Register 3D volumes¶

Avizo project with sample data

Data description¶

Sample: Neutron Images of Titanium 3D print object
Dataset #1: Full sample, high-res, no grating, absorption image
Dataset #2: Half sample, low-res, interferometry with grating, absorption and darkfield images

After some cleanup (crop, extract subvolumes) of the two datasets, we have good slices, and similar tight bounding box of the sample, ready to be further processed and compared. Dataset #1 is now about 800x800x607 in 3D dimension, Dataset #2 about 325x315x103. You can see that Dataset #2 is almost 3x3x3 subsample of Dataset #1.

Prepare data¶

Masking¶

To remove noise surrounding the sample, we apply a mask:

Use Interactive Threshold module to make binary image
Use Image Morphology modules (e.g. dilation/erosion) to create a mask
Use Arithmetic module to multiply mask to original data

Extract subvolume¶

Dataset #2 has two samples stacked together, we separate the sample we need using Extract subvolume module. Do same to both absorption and dark field images.

Crop¶

Two kinds of cropping are applied with crop editor

Crop bad slices (glue, not sample)
Crop the full sample Dataset #1 to contain the same part as half sample Dataset #2

Flip¶

In crop editor, do “flip z” to Dataset #2, so the vertical direction of the sample matches Dataset #1.

If you go through both dataset slice by slice in sample vertical direction, you should easily see correspondence of the defection regions.

Note

Dataset #1 (high-res) has more slices than Dataset #2 (low-res), but the spatial dimension (slice * voxel size) should be same – they are the same sample physically

Segmentation¶

Absorption images for both dataset are segmented so we can do auto skeletonization.

Howto:

open segmentation editor, use magic wand to select the defected region slice by slice
smooth segmented labels if necessary

Visualization¶

We mainly use Volume Rending, bounding box, and ortho slice modules for visualization.

To show the defects with volume rendering, we use “volrenWhite” colormap ( or “volrenRed”, “volrenGreen”, in the same category), but set the color used for values greater than max range to be completely transparent (A=0) by clicking the color icon.

For the colormap data range, set the initial values by clicking “Edit -> Adjust rang to -> Data window” first, then adjust manually using the slider. See table below for numbers used in this tutorial.

volume rendering parameter setup¶
dataset	colormap min range	colormap max range	alpha
Dataset #1 abs image	data window min	1.28 e-03	0.1
Dataset #2 abs image	2.84 e-03	2.95 e-03	0.25
Dataset #2 df image	1.36 e-03	data window max	1

The screenshot below is a split-window view of Dataset #1 (top-left) and Dataset #2 (top-right:abs, bottom-left:darkfield, bottom-right:label), visualized with volume rendering and bounding box.

About colormap legend¶

The Avizo colormap editor works great, but when you want to save the current colormap and plot its legend, some bugs are present and may cause confusion to users. Here is a workaround I use to get a reasonable colormap legend plot:

Select “volume rendering” module, in its “Properties” panel, adjust colormap range so your volume rendering looks good. Then click Colormap -> Edit -> Options -> Edit colormap. The colormap editor window will show up. At the top left corner, click the “save colormap” icon, save to your project folder. The saved colormap module will show up in your Project view, but when you select this module, the range shown at its “Properties” ports sometimes are not the same! Don’t panic. Here is what you do:

Make sure the port Datafield is set to “NO SOURCE”.
Go ahead and edit the port range boxes (left and right) so it’s the same as the corrent range you are using
Click the “connect” icon (right arrow in a circle) from colormap module, choose Annotate -> Colormap Legend module, then click “create”. The legend will show up in view window, and you can configure how it looks from its “Properties” panel ports.

Manual registration¶

Reference: Avizo user guide (version 9), Chapter 6.1 Getting started with spatial data registration using the Transform Editor

Howto:

Use Dataset #1 as reference, don’t transform this data
Open transformation editor for Dataset #2, use the “Transformer” tool to translate/rotate/iso-scale the data until it matches closely to Dataset #1

Note

Click the “3D cube wireframe with green nodes” icon in the “Properties” panel toolbar to open the transformation editor. Click the icon again to close it. You can Copy/Paster transform data between different datasets

Auto registration¶

After manual registration to a fairly close point, we use auto registration to refine the result.

Reference: Avizo user guide (version 9), Chapter 6.4 Registration of 3D image data sets

After many trying, turns out registration using abs or darkfield data directly don’t work. On the other hand, use segmented data (label) for registration gives pretty good result, once manual registration is relatively close. This is not bad, consider we need to do segmentation anyway, and we expect better segmenation to result in better registration.

Add Register Image module to label data (segmented binary) of Dataset #2. Set the ports like below:

Model: the data to be transformed by registration – label data of dataset #2
Reference: the data to be registered to – label data of dataset #1
Metric: “Normalized mutual information” or “correlation”
Transform: “rigid”, and “iso” if you want scale
Action: “register” (don’t use “align centers” and “align principle axes” in this case)

The lattice and transform data of Dataset #2:

Lattice info: 103 x 325 x 315, uniform coordinates
voxel size: 1 x 1 x 1
translation: 169.677 32.4205 -245.503
rotation: 0.594714 -0.523558 0.610083 125.351
scaleFactor: 2.6023 2.60231 2.60232
scaleOrientation: 0.890414 -0.259096 -0.374208
center: 205 372 507

Landmark-based registration¶

Reference: Avizo user guide (version 9), Chapters

6.3 Registration with landmarks, warping surfaces and image.
6.3.1 Creating landmark sets
6.3.2 Registration with Rigid Transformation