Registration method and device for CT image and MRI three-dimensional image

Abstract

The invention provides a CT image and MRI three-dimensional image registration method and device, and the method comprises the steps: obtaining an originally collected CT image and an originally collected MRI three-dimensional image, and carrying out the preprocessing operation, and obtaining a CT processing image and an MRI processing image; performing modal conversion on the CT processing image and the MRI processing image by using a modal conversion generator obtained by pre-training to obtain a CT modal conversion image and an MRI modal conversion image; predicting the CT processing image and the MRI processing image by using a single-mode registration network obtained by pre-training to obtain a single-mode registration confidence coefficient; and performing registration on the CT processing image, the MRI processing image, the CT mode conversion image, the MRI mode conversion image and the single-mode registration confidence coefficient by using a multi-mode registration network obtained by pre-training so as to realize automatic registration of the CT image and the MRI three-dimensional image.

Description

technical field [0001] The present disclosure relates to the field of digital medical technology, and in particular to a registration method and device for CT images and MRI three-dimensional images. Background technique [0002] Medical image registration is an important step in medical image processing. Traditional medical image registration methods include iterative closest point method, Gaussian mixture model method, congruent four-point set method, etc. These methods generally have the problem that the iterative optimization process is computationally complex, and it is easy to fall into local optimum and cause registration failure. In recent years, with the widespread application of deep learning in the field of image processing, such traditional registration methods have gradually been replaced by deep learning methods. [0003] Deep learning methods use convolutional neural networks to learn input image features and are widely used in image processing. Among them, ...

AI Technical Summary

Problems solved by technology

However, current deep learning methods generally require a large amount of existing medical image data for network training. Considering that medical image data is difficult to obtain, and non-rigid registration deformation results between images are difficult to manually label, supervised deep learning registration methods are usually difficult. practical application

However, the unsupervised learning registration method needs to calculate the similarity loss function according to the appearance difference between the images to be registered. When the images to be registered have obvious differences in appearance in different modalities, it is difficult for existing methods to measure their similarity.

[0004] Therefore, such methods are difficult to apply to computer tomography (CT) and magnetic resonance (MRI) cross-modal image registration problems

Images