Research Of Brain Structure Segmentation Algorithms Of Infant Based On Low-Rank Decomposition Improved Dictionary Learning

Infantile period is a critical period of brain development.During this period,the brain develops rapidly,the plasticity is strong,and the risk of brain disease is much higher than that of adults.Therefore,the early diagnosis methods for exploring brain diseases in infants and young children are significance.Modern medical imaging technology has become an important means of clinical diagnosis of brain diseases.The use of nuclear magnetic resonance MRI images to segment the brain structure of infants is the basis for quantitative analysis of brain structure in infants and brains to study brain development and disease diagnosis.The deep brain structure includes the hippocampus,the corpus callosum,and the amygdala.The deep structure of the brain of infants and young children is still in the developmental stage,the gray-scale and shape information is more complicated and irregular than that of adults,so it is more difficult to segment the adult brain structure.In this paper,the hippocampus and corpus callosum are taken as examples.Under the framework of multi-atlas,the low-rank decomposition method is used to study the three-dimensional automatic segmentation method of infant brain structure.The main work and research results of this issue are as follows:(1)The low rank decomposition algorithm based on multi-atlas segmentation model is studied and implemented.The data used in the experiment is preprocessed first,and then an improved algorithm of low rank decomposition is proposed based on the dictionary learning algorithm model.After extracting the target image block and the background image block,dictionary training using the low rank decomposed image block makes the dictionary more discriminative.The method takes into account the specific part of the brain image map more comprehensively,so that the reconstruction error of the brain structure after training is reduced,and the segmentation precision is improved.(2)Aiming at the characteristics of the brain structure of the brain MR image,the improved dictionary learning algorithm based on low rank decomposition is proposed and introduced.The low rank vector is introduced into Fisher discriminative dictionary learning.The introduction of low rank vector can further enhance the discriminative ability of the dictionary.This method takes into account the local information of the brain image more comprehensively,and has a greater improvement on the discriminating ability of the edge label of the brain structure.Experiments show that the proposed improved algorithm has a better segmentation result for the brain structure of infant brain MR images,and can use the image itself to correct the general algorithm segmentation.(3)Aiming at the low contrast and poor image quality of infant brain MR images,a multi-modal low rank dictionary learning algorithm based on multi-atlas fusion algorithm is proposed.In the dictionary construction stage,the brain nuclear magnetic resonance T1 and T2 images are used to construct a dictionary together,and then the low-rank dictionary learning method is used to carry out the dictionary training process.The multi-modal atlas image can be used to guide and correct the segmentation of the brain structure.The multi-modal improvement algorithm further improves the discriminability of the dictionary,overcomes the disadvantages of low image contrast,and further improves the segmentation accuracy of the brain structure.