Construction Of Interactive Tool For Mouse Brain Image Registration And Mapping

Brain function and structure are highly correlated,and the key to understanding brain function is to analyze neuroanatomical differences on a large scale.However,the differences in morphological characteristics of brain samples hinder the statistics of neural circuit data of multiple samples in a uniform anatomical space.To align cross-sample and cross-modal images,many studies have developed a number of professional tools for automatic or manual registration based on image registration algorithms to map mouse brain images to reference atlas.However,the traditional registration method based on grayscale only applies the low-level local grayscale information,which is susceptible to the sample morphological characteristics and image quality.Therefore,these automatic registration tools are not suitable for non-ideal data with large non-uniform deformation,sample defects and lowcontrast texture,while manual registration tools that rely solely on artificial markers are too inefficient.The purpose of this study is to develop a new method and a new tool for robust registration and mapping of multi-modal mouse brain images,particularly non-ideal data,to atlas.The main research contents are as follows:Firstly,a registration method based on Anatomic Context(AC)is proposed.The most significant sign of image misregistration is the anatomical context in the registered image,that is,the original constraint relationship of the key anatomical structure(the parallel or vertical relationship of multiple planes)is disordered.In this paper,the constraints of anatomic context are introduced into the registration process as grayscale features to resist the disorder and mismatch of anatomic structure caused by unconstrained registration transformation.The specific scheme is as follows: firstly,pre-alignment and texture enhancement are used to reduce the uncertainty of registration;Then,the key anatomical structures are extracted semi-automatically,and to blended into the original image by gray inversion,making it show gray mutation in these positions,so as to give priority to alignment in the subsequent mutual information based registration and achieve robust global matching.Finally,the registration is further refined through interactive landmarks.Qualitative comparison and quantitative evaluation verify the effectiveness of each step of the scheme.Secondly,a general registration and mapping framework for multimodal mouse brain image was developed.This paper develops a fully interactive and efficient processing tool called the Brain Registration,Identification,and Encoding Framework(BRIEF for short).The tool provides a friendly interface and interaction mode to annotate,encode,and manage anatomical features and landmarks,which are introduced into the registration process through a modular,multi-page interface.Using a custom atlas and Imaris API,BRIEF quantifies and visualizes the distribution of neurons in the entire brain at a cellular resolution within a few hours.In this thesis,BRIEF was used to register multimodal data,and demonstrated the application of research in cell type-specific neural circuits and imaging data reconstruction.The research in this thesis will provide new methods and tools for neuroscientists to map neural circuits efficiently and promote the integrated utilization of mouse brain images obtained under non-ideal conditions.BRIEF have been applied to several topics of specific neural circuit analysis,and the achievements have been transformed and applied to BioMapping9000,a commercial product.