Research On 3D Mapping Of Mouse Brain Coordinate Framework Based On Bi-channel Registration And Deep Learning

The 3D mapping of the mouse brain coordinate framework is to define the 3D anatomical space of the acquisition dataset according to the standard anatomical coordinate framework.And the biomedical analysis under the achieved accurate anatomical positioning across the entire experiments has been indispensable in the brain research and also been the basis of neuroscience.With the launch of various large-scale brain projects and the gradually deepening of brain research,the rapid development of microscopic imaging technologies and biological processing methods have enabled the acquired cross-modality datasets to be higher resolution,larger-scale and obvious heterogeneous which has brought greater challenges to the synchronization and integration of brain coordinate framework mapping.The following problems still exist in the brain mapping:low accuracy of multi-modality data mapping,lack of the flexibility to register incomplete brain datasets to atlas in 3D,and a challenge to build a standardized and automated computational pipeline under the common coordinate framework.This dissertation focus on these problems,mainly including the followings:First,based on the conventional registration algorithm,a novel bi-channel registration scheme is proposed for mapping the multi-modality mouse brain data.The scheme first optimizes the data preprocessing process of large-scale brain images,and a dynamic re-sampling ratio is applied along anterior-posterior axis to mitigate the nonuniform deformation of brain samples relative to the reference template.Then,the three-dimensional geometric features and the two-dimensional edge texture features of mouse brain are simultaneously extracted to form a new channel to augment the image registration.The assistant channel not only improves the registration accuracy of brain images under the same modality of the reference template,but also enhances the mapping accuracy of cross-modality clarified mouse brain samples.Under the same cross-modality brain datasets,compared with a series of classic brain segmentation tools based on registration,the bi-channel registration is superior to other tools with thes higher accuracy.Then,to overcome the limitation that it is difficult to mapping the incomplete brain data due to the missing morphology information,we introduce a neural network to directly segment the anatomical regions in incomplete brains with minimal supervision.Benefiting from the bi-channel registration,we combine it with human simulation as a label data generator which can avoid experiencing time-consuming manual annotation.And in the network,each pixel is allocated,classified,and segmented into brain region after a series of feature processing stages.Under the verification of experiments,the incomplete brain datasets are segmented into anatomical structures with satisfied accuracy,and trained network is applied to random incomplete mouse brain data in 3D.In comparison with other segmentation tools based on neural network,the semantic segmentation framework we used has more obvious advantages in accuracy.Finally,we develope a bi-channel image registration and deep-learning segmentation pipeline named BIRDS for the mapping and analysis of three-dimensional microscopy data of mouse brain.The BIRDS pipeline includes image pre-processing,bi-channel registration,automatic segmentation and annotation based on neural network,creation of a 3D digital frame,high-resolution visualization,and expandable quantitative analysis.This pipeline combines the bi-channel registration with semantic segmentation network which not only provides accuracy-improved anatomical segmentation for cross-modality whole mouse brain but also gives the real-time and flexible segmentation inference for incomplete brain.BIRDS is integrated on the open-source FIJI platform and linked to Imaris software module,which is friendly for job submitted and implemented in most laboratory computing environments and demonstrates the prospects in brain neuroscience.