Properties- not getting output in Consensus

I generated three label files as per the Manual segmentation rule provided in AtlasTool kit. After that I used tool kit and 3 D label registration and open those three labels. After that used TOOL and Synchronized the labels then Image- Properties but I am not getting anything in Consensus. window.
Is my segmentation process is wrong ? Please advice.

Welcome to the forum, @Sonali!

I have invited the author of the Atlas Toolkit plugins, Timothy Grocott, to join the discussion here on the forum. Hopefully he can help answer your questions.

Hi Sonali,

Although we’ve been emailing, I should put my responses here so other users can benefit. I think I see the problem from the three ‘.label’ files you kindly emailed.

You have created three ‘.label’ files, each one labelling a different structure within the same heart (e.g. left ventricle, right ventricle, etc.) and then attempted to use Label Registration 3D to align those structures, one on top of another.

Having repeated this myself, I can confirm that the resulting consensus.tif file is an empty image stack, in other words no consensus morphology was found. The reason it failed is because the plugin is not designed to force alignments between non-homologous structures (e.g. left ventricles with right ventricles).

Here is an example of what the plugin is designed to do:

  • Take three different image stacks of three different hearts.
  • For each heart, use the Segmentation Editor to label the left ventricles, which are homologous structures sharing a common morphology.
  • Use Label Registration 3D to align those left ventricles, one on top of another. You may have to play around with the parameters to get the best result (I need to write a guide for choosing parameters).

Did you try aligning the test data supplied with the paper? I emailed a Dropbox link as an alternative download source.

1 Like

Technical discussion on properties of the consensus.tif:

You are correct that consensus.tif is not properly calibrated. You can see in the screencast that I manually adjusted the calibration after registering to ensure correct 3D rendering of the object.

The reason for this is that I was unsure how to handle the case were ‘.label’ files with different calibrations are group-wise aligned. In pair-wise alignment the moving image would simply adopt the calibration of the fixed image, but what to do when all of the images are moving? Moreover, what to do when ‘.label’ files with non-isotropic calibrations are rotated? And how to detect the a rotation anyway, since that is done implicitly by bUnwarpJ?

In hindsight each ‘.label’ file should simply be re-sampled so that all have identical and isotropic calibrations prior to registration. Of course that means the ‘.ots’ files ought to include information about the re-sampling. Unfortunately, making that change will break backwards compatibility of ‘.ots’ files as they’re implemented as serialized objects.

I have just fixed an issue with the Cluster Viewer tool so it works with the new 3D Viewer plugin (and presumably breaks compatibility with the old 3D Viewer). Once I’ve got to grips with Maven versioning and uploading that fix, I’ll turn my attention to solving the calibration issue.

1 Like

Tim,
Thanks a lot for your reply. I tried the test data sent by you. I was wondering after seeing the test data because all three label files are very similar. I got it where I am going wrong. If I take two different hearts, and label each region in the heart such as right ventricle , left ventricle, Atrium etc then create two label files for two heart. Instead of going for one region alignment can go for multiple regions at time by using 3D registration?

Once again thanks

Thanks Leader. I appreciate your help.

Yes, but with one caveat: the Label Registration 3D tool currently allows you to select only one region for registration. You can see in the screen cast that the ‘Select Channel (0-255)’ parameter is set to one (meaning the first labelled region).

On possible workaround is to set the ‘Select Channel’ parameter to zero, meaning you will register the empty non-labelled region, which is roughly equivalent to registering all labelled regions. I just tried that with my test data, and it kind of works: the ‘.ots’ files are calculated correctly, and can therefore be used to transform the original image stacks, but the ‘consensus.tif’ is a mess and isn’t much use for some of the other tools like Cluster Viewer.

A more satisfactory solution might be to label the entire heart as one region.

I’ll add multiple region selection to the list!

1 Like