DeepMapi: a Fully Automatic Registration Method for Mesoscopic Optical Brain Images Using Convolutional Neural Networks

Hong Ni,Hui Gong,Xiangning Li,Anan Li,Tao Jiang,Yue Guan,Wu Chen,Qingming Luo,Zhao Feng,Jing Yuan,Qiuyuan Zhong,Xueyan Jia,Miao Ren

doi:10.1007/s12021-020-09483-7

Abstract

The extreme complexity of mammalian brains requires a comprehensive deconstruction of neuroanatomical structures. Scientists normally use a brain stereotactic atlas to determine the locations of neurons and neuronal circuits. However, different brain images are normally not naturally aligned even when they are imaged with the same setup, let alone under the differing resolutions and dataset sizes used in mesoscopic imaging. As a result, it is difficult to achieve high-throughput automatic registration without manual intervention. Here, we propose a deep learning-based registration method called DeepMapi to predict a deformation field used to register mesoscopic optical images to an atlas. We use a self-feedback strategy to address the problem of imbalanced training sets (sampling at a fixed step size in nonuniform brains of structures and deformations) and use a dual-hierarchical network to capture the large and small deformations. By comparing DeepMapi with other registration methods, we demonstrate its superiority over a set of ground truth images, including both optical and MRI images. DeepMapi achieves fully automatic registration of mesoscopic micro-optical images, even macroscopic MRI datasets, in minutes, with an accuracy comparable to those of manual annotations by anatomists.

Highlights

Deconstructing fine neural structures at the cellular level is critical in understanding the connections and collaborations of brain networks (Economo et al 2015; Gong et al 2016; Electronic supplementary material The online version of this article contains supplementary material, which is available to authorized users.Huang and Luo 2015; Li et al 2015; Oh et al 2014)
We present a method, called DeepMapi, which is based on a convolutional neural network (CNN) to predict the deformation field corresponding to each pair of images and used to automatically register mesoscopic micro-optical imaging datasets to a reference atlas
The recently developed deep learning registration methods have concentrated on improving the loss function and refining

Summary

Introduction

Neuroscientists often manually delineate the brain regions and nuclei in which neurons are located (Fürth et al 2018; Lein et al 2007; Lin et al 2018; Osten and Margrie 2013) with the help of a brain stereotactic reference atlas. The rapid development of neural circuit labeling methods and whole-brain imaging technologies (Gong et al 2016; Li et al 2010; Ragan et al 2012) have resulted in brain images becoming increasingly complicated at the mesoscopic level. The advent of the terabyte-scale (TB-scale) mouse brain dataset (Landhuis 2017) has promoted the need for a stable and reliable high-throughput automatic registration method suitable for these large datasets

Methods

Results

Conclusion