Scalable mapping of myelin and neuron density in the human brain with micrometer resolution

Abstract

Optical coherence tomography (OCT) is an emerging 3D imaging technique that allows quantification of intrinsic optical properties such as scattering coefficient and back-scattering coefficient, and has proved useful in distinguishing delicate microstructures in the human brain. The origins of scattering in brain tissues are contributed by the myelin content, neuron size and density primarily; however, no quantitative relationships between them have been reported, which hampers the use of OCT in fundamental studies of architectonic areas in the human brain and the pathological evaluations of diseases. Here, we built a generalized linear model based on Mie scattering theory that quantitatively links tissue scattering to myelin content and neuron density in the human brain. We report a strong linear relationship between scattering coefficient and the myelin content that is retained across different regions of the brain. Neuronal cell body turns out to be a secondary contribution to the overall scattering. The optical property of OCT provides a label-free solution for quantifying volumetric myelin content and neuron cells in the human brain.

Enhanced Multiscale Human Brain Imaging by Semi-supervised Digital Staining and Serial Sectioning Optical Coherence Tomography

“…Comparing the profiles between OCT-SC and DS in different layers suggests that our DS model works beyond our approximate linear biophysical model (21) and increases the local contrast by a nonlinear mapping function expressed by our neural network.…”

“…This approach proves effective because the OD image exhibits similar image contrast and feature distribution as the OCT-SC across various cortical regions. Additionally, the OCT-SC demonstrates an approximate linear relationship with the OD of the Gallyas silver stain (21). Furthermore, since the OD map is naturally pixel-aligned with the PS image, it facilitates supervised learning and provides additional semi-supervision and alignment constraints for the main DS model.…”

“…This enables accurate reconstruction of the complex 3D structures of brain tissues without requiring sophisticated inter-slice registration algorithms. Despite its ability to routinely generate large-scale volumetric brain imaging data, S-OCT still requires considerable expertise to identify and annotate anatomical and neuronal features for further analysis (11,14,17,21). Our goal is to augment S-OCT with a digital staining (DS) technique that enables straightforward 3D histology on large-scale human brain tissues.…”

“…Firstly, we devise a pseudo-supervised learning module by training the DS network on a pseudo-paired training dataset that is generated using our previously established biophysical model. Our previous work has revealed a linear correlation between the OCT scattering coefficients (SC) and the optical density (OD) computed from the Gallyas silver stained image (21). Based on this similarity prior, this module learns to translate the generated OD back to the Gallyas silver stain, acting as a proxy supervision for learning the translation from OCT-SC to Gallyas silver stain.…”

Segmentation of supragranular and infragranular layers in ultra-high resolution 7Tex vivoMRI of the human cerebral cortex

“…Using LSFM registered into the MRI space, we demonstrate that the MR contrast boundary in cortical gray matter corresponds to the cytoarchitectural boundary between layers III and V, visible on the NeuN stain (Figure 2(A)). Based on the myelin density differences [36,37] and their resulting contrast in MRI, we group layer I, II, and III together as the supragranular layer, and layer IV (absent in some regions), V, and VI together as the infragranular layer. Bias correction was applied for all scans.…”