A multimodal computational pipeline for 3D histology of the human brain

Abstract

Ex vivo imaging enables analysis of the human brain at a level of detail that is not possible in vivo with MRI. In particular, histology can be used to study brain tissue at the microscopic level, using a wide array of different stains that highlight different microanatomical features. Complementing MRI with histology has important applications in ex vivo atlas building and in modeling the link between microstructure and macroscopic MR signal. However, histology requires sectioning tissue, hence distorting its 3D structure, particularly in larger human samples. Here, we present an open-source computational pipeline to produce 3D consistent histology reconstructions of the human brain. The pipeline relies on a volumetric MRI scan that serves as undistorted reference, and on an intermediate imaging modality (blockface photography) that bridges the gap between MRI and histology. We present results on 3D histology reconstruction of whole human hemispheres from two donors.

Figure 1

Workflow of data acquisition (orange) and computational processing (green). The ex vivo brain is scanned, dissected, sectioned and stained, providing data for the pipeline: the MRI volume, the whole and blocked slice photographs (dissection), the blockface photographs (sectioning), and the stained sections. The flowchart illustrates the main steps of the pipeline: stacking of blockface photographs; registration of blockface volumes to slice photographs; blocks initialization; block mosaic-preprocessed MRI alignment; and MRI-histology registration.

Figure 2

Overview of the main MRI preprocessing steps. (a) The raw ex vivo MRI data as acquired, showed in sagittal (left), axial (middle) and coronal (right) views; (b) brain extraction and bias field correction given by SAMSEG; (c) segmentation or subcortical structures given by SAMSEG, combined with cortical segmentation and parcellation provided by FreeSurfer, with the outline of the reconstructed surfaces in yellow; (d) 3D rendering of the cortical surface from lateral (left) and medical (right) views. The color coding of the segmentation and parcellation follows the FreeSurfer convention.

Figure 3

Overview of the dissection procedure. (a, b) Brain hemisphere from lateral (a) and medial (b) views; (c) incision; (d) subsequent brainstem transection below the mamillary body; (e) overview of how the anterior and posterior slices are cut: the forebrain is cut coronally; (f) forebrain slices and remaining brainstem and cerebellum portions: the brainstem is cut transversely while the cerebellum is cut sagittally; (g) example of block cut planning; (h) example of blocked slice photograph; (i) example of blockface photograph, including the microtome setup.

Figure 4

A flowchart of the computational pipeline. This flowchart represents the main processing steps, subdivided by the related subsections in the main text. Where relevant, the flowchart specifies what algorithm or software was used, or if any manual intervention was needed.

Figure 5

An example of blocked slice photograph to blockface photograph registration. The three images on the top show the three types of photograph we are dealing with: (a) a sample whole slice photograph; (b) the related blocked slice photograph; (c) corresponding blockface photograph of the bottom left block. The three images on the bottom show the outcome of the registration: (d) block of interest in the blocked slice photograph; (e) aligned blockface photograph; (f) blocked slice photograph with blockface photograph overlaid in transparency.

Figure 6

Blockface and histological volumes. Example of reconstructed blockface volumes (a) from three different views (coronal on the left, axial in the centre, and sagittal on the right). The same views are also showed for the correspondent histological volumes, stained with LFB (b) and H&E (c).

Figure 7

Initialization and refined blockface mosaics. A surface rendering of the whole brain mask derived from the MRI is overlaid in green. (a) Blockface volume for the whole hemisphere resulting from the initialization with the slice photographs; the color for rendering is taken from the blockface photographs. (b) Same volume as (a), where each block is rendered in a different color. The color coding is random and simply emphasizes the block cutting profiles. (c,d) Same volumes as (a,b) after joint registration.

Figure 8

Registration error in 3D MRI-histology alignment. (a) 3D visualization of the landmark pairs after registration. The red spheres represent the landmarks annotated on the MRI. The green spheres represent the corresponding landmarks on the histology, after shifting them with the deformation field estimated by our registration algorithms. The blue lines connect landmark pairs, thus illustrating registration error. (b) Distribution of the landmark registration error (mean: 1.50 mm, standard deviation: 1.61 mm; median: 0.89 mm). (c,d) 3D visualization of landmark pairs after registration and distribution of the landmark registration error for the second processed hemisphere (mean: 1.44 mm; standard deviation: 1.80 mm; median: 0.80 mm).

Figure 9

MRI-histology alignment. Histological sections from a sample block aligned with the MRI volume as a result of the refinement algorithm, showed from three different views (top: axial; middle: coronal; bottom: sagittal) and for the two stains (left: LFB; center: H&E). On the right, we have overlaid semi-automated segmentations of brain structures made on the LFB sections, including the hippocampus (yellow), cerebral cortex (red), cerebral white matter (white), pallidum (dark blue), putamen (bright pink), claustrum (green), thalamus (dark green), caudate (light blue), hypothalamus (light pink), substantia nigra (bright green), and subthalamic nucleus (violet).

Figure 10

Registration between MRI and histological sections. (a) Sample MRI slice with registered LFB section overlaid; (b) magnified view of the 5 $\times$ 5 mm area marked in red; (c) corresponding region on LFB section; (d) MRI and LFB overlapped in transparency; (e–g) magnification of the 30 $\times$ 30 mm area marked with the green square in (a); (h) another sample coronal MRI slice; (i) corresponding slice in reconstructed LFB mosaic; (j) overlay of (h) and (i) in transparency.