Methods for Deep-Brain Connectomics Combined with Functional Imaging, Applied to Grid Cells; and Machine Vision Approaches to Calcium Imaging Segmentation

Abstract

A major ambition at the heart of modern neuroscience is to understand how neurons interact to produce circuit-level activity, behaviors, and states of cognition. Current limitations in experimental technologies often preclude direct testing of theoretical models of these phenomena. This thesis creates and refines methods that narrow this technological gap in systems neuroscience, to allow for improved testing of theories about neural function and anatomy. First, we develop deep learning methods for automated neuron detection in large-scale calcium imaging. We find that deep convolutional networks can achieve near-human accuracy at superhuman speed, surpassing the popular PCA/ICA method. Second, we refine methods for large-scale connectomics on functionally identified neurons, and apply them to sequence neurons in retrosplenial cortex during a working memory task. 2-photon calcium imaging was combined with virtual reality behavior, multimodal image registration, and microCT-assisted heavy-metal staining. This enabled collection of a ~1mm³ electron microscopy (EM) tissue volume containing hundreds of functionally-characterized cells. Finally, we establish methods for deep-brain connectomics on functionally identified neurons, and apply them to grid cells in entorhinal cortex. Using microprism implants with specialized landmark registration and extraction techniques, we measured the responses of individual grid cells during virtual navigation, then successfully recovered a ~1mm³ EM volume encompassing the recorded cells. Overall, these experiments demonstrate the feasibility of combined functional imaging and complete circuit recovery in mammals during both complex behavioral tasks and in deeper brain regions. By advancing methods for the direct testing of neural circuit theories, we anticipate that this work will aid in providing fundamental insight into the circuit architectures underlying memory and cognition.