A method and system for real-time image registration

Abstract

The invention discloses an image real-time registration method and system. The method comprises the following steps of (1) acquiring data of a to-be-registered 2D image; and (2) inputting the data to a registration neural network obtained according to training of a target 3D image, and obtaining position information of the 2D image in the target 3D image, wherein the position information comprises an eigenvector of the to-be-registered 2D image in a coordinate system of the target 3D image and a distance between the to-be-registered 2D image and an origin of the coordinate system, and the eigenvector is a vector passing through the origin of the coordinate system and perpendicular to a 2D image plane, and comprises two parameters including an elevation angle and an azimuth angle. The system comprises an image acquisition module and a registration module. According to the method and the system, deep learning is introduced in the 2D-3D registration problem, a corresponding relationship between 2D and 3D is expressed as a deep convolutional neural network, and a policy for solving the exhaustion calculation problem is explored theoretically, so that the purpose of real-time registration is achieved.

Description

technical field [0001] The invention belongs to the field of image registration, and more particularly relates to a method and system for real-time registration of medical images. Background technique [0002] Image registration is widely used in clinical medicine, aerial remote sensing, public security criminal investigation and other fields, and a large number of in-depth researches have been carried out around this issue. There are a large number of 2D-3D image registration problems in medical images. The current 2D-3D registration methods can basically be described as: obtain slices of 3D data from different angles and positions, and compare the slices with the actual 2D images. , according to the principle of optimal similarity, the position of the 2D image in the 3D data is obtained. [0003] According to the above analysis, the existing 2D-3D image registration, strictly speaking, is coarse registration, and the main strategy used is a method close to exhaustion or a...

AI Technical Summary

Problems solved by technology

[0004] In view of the above defects or improvement needs of the prior art, the present invention provides a 2D-3D real-time registration method based on convolutional neural network, which introduces volume and neural network into the registration, and its purpose is to simulate the neural network of actual organisms. Network, through its excellent feature learning ability to achieve real-time 2D-3D registration, thereby solving the technical problems of low accuracy and low robustness of existing registration methods

Images