Automatic segmentation method of three-dimensional magnetic resonance image of brain structure

Abstract

An automatic segmentation method of a three-dimensional magnetic resonance image of a brain structure utilizes a manual image segmentation method to obtain a plurality of maps. The plurality of maps are gradually registered to images to be segmented; initial segmentation is performed on the images to be segmented, and image voxel required to be segmented further is determined; voxel of all maps in a specific space field corresponding to the image voxel required to be segmented further serves as a candidate training sample set; image gradation and textural features are extracted from the image voxel required to be segmented further and the image voxel in the candidate training sample set; and K pixels closest to the image voxel required to be segmented further are searched in the candidate training sample set so serve as a training set, and a supporting vector machine is utilized to train a sorter to segment the training set. The automatic segmentation method of the three-dimensional magnetic resonance image of the brain structure extracts various feathers comprising grey level and texture of each sample and does not require that the image voxel required to be segmented corresponds to the voxel in the maps one by one. The number of training samples can be more than that of the maps, and the training sorter is robust.

Description

technical field [0001] The invention relates to the technical field of medical image processing, in particular to a method for automatically segmenting substructures of three-dimensional magnetic resonance brain images. Background technique [0002] In the basic and clinical research of medical imaging, the segmentation of MRI brain images is of great significance. Reliable and accurate brain segmentation is widely used in many medical imaging applications, such as surgical planning, study of disease course, brain development of old or young people, etc. Brain segmentation involves segmenting brain tissue and brain structures. In traditional research, the segmentation results obtained by manual calibration of brain MRI images by trained experts are the gold standard for image segmentation. However, this method is very time-consuming, especially as the data set continues to grow, this method becomes less and less feasible. In addition, manual calibration is also prone to i...

AI Technical Summary

Problems solved by technology

However, most of the current label fusion methods rely on the preset model, which cannot guarantee that this model is optimal. In addition, during the fusion process, it is strictly required that the voxels of the image to be segmented correspond to the voxels in the atlas, so that the voxels used for fusion The number of samples is limited to the number of maps. Finally, the current label fusion method only uses the gray level information of voxels, and many useful texture features can provide important information for fusion.

Images