Multi-modal brain image registration method based on GAN

Abstract

With the development of the imaging technology, the appearance of various imaging devices makes a great contribution to the progress of modern medicine, but due to the limitation of the imaging principle, the single-mode imaging technology generally can only provide single and limited information, and therefore, in order to improve the accuracy of diagnosis and the effectiveness of treatment, doctors often need to fuse information of images of different modalities to know comprehensive information of diseased tissues or organs. The invention provides a registration method based on a generative adversarial network in order to realize efficient and rapid registration of brain images and facilitate treatment of doctors. A U-Net encoder-decoder structure is adopted in a generator, an encoder obtains transformation parameters from a floating image to a fixed image, a decoder recovers the size of a feature map, medical image registration based on normalized mutual information is used in similarity measurement, a gradient descent method is used as an optimization algorithm of image registration, and tests are respectively performed in CT single-mode and CT-MRI multi-mode sequence images, and registration results obtained by using an original normalized mutual information calculation method and an improved normalized mutual information method are compared. According to the disclosed method, the generalization ability of the model can be enhanced by automatically learning the mapping relation between the same data set. According to the invention, accurate and rapid registration can be realized.

Description

technical field [0001] The invention belongs to the technical field of medical image processing, and in particular relates to a method for brain image registration. Background technique [0002] As a visual representation of the internal structure and internal functions of anatomical regions, medical images convey vital information to people and are an important basis for doctors to observe, diagnose and treat diseases. However, different imaging devices have different advantages and disadvantages. For example, ultrasound imaging is low-cost and fast-acting. It is currently the main method for some disease screening and fetal examination for pregnant women. However, the ultrasound image is noisy, and the imaging site and depth have great limitations. ;MRI is very sensitive to the change of water content in human tissue, and has a good imaging effect on human soft tissue, but it is easily affected by metal, and the scanning time is long, and there are many artifacts; CT is go...

AI Technical Summary

Problems solved by technology

Feature-based non-rigid medical image registration has a small amount of calculation and a fast calculation speed, but usually the feature point selection process is cumbersome and requires manual operations that lack automation, and the difference in operator level directly affects the accuracy of feature point selection

Images