QSM deep brain nucleus automatic segmentation method based on deep learning

Abstract

The invention discloses a QSM deep brain nucleus automatic segmentation method based on deep learning, and the method is realized based on a quick segmentation tool for a QSM deep brain grey matter nucleus structure, and is named as DeepQSMSeg. Based on an end-to-end model DeepQSMSeg of deep learning, five pairs of ROI (Region of Interest) structures (caudal nuclei CN, shell nuclei PUT, pale globe GP, nigra SN and red nuclei RN in left and right hemispheres) of a deep grey matter nucleus can be accurately, stably and quickly segmented by utilizing a QSM technology. The deep grey matter nuclei are accurately segmented, which is beneficial to the development of brain-iron related researches, especially nervous system degenerative diseases closely related to the deep brain nuclei, such as Parkinson's disease.

Description

technical field [0001] The invention belongs to the technical field of neuroimaging, and in particular relates to an automatic segmentation method of QSM brain deep nuclei based on deep learning. Background technique [0002] Magnetic resonance imaging (MRI) technology provides a non-invasive and highly reproducible method for quantifying tissue magnetic susceptibility, especially important tissue components of brain tissue, including brain iron and myelin. In recent years, with the continuous replacement of MRI technology, quantitative susceptibility mapping (QSM) has become the gold standard for quantitative tissue susceptibility of MRI due to its high magnetic susceptibility contrast and tissue quantitative characteristics. It is worth mentioning that iron is distributed in various regions of the brain, most notably in the subcortical deep gray matter (DGM), and many studies have confirmed that it is closely related to human learning, planning and cognition. On the contr...

AI Technical Summary

Problems solved by technology

But graph-based methods require a series of difficult operations that are computationally expensive and time-consuming

In addition, atlas-based segmentation methods are highly dependent on the stability of atlases and registration algorithms, data changes between atlases or individual images (such as contrast changes or scanner parameter settings) can easily affect performance, and the necessary manual correction is inevitable

In the field of deep learning, some scholars have proposed to use a two-dimensional fully convolutional neural network to segment DGM structures in QSM, but two-dimensional networks cannot capture the spatial information between slices, which is crucial in pixel-level volumetric image segmentation

In addition, it only uses manually selected slices containing DGM structures during training and testing, thus, this segmentation method is not fully automatic

Images