Multi-map label fusion method based on map combination information sparse representation

Abstract

The invention relates to a multi-map label fusion method based on map combination information sparse representation. The multi-map label fusion method comprises the steps that 1, registering N maps tothe space where a to-be-segmented brain MR image T is located; step 2, extracting combination information of map voxels; 3, extracting combination information of voxels to be marked; 4, forming a data dictionary by taking the voxel combination information extracted from each atlas as a unit, performing sparse representation on the combination information of the voxels to be marked by utilizing the data dictionary, and obtaining corresponding sparse coefficients; and 5, performing weighted fusion on the map label by using the obtained sparse coefficient to obtain label information of the voxelto be marked, thereby obtaining a final segmentation result of the brain MR image. According to the method, the map image information and the map label information are effectively combined, so that the utilization rate of the prior information provided by the map is improved, and particularly the boundary information of the target brain tissue provided by the map is improved.

Description

technical field [0001] The invention relates to the field of brain MR image segmentation of medical images, in particular to a multi-atlas label fusion method based on sparse representation of atlas combination information. Background technique [0002] Brain-related diseases are affecting the life and health of our country and the global population. By analyzing the characteristics and rules of changes in brain anatomical structures, the relationship between brain diseases and specific brain anatomical structures can be discovered (for example, the onset of Alzheimer's disease is related to the lesion of the hippocampus in the brain anatomical structure, Pa Kinson syndrome has a certain relationship with the lesions of the anatomical structure of the brain, globus pallidum, etc.). Quantitative analysis of the anatomical structure of the brain can not only provide a reliable basis for the diagnosis of brain-related diseases, but also serve as the basis for follow-up surgery...

AI Technical Summary

Problems solved by technology

At present, there are many disadvantages in the manual segmentation of brain MR images: manual segmentation of a brain MR image takes a lot of time, physical and mental effort; manual segmentation of brain MR images has a strong subjective Different people or the same person will get different segmentation results when they segment the same brain MR image at different times, making the results obtained by manual segmentation not reproducible; a large number of brain MR images are generated every day in clinical applications. MR images, which makes manual segmentation impractical

The method based on multi-atlas label fusion has been proved to be an effective method for automatic segmentation of brain MR images, but the existing multi-atlas label fusion method only performs the brain MR image information corresponding to the atlas in the process of calculating the atlas label weight. However, the importance of brain tissue boundary, shape and other information contained in the corresponding labels of the map is completely ignored in the process of calculating the weight of the map label, which leads to a decrease in the utilization rate of the map information, and also reduces the reliability of the map label weight.

Images