Brain magnetic resonance image segmentation method and device based on unsupervised learning

Abstract

The invention provides a brain magnetic resonance image segmentation method and device based on unsupervised learning, and the method comprises the steps: carrying out the initial segmentation of an original brain magnetic resonance three-dimensional image, and generating a plurality of supervoxels; obtaining a basic feature of each super voxel, and constructing a super voxel feature matrix basedon the basic features corresponding to the plurality of super voxels; inputting the supervoxel feature matrix into a deep subspace clustering network for training to obtain a clustering result of cerebrospinal fluid, gray matter and white matter; and mapping the clustering result to the original brain magnetic resonance three-dimensional image so as to perform image segmentation on cerebrospinal fluid, gray matter and white matter in the original brain magnetic resonance three-dimensional image. According to the embodiment of the invention, three kinds of tissue cerebrospinal fluid, gray matter and white matter in brain magnetic resonance can be accurately and effectively segmented by utilizing the deep subspace clustering network under the condition that a label is not needed.

Description

technical field [0001] The present application relates to the technical field of medical image processing, in particular to a brain magnetic resonance image segmentation method and device based on unsupervised learning. Background technique [0002] In recent years, with the development of medical imaging technology and driven by the era of big data, more and more medical imaging technologies have been collected and stored. The rapid development of biomedical imaging technology enables computers to easily enter the medical field to assist diagnosis and treatment. The brain is an important organ of the human body, and brain tissue plays a vital role in human health. Therefore, accurate and effective techniques are needed to assist in the diagnosis of diseases. Magnetic Resonance Imaging (MRI) has the characteristics of high resolution and low radiation to the human body, and is widely used in the diagnosis and treatment of clinical diseases. However, brain MRI images are ea...

AI Technical Summary

Problems solved by technology

[0006] The disadvantage of this scheme is that the segmentation effect is not very ideal because the gray level difference in different regions of the brain magnetic resonance image is not very large.

[0009] The disadvantage of this scheme is that the image segmentation scheme based on map features actually converts image segmentation into image registration, so that the final segmentation result is very dependent on the similarity between the template image and the image to be registered

[0012] The disadvantage of this scheme is that most of the image segmentation methods based on deep learning are based on strong supervision schemes, relying on large-scale images and high-quality labeled images; the labeled images depend on the annotations of professional doctors, and the results depend on the doctor's Experience level, which undoubtedly increases the difficulty of obtaining high-quality annotated images

Images