three-dimensional medical image registration method based on a U-NET neural network

Abstract

A three-dimensional medical image registration method based on a U-NET neural network is disclosed. The method comprises the following steps of: firstly, using a training image to perform U-shaped registration on a U-shaped image; carrying out training optimization on the NET architecture neural network; and applying the trained model to the registration of the new image, so that the technical limitation in the traditional registration method can be overcome, a more accurate registration effect can be obtained by a neural network-based cyclic optimization method, the speed is high, the speed is improved by more than ten times compared with that of the traditional registration method, and the image registration efficiency is greatly improved.

Description

technical field [0001] The invention belongs to the field of image registration, and mainly relates to a three-dimensional medical image registration method based on a U-NET neural network. Background technique [0002] In medical applications, two sets of (three-dimensional) images of the same part of the same patient recorded at different times and by different imaging devices (CT, MRI, etc.) need to be matched and superimposed. This process is called image registration. Traditional registration methods are mainly divided into three categories: [0003] 1. Feature point registration. The feature point registration method extracts the feature points of the image, matches the feature point pairs through the similarity measure, and obtains the spatial coordinate transformation parameters of the image according to the matched feature point pairs. [0004] 2. Rigid registration. The rigid registration method regards the object to be registered as a rigid body, and only uses ...

AI Technical Summary

Problems solved by technology

[0006] Among them, feature point registration is limited by feature point extraction methods and technologies, requiring more experience and long-term debugging; rigid registration is limited by deformation means, and the effect is not ideal for soft and easily deformable objects (such as human internal organs); Elastic registration overcomes the problem of rigid registration, but it needs to optimize each point of the deformation field, but its optimization has a large degree of freedom, a large amount of calculation, and takes a long time.

Images