Reconstruction of 3-Dimensional Histology Volume and its Application to Study Mouse Mammary Glands

Summary

We present an image registration approach for 3-dimensional (3D) histology volume reconstruction, which facilitates the study of the changes of an organ at the level of macrostructures made up of cells . Using this approach, we studied the 3D changes between wild-type and Igfbp7-null mammary glands.

Abstract

Histology volume reconstruction facilitates the study of 3D shape and volume change of an organ at the level of macrostructures made up of cells. It can also be used to investigate and validate novel techniques and algorithms in volumetric medical imaging and therapies. Creating 3D high-resolution atlases of different organs^1,2,3 is another application of histology volume reconstruction. This provides a resource for investigating tissue structures and the spatial relationship between various cellular features. We present an image registration approach for histology volume reconstruction, which uses a set of optical blockface images. The reconstructed histology volume represents a reliable shape of the processed specimen with no propagated post-processing registration error. The Hematoxylin and Eosin (H&E) stained sections of two mouse mammary glands were registered to their corresponding blockface images using boundary points extracted from the edges of the specimen in histology and blockface images. The accuracy of the registration was visually evaluated. The alignment of the macrostructures of the mammary glands was also visually assessed at high resolution.

This study delineates the different steps of this image registration pipeline, ranging from excision of the mammary gland through to 3D histology volume reconstruction. While 2D histology images reveal the structural differences between pairs of sections, 3D histology volume provides the ability to visualize the differences in shape and volume of the mammary glands.

Introduction

IGFBP7 (insulin like growth factor binding protein 7) is a member of the IGF-binding protein family, and has been shown to bind the IGF1 receptor⁴. Down-regulation of IGFBP7 is known to be correlated with poor prognosis in breast cancer⁵, while the reintroduction of IGFBP7 in Xenograft tumor models greatly inhibits the tumors growth⁶ through induction of apoptosis and cellular senescence⁷. In order to study the effects of IGFPB7, an Igfbp7-null mouse was created⁵ (unpublished data). While these mice do not develop tumors, they show changes in histology of the ovary, muscle and liver as well as defects in mammary gland developmental patterning (unpublished data). The defective phenotype was first indicated as the null mice have smaller litter sizes and are unable to sustain multiple large litters (unpublished data).

3D histology volumes have the potential to provide useful information for quantitative and comparative analyses and assessment of pathologic findings in volumetric medical images. Three-dimensional confocal, two-photon microscopy can provide high-resolution cell morphological information of the gland at local extent¹⁴ , but it has a limited field of view and depth. Histology volume reconstruction provides more information over a much greater spatial extent. Using traditional approaches some distortion is anticipated during the preparation of histological sections, such as shrinkage, expansion, tears, and folds. These distortions make it difficult to register serial histological images into a 3D stack to reconstruct a 3D volume. As the number of consecutive sections with defects increases the similarities between intact sections is reduced and consequently makes the registration process more complicated.

Different methods have been proposed to register histological sections and to create a continuous histology volume. Some techniques depend on intensity variations⁸, and others are based on the shape of the sections⁹. For some specimens the anatomical structures can be used as landmarks^10,11 along with landmark-based registration methods^12,13. But these internal structures might not be detectable throughout the whole volume and for some specimens no reliable anatomical structures can be identified. Some groups have used a pair-wise registration approach and registered consecutive histology images one to another using contours or anatomical structures^16-18. Registering serial histology sections to one another without the use of reference images may propagate registration error and change the actual shape of the histology volume. Pair-wise registration approach relies on the consistency of shape of the histology sections and the internal structures throughout the stack of the images; therefore it requires dense sampling of the specimen, which might not always possible, e.g., for clinical specimens.

In this pipeline we use blockface images as a set of reference images for histology volume reconstruction¹⁹. Blockface images are taken of the paraffin tissue blocks after mounting on the microtome and before each section is cut. Thus, damage to individual serial sections cut does not interfere with registration of serial sections^8,11,15. We capture the blockface images in a different way from the other groups. The optical block face images are obtained by a telecentric lens to eliminate or minimize the barrel and perspective distortion, which usually occurs when using regular lenses in optics. This is one of the advantages of the proposed approach over the other published methods, which perform blockface imaging using regular lenses. The images are taken at a slight oblique angle to use the reflection of the surface of the block for contrast enhancement between the tissue and paraffin surface and to eliminate the shadow of the tissue in depth, under the paraffin surface. A photographic filter is also used to polarize the light coming from the block surface and the tissue to balance the contrast¹⁹. To correct for the displacement of the block on the rotary microtome, two to three holes are drilled in the corners of the block, which are easily detectable in the blockface images. The centroids of these holes are used along with landmark-based rigid registration to align the blockface images.

Protocol

1. Specimen

1. Excise the mammary glands surgically from wild-type CDH1 as well as Igfbp7-null mice three days post onset of lactation.
2. Spread the glands onto glass slides to help regain native mammary gland morphology.

2. Fixation and Tissue Processing

1. Fix the mammary glands in neutral buffered 4% PFA O/N at 4 ^oC.
2. Store the glands in 70% ethanol prior to tissue processing.
3. Transfer the glands to small tissue processing cassettes.
4. Process the tissues using an automated tissue processor
   1. Dehydrate the tissues in increasing ethanol and xylene baths of 70% ethanol for 45 min, 2 times in 95% ethanol for 45 min, 3 times in 100% ethanol for 1 hr and 2 times in xylene for 45 min.
   2. Permeate the tissues with paraffin 3 times for 1 hr each in a vacuum with applied pressure.
5. Embed the tissues in paraffin to form blocks, for sectioning.

