Research And Software Design Of Mouse Brain Segmentation And Annotation

The basis of any mouse brain efforts requires objective and accurate determination of the brain anatomical boundaries,which directly defines the identify of cells or neuronal connections,a process that crosses the interpretation and comparison of the entire experiment.In recently years,large-scale brain mapping has become a major endeavor in neuroscience research.The purpose is to understand fundamental and pathological processes in the brain,and the precise anatomical location is particularly important for neuron mapping and loop analysis.Some mouse brain mapping projects have made great achievements,but developing a robust and high-precision system in these projects for the segmentation and annotation of mouse brain datasets is still a major challenge.This paper focuses on anatomical boundary segmentation and annotation problem from two aspects: improved registration algorithm and image segmentation techniques of deep learning.It is difficult for traditional registration method to ensure enough accuracy on the obviously deformed brain dataset,and the improvement of the registration algorithm is significant important.In addition,considering the time-consuming and lacking of data generality of the registration algorithm,we try to introduce deep learning technology to solve the classic segmented problem.Finally,a highly operable,interactive software system was built to implement the functions and can be used in the laboratory.The main research contents of this paper include three parts.In the first part,the brain region segmentation and annotation are based on the image registration algorithm.During the registration,the inherent morphological features of the mouse brain were additionally used to generate a binary mask-assisted registration.Finally,the accuracy of the proposed registration scheme is apparently improved as compared with the conventional method.In the second part,we trained a deep neural network based on the registered datasets,and applied it to the direct segmentation of the mouse brain region.We sufficiently verified the robustness and segmentation accuracy of this deep neural network segmentation method through a variety of experiments.The third part is to design a complete mouse brain segmentation and annotation software based on the Java language.Module encapsulation of each function,and multi-thread technology were used to improve the efficiency and feasibility of the system.