| The brain atlas based on magnetic resonance imaging(MRI)is an important tool for neuroscience research.Constructing high-quality brain atlas can accurately reflect the anatomical structure of a specific population of brains,provide standard space and corresponding anatomical information for the processing of different individual brains,and promote neuroimaging research.As non-human primates,macaque has a close phylogenetic relationship with humans,and is a widely used animal model in neuroscience research.The brain of developing macaque,especially early postpartum macaque,is in a relatively rapid regional heterogeneous dynamic development period,which brings challenges to image preprocessing and atlas construction.For macaque,the existing brain atlas are mainly based on adult individuals,showing a lack of multi-time point longitudinal brain atlas that can cover the dynamic changes of the brain during development.To fill this gap,we combined in deep learning and prior knowledge of brain imaging to process MR image data,and constructed the longitudinal brain atlas of cynomolgus macaque covering 12 postpartum time points(i.e.1,2,3,4,5,6,9,12,18,24,36 and 48 months).This paper focuses on the following three aspects:(1)Macaque brain extraction algorithm based on attention mechanism and cross-domain prior knowledge.Accurate brain extraction(or skull stripping)from MR images is the key preprocessing for subsequent neuroimage analysis.At present,the mainstream brain image processing tools are designed for human beings,and it is usually difficult to obtain ideal results when processing macaque brains.Therefore,brain extraction from MR images of developing macaque is challenging and of great significance.In order to solve this problem,this paper proposes the Domain-Invariant Knowledge-Guided Attention Networks based on attention mechanism and cross-domain prior knowledge,which effectively integrates image intensity information and cross-domain prior knowledge,namely Signed Distance Map(SDM)and Center of Gravity Distance Map(CGDM),and realizes the automatic brain extraction of MR brain images of developing macaque.(2)Macaque brain tissue segmentation algorithm based on anatomical prior knowledge and full resolution architecture.Accurately dividing three types of brain tissues(gray matter,white matter and cerebrospinal fluid)in brain images is very important for constructing brain atlas and studying early brain development.However,the changes of tissue contrast and anatomical morphology caused by severe brain development have brought great challenges to the automatic segmentation of brain tissue.In order to solve this problem,this thesis proposes Anatomy-guided Full-resolution Attention Networks to realize automatic segmentation of the brain tissue of early postpartum macaque.Specifically,we use signed distance maps based on the Outer Cortical Surface as anatomical prior knowledge,and embed these prior information into the proposed full-resolution attention network to guide brain tissue segmentation in unclear regions.(3)Construction of multi-time point longitudinal brain atlas of cynomolgus macaque from 0 to 48 months after delivery.This paper uses the above algorithm to preprocess a large-scale(39 subjects,a total of 175 scanned images)longitudinal multi-time scanning developing cynomolgus macaque data set.Based on the preprocessed brain images and brain tissue segmentation maps,we complete the construction of longitudinal brain atlas.The 4D brain atlas reflects the typical anatomical structure of the macaque population at a specific time point in the development period,and shows highly efficient cross-individual effectiveness.In addition,the brain atlas covers the dynamic developmental changes of macaques throughout the development period,so it is suitable for macaque brains of any developmental level,and can serve as a reference for brain anatomical structure information and a brain template for neuroimaging research. |
PDF Full Text Request