3. Histology and Blockface Imaging

1. Trim the paraffin blocks using a rotary microtome until the excess paraffin is removed.
2. Use a vertical milling machine to drill 1 mm holes in at least two corners of the paraffin block perpendicular to the cassette.
3. Mount the tissue block on the rotary microtome.
4. Set up the blockface imaging system¹⁹ in front of the microtome.
5. Capture optical blockface image prior to sectioning.
6. Cut ribbons of four sections at 5 μm thickness on the microtome.
   1. Transfer the ribbons to the cold water bath.
   2. Separate the second and fourth sections of the ribbon and mount them on microscope slides. Choosing the second and fourth sections provides a 5 µm gap between sections.
   3. Expand each section in a warm water bath (48 ^oC) to unwrinkle it, then re-mount it on the microscope slide.
      NOTE: Cutting, mounting, unwrinkling the sections cause some distortions on the section, such as tear, fold, shrinkage, and expansion. These artifacts complicate the registration of the histology sections.
   4. Stain the sections with H&E using an automatic stainer.
   5. Coverslip the slides using an automatic coverslipper.
   6. Digitize the slides using a digital histology slide scanner at the resolution of interest. For this protocol the magnification is 20x and the resolution is 0.47 μm.
   7. Down-sample the histology images to the resolution of blockface images, 18 μm.

4. Image Registration

1. Image Segmentation and point Selection
   1. In blockface images measure the pixel values of the registration holes and use the average value as a fixed threshold to segment the registration holes in the corners of the paraffin block.
   2. Since some additional parts might also be segmented by using the fixed threshold, use the circularity and the area of the segmented objects to find the holes and discard the extra objects. To do this, write a small code and find the ratio of (4π x area)/(perimeter)² for the segmented objects. This ratio for round objects is 1.
   3. For each mammary gland, select one blockface image as reference and align the rest of the blockface images to the reference by using the centre of the registration holes and landmark-based registration techniques.
   4. For the aligned blockface images, manually segment or extract the tissue from the background. Use the most sizable object in the mask for the rest of the protocol.
   5. For H&E sections follow the steps below for automatic segmentation.
      1. Use Otsu thresholding technique²⁰ to segment images from the background and create binary masks of the histology images.
      2. Identify and select the most sizable object in each mask using the histogram of the labeled objects.
      3. Extract the one pixel wide boundary points from both histology and blockface masks.
      4. Use Chain code algorithm²¹, to represent the boundary points by a sequence of piecewise linear fits.
2. Initial Rigid Registration
   1. Use Fourier Descriptors algorithm²², to find the initial rigid transform between the boundary points of histology and their corresponding blockface images. This initial transform includes the translation, rotation and scale factors.
   2. Transform each histology image with the initial transform obtained from the previous step.
3. Refinement of the Rigid Registration
   1. Remove the high curvature edge sections from the histology contour using a rolling ball filter²³.
   2. Select 500 points from the remaining histology boundary points randomly using uniform distribution.
   3. Transform the the histology random boundary points with the initial transformation obtained from Fourier descriptors.
   4. Select the whole set of blockface boundary points and use Iterative Closest Points (ICP) algorithm²⁴ to find the rigid transformation between the blockface boundary points, destination, and histology random boundary points.
   5. Transform the aligned histology images obtained from the previous step and the stack of aligned histology images creates the histology volume.
   6. Use a 3D visualization software to create a visual image of the histology volume.
4. Viewing the Stack of Images at 5x magnification
   1. Down-sample the original histology images to 5x magnification.
   2. Crop the region of interest in one of the histology images.
   3. Calculate the location of that region in other 5x histology images using the combination of the rigid transformations from the two steps of registration.
   4. Crop the regions of interest to the same size region in all other histology images.
   5. Finally refine the alignment between the regions manually. Write a program that overlays two images and allows for selecting the values for rotation and translation of one of the images over the other one and then saves the transformed image when the alignment is accepted.
   6. View the stacks of the aligned 5x histology regions using a 3D visualization software.

Results

A pitfall of traditional microscopy techniques is that the understanding of an organ at the microscopic level is limited to one field-of-view at a time. Even “total disclosure” slides, which provide entire slide sections, fail to provide three-dimensional information. With the development of whole slide, dynamic scanning technologies, our ability to see a section in its entirety has increased, however extrapolating structures requires 3D histology volume reconstruction.

To better characterize ...

Discussion

In this study, we have developed an image registration workflow to reconstruct a 3D histology volume from serial 2D histology images, which does not require internal randomly selected landmarks or implanted fiducial markers within the tissue, which might distort the tissue. By the method described, optical blockface images themselves are used as the reference images prior to sectioning. We use external holes drilled in the paraffin block to aid in aligning the blockface images and to correct for the 2D transversal moveme...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
16% PFA	VWR International	15710	16% Paraformaldehyde solution
Small tissue processing cassettes	VWR International	CA95029-956
Leica ASP300 automated tissue processor	Leica	14047643515
100% Ethanol	Fisher Scientific	S25307B
Xylene	VWR International	CA95057-822
Paraffin	Thermo Fisher	39501006	Paraplast tissue embedding medium
Leica EG 1160 embedding center	Leica
Leica rotary microtome	Leica
Milling machine	Argo
Microscope slides	VWR International	CA48312-015
H&E stain	VWR International
Automatic stainer
Coverslips	VWR International	48404-452
MEDITE RCM 7000 glass coverslipper	MEDITE
Leica SCN400 slide scanner	Leica
MATLAB	MathWorks Inc	MATLAB 2007b	Development software
MeVisLab	MeVis Medical Solutions AG	MeVisLab 2.1	3D visualization software