Manual Registration and Pyre

When automated registration fails the Pyre (Python Registration) tool is used to repair the transform for later operations on the volume.

Pyre commands

The blinking point is the currently selected point

Mouse

Left Button

<Left Click> to select an existing point

Shift + <Left Click> to add a new point

Alt+Shift + <Left Click> to add a new point and auto-align

<Left Click> + drag to move point under the cursor

Ctrl + <Left Click> + drag to translate entire warped image

Alt + <Left Click> to move currently selected point to mouse position

Right Button

Shift + <Right Click> to delete point under the cursor

<Right Click> + drag to move the view

Scroll wheel

<Scroll wheel> zoom in or out

Ctrl + <Scroll wheel> to rotate warped image

Ctrl + Shift + <Scroll wheel> to rotate warped image slowly

Keys

A,W,S,D Move the view

Page Up/Down Change the magnification

M Match the view on all windows to look at the same point as the current window (Not Functional for Warped Image)

L Show transform mesh lines

f Flip the warped image

Space Auto-align the selected point

Shift + Space Auto-align all points

Ctrl+Z to undo a step

Ctrl+X to redo a step

Tab Change properties of the view. A warped image may be displayed as it appears registered. The composite view will switch to a different view.

Finding registration errors

Brute force registration is the first step in registration. If the initial rotation or translation is incorrect then later refinement stages will fail.

Each group of stos transforms lives in a directory at the same level as volume sections. Nornir creates overlay images for every transform in the group showing how well the slice aligned with the adjacent section.

The fastest way to detect misregistration is to open the overlay images for a stos-group and look for problems. For examples we will use a build of the 6263 PMG dataset found in test files and built with the CMPBuild and CMPAlign pipelines.

Examing the output of the final slice-to-slice grid refinement pipeline with a downsample of 2 the Glutamate (“E”) section not aligning well with a Glycine (“G”) section. The relative path of this image under the 6263 volume directory would be:

/TEM/Grid2/overlay_5-6_ctrl-E_LeveledShadingCorrected_map-G_LeveledShadingCorrected.png
../_images/overlay_5-6_ctrl-E_LeveledShadingCorrected_map-G_LeveledShadingCorrected.png

This mis-registration is close but still a failure. Modifying the pipeline by adding iterations to the refinegrid operation may fix this error, but that is time consuming.

Fixing errors

When working with large images the preferred method is to examine an early slice-to-slice registration stage, usually a group named “brute” or “grid” with the highest number. Pyre is used to quickly create a simple transform containing only a few points to provide the correct the angle and translation. The rest of the pipeline is then re-run. This is usually sufficient to allow later refinements to succeed which provide an alignment quality beyond what a human will typically take the time to do.

Stos Group Folders

Each stos group folder has the following content

  • <GroupName>

    • *.stos

    • Overlay_*.png

    • Warped_*.png

    • Automatic

    • Manual

The manual folder provides a replacement for the stosgroup’s normal output. By placing a file here with the same name as an output .stos file we are telling the buildscripts to skip automatic registration and simply to copy the manual .stos file up to the group folders .stos files.

The automatic sub folder contains the input stos-group transforms adjusted to contain the proper input filters and scale. It does not need any interaction.

Repairing a .stos file

Pyre is a GUI provided with the Nornir for the manual generation of stos files. Pyre can write both .stos files or registered images. This enables one to bypass the entire nornir-buildmanager if desired.

To fix our misregistration we will launch Pyre from the Python scripts directory:

python start_pyre.py

Loading data into pyre

Drag and drop the stos file of interest from the stos group directory onto the Pyre UI. (I have seen bizarre behavior from Drag and Drop in Wx Python. If Drag Drop does not work please load using the File -> Open Stos menu option)

If there is no .stos file available you may drag images into the moving or fixed image windows of Pyre directly.

The result is a window containing images to register. Use the Window menu to automatically arrange the Pyre windows.

../_images/6263_Pyre_start.png

Clear points

The first step is to clear the points from using the Operations -> Clear all points menu item

../_images/6263_Pyre_ClearedPoints.png

This leaves four points about the edge. I then place new points at the correct corners. The fastest method is shift + left click on the first image. Find the same spot on the second image and bring the most recently created point to that position by using alt + left click.

Once at least three points are placed we can delete the original three points by selecting them with shift + <right click>.

The result is a roughly aligned image.

../_images/6263_Pyre_RoughAlignment.png

Auto placement

Pyre helps you align the points if desired. Hitting spacebar will attempt to automatically align the current blinking point. CTRL+Z will undo the last change for the point.

If this is the first stage of slice-to-slice registration this registration is likely sufficient since it will be refined by later stages. In our case I loaded the .stos file from the last refinement stage so whatever I save to the manual registration folder will be the final product. Because of this I continued placing points until the registration was quite accurate.

../_images/6263_Pyre_final2.png

Export

When you are done select File -> Save Stos from the menu. Save the .stos file in the manual directory of the stos group you loaded the original .stos file from.

File -> Save Warped Image saves the registered image directly to disk.

Final Result

Once all manual registrations are in place one can re-rerun the CMPAlign script. This produces new output images incorporating the transform fixes generates new registered images.

The image below shows the final product after using the manual corrections from this tutorial. GABA, Glutamate, and Glycine were assigned to Red, Green, and Blue channels respectively. Glutamate to Gl

../_images/6263YEG-RegistrationSample.png