MEMRI offers the exciting possibility of tracing neuronal circuits in living animals by MRI. Here we use the power of mouse genetics and the simplicity of the visual system to test rigorously the parameters affecting Mn 2+ uptake, transport and trans-synaptic tracing. By measuring electrical response to light before and after injection of Mn 2+ into the eye, we determine the dose of Mn 2+ with the least toxicity that can still be imaged by MR at 11.7T. Using mice with genetic retinal blindness, we discover that electrical activity is not necessary for uptake and transport of Mn 2+ in the optic nerve but is required for trans-synaptic transmission of this tracer to distal neurons in this pathway. Finally, using a kinesin light chain 1 knock-out mouse, we find that conventional kinesin is a participant but not essential to neuronal transport of Mn 2+ in the optic tract. This work provides a molecular and physiological framework for interpreting data acquired by MEMRI of circuitry in the brain.

Microtubule-based motors carry cargo back and forth between the synaptic region and the cell body. Defects in axonal transport result in peripheral neuropathies, some of which are caused by mutations in KIF5A, a gene encoding one of the heavy chain isoforms of conventional kinesin-1. Some mutations in KIF5A also cause severe central nervous system defects in humans. While transport dynamics in the peripheral nervous system have been well characterized experimentally, transport in the central nervous system is less experimentally accessible and until now not well described. Here we apply manganese-enhanced magnetic resonance (MEMRI) to study transport dynamics within the central nervous system, focusing on the hippocampal-forebrain circuit, and comparing kinesin-1 light chain 1 knock-out (KLC-KO) mice with age-matched wild-type littermates. We injected Mn2+ into CA3 of the posterior hippocampus and imaged axonal transport in vivo by capturing whole-brain 3D magnetic resonance images (MRI) in living mice at discrete time-points after injection. Precise placement of the injection site was monitored in both MR images and in histologic sections. Mn2+-induced intensity progressed along fiber tracts (fimbria and fornix) in both genotypes to the medial septal nuclei (MSN), correlating in location with the traditional histologic tract tracer, rhodamine dextran. Pairwise statistical parametric mapping (SPM) comparing intensities at successive time-points within genotype revealed Mn2+-enhanced MR signal as it proceeded from the injection site into the forebrain, the expected projection from CA3. By region of interest (ROI) analysis of the MSN, wide variation between individuals in each genotype was found. Despite this statistically significant intensity increases in the MSN at 6 hr post-injection was found in both genotypes, albeit less so in the KLC-KO. While the average accumulation at 6 hr was less in the KLC-KO, this difference did not reach significance. Projections of SPM T-maps for each genotype onto the same grayscale image revealed differences in the anatomical location of significant voxels. Although KLC-KO mice had smaller brains than wild-type, the gross anatomy was normal with no apparent loss of septal cholinergic neurons. Hence anatomy alone does not explain the differences in SPM maps. We conclude that kinesin-1 defects may have only a minor effect on the rate and distribution of transported Mn2+ within the living brain. This impairment is less than expected for this abundant microtubule-based motor, yet such defects could still be functionally significant, resulting in cognitive/emotional dysfunction due to decreased replenishments of synaptic vesicles or mitochondria during synaptic activity. This study demonstrates the power of MEMRI to observe and measure vesicular transport dynamics in the central nervous system that may result from or lead to brain pathology.

The performance of modern face recognition systems is a function of the dataset on which they are trained. Most datasets are largely biased toward "near-frontal" views with benign lighting conditions, negatively effecting recognition performance on images that do not meet these criteria. The proposed approach demonstrates how a baseline training set can be augmented to increase pose and lighting variability using semisynthetic images with simulated pose and lighting conditions. The semi-synthetic images are generated using a fast and robust 3d shape estimation and rendering pipeline which includes the full head and background. Various methods of incorporating the semi-synthetic renderings into the training procedure of a state of the art deep neural network-based recognition system without modifying the structure of the network itself are investigated. Quantitative results are presented on the challenging IJB-A identification dataset using a state of the art recognition pipeline as a baseline.