Image non-rigid registration method and system based on generative adversarial network

Abstract

The invention discloses an image non-rigid registration method and system based on a generative adversarial network. The method comprises the steps: acquiring a synthesized medical image data set based on a real medical image data set, the synthesized data set serves as a training sample of a generative adversarial network model, and the real data set serves as a verification and test data set foroptimizing the generative adversarial network model; constructing a structure of the generative adversarial network model, and performing iterative training on the generative adversarial network model based on the synthesized medical image data set to obtain an optimized generative adversarial network model; and performing image non-rigid registration on the to-be-registered image pair in the real medical image data set based on the optimized generative adversarial network model to obtain a corrected image close to the reference image. According to the method provided by the invention, the precision and speed of non-rigid registration of the medical image can be improved under the condition that the number of training samples of the medical image is small and labeling information is lacked, so that the generalization ability of the registration method and system is enhanced.

Description

technical field [0001] The present invention relates to the field of image processing, in particular to an image non-rigid registration method and system based on a generative confrontation network. Background technique [0002] The development of medical image registration technology can be divided into three stages: in the 1980s, due to the backwardness of imaging technology at that time, most of the research was limited to the rigid registration between images of the same modality; in the 1990s, with various modality With the emergence of multi-modal images, registration is no longer limited to single-modal images, and multi-modal image registration and fusion technologies have emerged. However, the spatial transformation relationship between images at this time is still rigid; at the beginning of the 21st century, Clinical applications in medicine have put forward higher requirements for image registration. Although rigid registration technology has reached maturity, it ...

AI Technical Summary

Problems solved by technology

Although this method achieves better registration results, the final registration results still depend on other feature-based registration methods; at the same time, block-based unsupervised registration methods require post-processing, and post-processing cannot It is performed in a convolutional neural network, so the operation steps are cumbersome, time-consuming and labor-intensive, and the registration efficiency is low

②The unsupervised registration method based on the learning method directly learns the deformation of multi-modal registration image pairs and optimizes the similarity measurement function. It is difficult to directly obtain accurate pixel relative displacement vector fields. At the same time, compared with image classification and segmentation tasks, Labels for image registration are difficult to obtain directly

[0004] Although the above method achieves good results in the single-modal image registration task, it is still difficult to directly use it to achieve multi-modal image registration.

On the one hand, the reason is the structural complexity of actual medical images, such as: local deletion, large deformation and irregular deformation caused by the physiological movement of each organ; on the other hand, it is difficult to define a robust image similarity measure , it is difficult to evaluate the registration accuracy

Images