S1, extracting and fusing features of an input image by using a stacked hourglass network; S2, inputting the features extracted in the step S1 into three prediction branches to predict a central point heat map, local offset of a central point and a target size; and S3, obtaining a classification result and position information of the blood vessel feature points from the three prediction branches. According to the method, the blood vessel bifurcation points and the cross points in the OCTA image can be directly detected and classified at the same time; and a stacked hourglass network in the Center Net is improved by using an attention gate, and the attention gate can highlight valuable features in the OCTA image and inhibit irrelevant feature activation, so that the detection performance is improved.

Description

technical field [0001] The invention relates to the field of biomedical engineering, in particular to a method for detecting bifurcation points and intersection points of retinal blood vessels in OCTA images. Background technique [0002] Retinal vasculature is an important component of clinical pathology and is associated with many ocular diseases. Quantitative analysis of retinal vasculature can facilitate the diagnosis and treatment of diseases involving vascular abnormalities. Optical coherence tomography angiography (OCTA) is a new ophthalmic examination technique developed in recent years. Compared with traditional fluorescein angiography, OCTA is a non-invasive and safe imaging technique that requires no intravenous dye and is faster and easier to perform. OCTA can generate data for multiple retinal and choroidal layers and can reveal structural and blood flow information. Currently, OCTA has become a valuable imaging tool for observing retinal blood vessels. [0...

AI Technical Summary

Problems solved by technology

[0004] Accurate segmentation of OCTA images is a great challenge due to the complex vascular network and blurred contrast between vessels and background

Using traditional image processing methods such as filtering, thresholding, and edge detection to segment vessels in OCTA images can lead to glitches, vessel discontinuities, and loss of vessel detail

Although some deep learning-based methods are effective in OCTA image segmentation, manual annotation of blood vessel segmentation maps for training is quite time-consuming and labor-intensive.

Images