Simplified weighted-undirected graph-based statistical averaging model construction method

Abstract

The invention discloses a simplified weighted-undirected graph-based statistical averaging model construction method. With the method adopted, problems such as the problem of deviation caused by the selection of a specific template image and the problems of low convergence speed and low efficiency of a template-free GroupWise registration can be solved. The method of the invention includes the following steps that: with a rigidly-aligned image set adopted as input, a vertex set of a graph model is constructed, a threshold value is solved, edges between images are constructed, so that an undirected graph model can be built; based on image affinity propagation clustering results, intra-cluster redundant edges and inter-cluster redundant edges are deleted, so that the undirected graph model can be simplified; based on the simplified graph model, the weights of edges are set according to the difference of each image and adjacent images connected with the image; deformation fields between each image and the adjacent images connected with the image are calculated through registration; the deformation fields are compounded according to the weight of the corresponding graph model; the compounded deformation fields act on the images, so that the differences between the images are gradually decreased, and the simplified weighted undirected graph model gradually shrinks; and the average image of the image set is solved, so that an atlas can be obtained.

Description

technical field [0001] The invention relates to the technical field of image registration, in particular to a method for constructing a statistical average model based on a simplified weighted undirected graph. Background technique [0002] Medical image registration refers to the transformation of images acquired under different conditions into the same coordinate space to achieve the fusion of multiple different information. By registering the data of different modalities, different times, or different patients, different data can be spatially consistent at corresponding points, providing more comprehensive information for doctors to diagnose diseases. It has broad application prospects in the field of computer-assisted therapy. [0003] With the development of medical image registration technology, a large number of accurate registration algorithms have been proposed. In 2002, Shen proposed the HAMMER elastic registration algorithm, which uses the boundary type, image g...

AI Technical Summary

Problems solved by technology

It is obvious that the ABSORB algorithm converges faster, but the results are biased

The HUGS algorithm avoids the deviation, but the undirected graph it establishes has a slow convergence speed, which affects the efficiency of the algorithm.

Images