Automatic Whole-Brain Mapper Using Localization and Registration

Abstract

Brain imaging is an indispensable tool in neuroscience research. In many cases, research in this field involves capturing images of thin slices of animal brains using powerful high-resolution microscopes. The brain consists of billions of neurons interconnected in a highly sophisticated manner. To understand the role of a group of neurons in animal behavior, target neurons are fluorescently labeled. Then the brain images are compared to standard brain atlases to identify the spatial regions in the brain where these target neurons are expressed. Brain atlases consist of structural outlines labeling the different regions in the brain. Deforming an atlas to align the regions in a brain image is also known as image registration. In this work, we explore techniques to automatically identify the atlas that a brain image belongs to, register the atlas to the brain image, and count the number of expressed neurons in the sub-regions of the brain. The final deliverable is a software application capable of performing all the tasks with minimal human supervision.