Hippocampus three-dimensional semantic network segmentation method based on multi-scale feature multi-path attention fusion mechanism

Abstract

The invention discloses a hippocampus three-dimensional semantic network segmentation method based on a multi-scale feature multi-path attention fusion mechanism, and belongs to the field of medical image processing. The method comprises the following steps: preprocessing public marked hippocampus image data, and cutting original data into image blocks containing a hippocampus; a brand new hippocampus three-dimensional semantic segmentation network structure is constructed based on multi-scale feature extraction, a multi-path attention fusion mechanism and a branch classifier ensemble learning strategy; dividing the data set; performing offline model training to obtain model weight parameters for the three-dimensional hippocampus structure; and segmenting the test set image by using the model file and evaluating a segmentation result. According to the method, the semantic segmentation network structure conforming to the characteristics of the three-dimensional hippocampus image is designed, so that the utilization rate of the network on multi-dimensional image information can be improved, the pixel intensive prediction capability is improved, and the hippocampus segmentation performance is improved.

Description

technical field [0001] The invention relates to the field of medical image processing, in particular to a hippocampus three-dimensional semantic network segmentation method based on a multi-scale feature multi-channel attention fusion mechanism. Background technique [0002] The hippocampus is an important structure in the brain. It is located between the thalamus and the medial temporal lobe. It is a part of the limbic system and is mainly responsible for the storage conversion and orientation of long-term memory. Its nerve cells are very fragile and extremely vulnerable. Once the nerve cells in the hippocampus die, people's memory will be lost. Its morphological changes are important biomarkers for studying long-term memory. For example, judging whether the volume of the hippocampus shrinks based on Magnetic Resonance Images (MRI) is one of the key techniques for diagnosing Alzheimer's disease. However, the hippocampus is not well distinguished from its surrounding brain ...

AI Technical Summary

Problems solved by technology

However, in this type of method, only the channel dimension features are weighted by the attention coefficient, and more feature information exists in the spatial dimension voxels. In the 3D segmentation network, the use of spatial voxel features can bring better segmentation performance. Improvement, there is still room for improvement in the segmentation network applying the attention mechanism

Images