Paul Lauterbur closed his seminal paper on MRI with the statement that "zeugmatographic (imaging) techniques should find many useful applications in studies of the internal structures, states and composition of microscopic objects" {Lauterbur, 1973 #967}. Magnetic resonance microscopy was subsequently demonstrated in 1986 by three groups{Aguayo, 1986 #968}{Eccles, 1986 #969}{Johnson, 1986 #970}. The application of MRI to the study of tissue structure, i.e. magnetic resonance histology (MRH) was suggested in 1993 {Johnson, 1993 #957}. MRH, while based on the same physical principals as MRI is something fundamentally different than the clinical exams which are typically limited to voxel dimensions of ~ 1 mm3. Preclinical imaging systems can acquire images with voxels ~ 1000 times smaller. The MR histology images presented here have been acquired at yet another factor of 1000 increase in spatial resolution. Figure S1 in the supplement shows a comparison of a state-of-the-art fractional anisotropy images of a C57 mouse brain in vivo @ 150 um resolution (voxel volume of 3.3 x10-3 mm3) with the atlas we have generated for this work at 15 um spatial resolution (voxel volume of 3.3 x 10-6 mm3). In previous work, we have demonstrated the utility of MR histology in neurogenetics at spatial/angular resolution of 45 um /46 angles {Wang N, 2020 #972}. At this spatial/angular resolution it is possible to map whole brain connectivity with high correspondence to retroviral tracers {Calabrese, 2015 #895}. But the MRH derived connectomes can be derived in less than a day where the retroviral tracer studies require months/years {Oh, 2014 #971}. The resolution index (angular samples/voxel volume) for this previous work was >500,000 {Johnson, 2018 #894}. Figure S2 shows a comparison between that previous work and the new atlas presented in this paper with a resolution index of 32 million. Light sheet microscopy (LSM) has undergone similar rapid evolution over the last 20 years. The invention of tissue clearing, advances in immuno histochemistry and development of selective plane illumination microscopy (SPIM) now make it possible to acquire whole mouse brain images at submicron spatial resolution with a vast array of cell specific markers{Ueda, 2020 #974}{Park, 2018 #953}{Murray, 2015 #952}{Gao, 2014 #973}. And these advantages can be realized in scan times of < 6hrs. The major limitation from these studies is the distortion in the tissue from dissection from the cranium, swelling from clearing and staining, and tissue damage from handling. We report here the merger of these two methods: 1. MRH with the brain in the skull to provide accurate geometry, cytoarchitectural measures using scalar imaging metrics and whole brain connectivity at 15 um isotropic spatial resolution with super resolution track density images @ 5 um isotropic resolution; 2. whole brain multichannel LSM @ 1.8x1.8x4.0 um; 3. a big image data infrastructure that enables label mapping from the atlas to the MR image, geometric correction to the light sheet data, label mapping to the light sheet volumes and quantitative extraction of regional cell density. These methods make it possible to generate a comprehensive collection of image derived phenotypes (IDP) of cells and circuits covering the whole mouse brain with throughput that can be scaled for quantitative neurogenetics.
Large-scale automated identification of mouse brain cells in confocal light sheet microscopy images
