Research On Hippocampus Segmentation Method For MRI Based On Multi-Atlas

Among the various organs of the human body,the brain is the most complex system currently studied,and it is one of the most popular research fields.Brain diseases usually have high morbidity,high disability rate,and low cure rate,which have become the focus of global medical care.With the advancement of science and technology,medical imaging technology has become an important auxiliary means for the diagnosis of brain diseases.Brain medical images can visually and clearly reflect the anatomical structure and functional metabolic information of various brain tissues.They need to be analyzed and processed in real time and accurately for use by clinicians.Brain medical image segmentation refers to the recognition and delineation of the tissue in the brain image or the boundary of the region of interest It is the premise for brain image analysis.Brain medical image segmentation plays a vital role in the diagnosis and treatment of diseases.Therefore,this paper mainly studies the segmentation of hippocampus in brain magnetic resonance imaging(MRI),providing a visual basis for medical research on hippocampus-related diseases.Due to the irregular shape of the hippocampus,the blurring of the surrounding tissues such as the amygdala boundary and the low contrast of the brain anatomy,it is difficult to segment the anatomical structure of the hippocampus.The earliest segmentation of specific anatomical structures was achieved by expert manual marking,but there were long-term and error-prone defects,which limited the application of manual marking in big data.The multi-atlas registration-based image segmentation(MAIS)method integrates the prior knowledge of medical maps into the segmentation process of a specific organization,and combined with the efficient label fusion algorithm,can obtain segmentation results that are quite accurate with manual segmentation,and is widely used in the segmentation of the hippocampus.Multi-atlas label fusion is a very important part of MAIS,which plays a crucial role in the segmentation performance of hippocampus.In order to improve the segmentation accuracy of hippocampus,this paper proposes an improved label fusion algorithm based on multi-atlas sparse patch and a label fusion algorithm based on distance metric learning.The traditional label fusion algorithm often ignores the defect of label information when calculating weights.In brain MRI,different label information represents different organizational structures,which plays a very important role in the similarity measure.At the same time,the traditional label fusion algorithm mainly estimates the label value in a local way,without considering global constraint information,such as shape prior information.To overcome these shortcomings,this paper propose a new sparse patch-based label fusion algorithm.In the process of label fusion,the shape of the prior information is introduced to obtain the maximum posterior probability of the label information,increasing the global constraint,and the initial segmentation result of the hippocampus is obtained quickly and accurately.For each label patch,this paper use the information to measure the similarity,and we combines the sparse patch-based label fusion algorithm to further improve the segmentation accuracy of the hippocampus.According to the experimental results,the algorithm can obtain more accurate segmentation results for the segmentation of the hippocampus.This paper proposes a new distance metric learning method in the label fusion stage based on multi-atlas segmentation.The existing label fusion algorithm usually uses a predefined distance metric model to calculate the similarity between the image patches and the label patches.However,the algorithm of this paper learn a distance model from the atlases,so that the structurally similar patches remain similar,and patches of different structures are separated.In the label fusion process,the learned distance metric model is used to perform the similarity measure.The experimental results show that compared with the traditional multi-atlas image segmentation method,this method has achieved statistically significant improvement in segmentation accuracy.