Nuclear magnetic resonance hippocampus segmentation algorithm based on dual dense context awareness network

Abstract

The invention discloses a nuclear magnetic resonance hippocampus segmentation algorithm based on a dual dense context awareness network. The nuclear magnetic resonance hippocampus segmentation algorithm comprises the following steps: segmenting a nuclear magnetic resonance hippocampus image and a corresponding standard segmentation image into two-dimensional image blocks with different sizes by utilizing different scales; judging whether a two-dimensional image block in the nuclear magnetic resonance hippocampus image is an effective block or not; carrying out data enhancement processing on all the effective blocks, then carrying out normalization processing, and forming an effective block set by all the normalized effective blocks; dividing the effective block set into a training set anda test set; inputting the effective blocks in the training set into a dual dense context awareness network for training to obtain a dual dense context awareness network training model, and obtaining an optimal weight vector and an optimal bias term by utilizing a loss function value; inputting the effective blocks in the test set into a dual dense context-aware network training model, and predicting by using the optimal weight vector and the optimal bias term to obtain a segmentation result graph corresponding to the effective blocks; the method has the advantage of high segmentation precision.

Description

technical field [0001] The invention relates to an image segmentation technology, in particular to an NMR hippocampus segmentation algorithm based on a dual dense context perception network. Background technique [0002] The hippocampus is a very important organization in the human brain, which is closely related to human cognitive functions (such as learning and memory). The shape analysis of the hippocampus is very important for the diagnosis and prediction of various neurological diseases. For example, the atrophy of the hippocampus is one of the characteristics of detecting schizophrenia, and the atrophy of the hippocampus can cause Alzheimer's disease. When judging whether the hippocampus is atrophied, doctors usually need to segment the hippocampal structure in MRI (Magnetic Resonance Imaging, MRI), and perform shape and volume analysis. However, since the hippocampus and its adjacent brain tissues such as the amygdala, fornix, and cingulate gyrus belong to gray matte...

AI Technical Summary

Problems solved by technology

The U-net network proposed on the basis of FCN (Fully Convolutional Networks, fully convolutional network) is a typical network that uses the long connection structure between the upper and lower sampling layers to achieve multi-resolution feature fusion, because its special U-shaped frame can Considering the context information of the entire image well, it has been widely used in the field of medical image segmentation. However, due to the small size of the hippocampus, it is difficult for this single-scale multi-resolution feature fusion network to accurately segment the hippocampus and background with large volume differences. part, resulting in lower segmentation accuracy

Images