A method and system for fundus image registration based on multispectral

Abstract

The present disclosure provides a registration method based on multispectral fundus images. A registration method based on multispectral fundus images, including: using RHA to obtain multispectral fundus images, selecting a template fundus image and a fundus image to be registered; constructing a regression network to obtain the deformation between the template fundus image and the fundus image to be registered field, and add a feature balance layer and pyramid structure to the regression network; construct a loss function, including the similarity loss of the blood vessel label map and the flattening constraint of the deformation field; train the network in a weakly supervised manner, and use the processed fundus image Segmentation maps serve as labels for training the network. This disclosure uses the segmentation map as a guide to train the registration model in a weakly supervised manner. Practice has proved that this method can effectively solve the problem of multispectral fundus image registration, and can achieve good results, and weakly supervised segmentation-oriented registration The method can be extended to more multimodal image registration problems.

Description

technical field [0001] The present disclosure belongs to the field of image registration, and in particular relates to a multispectral-based fundus image registration method and system. Background technique [0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art. [0003] Medical image registration is a basic task in medical image processing and a hotspot in medical image processing at present, which has very important significance in clinical diagnosis and disease treatment. Medical image registration is to find an ideal spatial transformation between two images, so that the corresponding points of the two images achieve a complete match in terms of spatial and anatomical positions, or at least all clinical diagnosis concerns are matched. [0004] The multi-spectral fundus image is a special fundus imaging, which uses different monochromatic LED light sources to project into...

AI Technical Summary

Problems solved by technology

The second difficulty is that the movement of the eyeball causes the deformation of the fundus

However, in fact, obtaining accurate segmentation information is very expensive, requiring corresponding experts to spend a lot of energy on labeling

The second is a learning-based registration method, which requires pre-aligned image pairs to train a multimodal registration network, but it is difficult to obtain precisely aligned image pairs.

Images