Elastic registration method of stereo MRI brain image based on machine learning

Abstract

The invention relates to a method for elastic registration of stereo NMR brain images based on machine learning. The machine learning method is used to obtain the best dimension of the computation attribute vector on each point in the reference image, from which the obtained best attribute vector keeps discrepancy of the greatest extent from the attribute vector on each point around, and conformability of the greatest extent with the attribute vector on the corresponding point of the training sample. Based on the significance and consistency condition of the attribute vector on each point of the image, a standard for evaluating a key point is defined. The key point is selected automatically and hierarchically in each registration stage via the machine learning method, thus preventing the registration process from trapping in a local minimum value point. Finally, the machine learning based frame is combined with the existing registration arithmetic to complete the elastic registration of stereo NMR brain images. The invention can enhance precision and robustness of registration of both real MR images and simulated MR images, thereby establishing a foundation for the feasibility and accuracy of subsequent clinical applications.

Description

technical field [0001] The invention relates to a machine learning-based elastic registration method for stereoscopic nuclear magnetic resonance brain images. With the help of machine learning methods, an optimal attribute vector is learned on each point of the stereoscopic brain image to accurately represent the characteristics of the point. And the key points in the image are selected hierarchically, so as to improve the accuracy and robustness of elastic registration. The present invention can lay the foundation for subsequent clinical applications such as image fusion, precise locating of lesions, formulation of surgical plans, and curative effect tracking, and involves fields such as image elastic registration, machine learning, and stereoscopic nuclear magnetic resonance brain (MR) images. Background technique [0002] Medical image registration has very important clinical application value. The registration of medical images obtained by using various or the same imag...

AI Technical Summary

Problems solved by technology

However, the disadvantage of HAMMER is that the size of the neighborhood for calculating GMI is determined in advance, and for each point in the image, the GMI is calculated in the neighborhood of the same size.

In areas with rich boundary information, such as near the corners of the cortical and ventricles, the GMI calculated in a smaller neighborhood can best reflect the anatomical structure information of these points; on the contrary, in the white matter (White Matter ) area, the grayscale of the image changes slowly. At this time, if the GMI is still calculated in a small neighborhood, the attribute vectors of these points will be very close to make it difficult to distinguish

Images