Multiscale 3-dimensional pathology findings of COVID-19 diseased lung using high-resolution cleared tissue microscopy

Guang Li; Sharon E. Fox; Brian Summa; Carola Wenk; Aibek Akmatbekov; Jack L. Harbert; Richard S. Vander Heide; J. Quincy Brown

Image Processing. The raw data acquired from the diSPIM system were transferred to a processing workstation (Dell T7910 with Xeon E5-2699 v3 processor and 128 GB RAM). The first step was to conduct affine transformation and interpolation to shear vertical light-sheet images by 45 degrees horizontally plane-by-plane with assistance of data batching and parallel computation in MATLAB. 15 Afterwards, illumination correction was conducted along the depth of each image, and all stacks of reconstructed volumes were stitched together to form the whole volume in FIJI. 16 Then, the two 3D volumes of fluorescence images corresponding to the eosin and TO-PRO-3 channels were combined into a pseudo-color image in order to mimic the color palette of hematoxylin and eosin using the Virtual Transillumination Microscopy approach developed by Giacomelli et al. 5 Visualizations of the 3D images were performed on both the single channel as well as pseudo-H&E volumes. Full volume visualizations were realized using Amira (Thermo Fisher Scientific) on 4 times downsampled images. After regions of interest were identified, sub-volumes were loaded at full resolution into Amira and the OpenViSUS 17 interactive framework for further visualization. Volume renders for Figure 2c were created with a mixture of the single channel nuclear (TO-PRO-3) and pseudo-H&E volumes to visualize the cellular and tissue architectural detail using standard texture-based volume rendering. 18 The visualizations for hyaline-fibrin aggregate and infected pneumocytes used the pseudo-H&E volume with the volume rendering transfer function designed to give semi-transparent stroma. The transfer function of the nuclear field was designed to give high intensity values an opaque, purple color while making mid to low intensity values transparency. For the megakaryocyte and infected cell cluster visualization, the single channel nuclear volume was added with standard volume shading applied to give perceptual hints of nuclei shape. Rendering order was enforced to make the nuclei field appear on top, avoiding nuclei color attenuation from the stroma in the pseudo-H&E volume. Rendering of the small vessel pulmonary microangiopathy was performed in a two-pass approach. First the single channel nuclear volume was thresholded to give all mid-to-high intensity nuclei maximum intensity. After this thresholding, this new field was blurred (Gaussian with a 30 pixel support). This blurring removed the contribution of regions with low concentrations of nuclei while maintaining the regions with large clusters to isolate the high cell density in the vessel thrombus. A similar transfer function as above was applied to this new field.