Stitching Multi ROI stacks

Hey Ilya,

while MIB offers so many fantastic tools for image alignment, I was wondering if there was an option to stitch multiple stacks of multi-ROI image acquisitions with serial EM data. We were imaging a blood vessel, that kept on “dividing” or rather it had to be imaged with several ROIs, before ultimately converging into one.

The problem is, that we did not know where the vessel would move to, so we were adjusting the FOV window a lot and not at all times, the ROIs do overlap, in order to keep imaging time as short as possible. Now all ROIs have a big canvas size with lots of empty space after alignment. Additionally, we were adding and removing ROIs when the vessel converged or split up. This is giving me a hard time with Big Stitcher and Fiji alternatives at least.

Are there any existing features that I miss, or any plans to add any in that regard? Maybe you have also have a tip which program can help me with that. I currently do not have access to Amira unfortunately.

Keep up the great work,

Leo

Hi Leo,
I clearly see your point. I had couple of times the similar problems. I was thinking to come-up with a solution, but I did not have time to allocate enough time to make such stitching tool. The maximum was to have a custom script for large scale 2D stitching.

On the other hand, TrakEM2 (have you tried it?) from Fiji worked quite fine to stitch a pair of ROIs. Not sure, whether it will work with ROIs that appear/disappear… I think in such cases, it may be required to do stitching of subvolumes and finally stitch them all together…

You should also realize that stitching of ROIs that are heavily spread produces datasets with large empty areas that have no information but take space on HDD or in memory. In my understanding, such scenarios would require a different approach. Each collected block is described with a bounding box information (which comes from x, y, z coordinates of the stage) and can be displayed with a client that renders the area relevant to the current view. The whole think should be organized similar to Google maps, but be compatible with 3D.

@normanrz Could webKnossos be a solution?

Ilya

1 Like

Thanks for the ping @Ilya_Belevich!

webKnossos would be a good tool for visualizing the data, once it is stitched. webKnossos will be efficient with HDD space due to compression. For visualization, it will only load the data that is required for the visible viewport.

Unfortunately, webKnossos doesn’t have a tool for interactive stitching at the moment. However, we are thinking about adding such a tool and would be grateful for feedback on requirements.
My company offers a commercial service for stitching/aligning multi-ROI of 3D-EM datasets. Let me know if that would be of interest to you.

Hi Norman,
Even though the stitching is good, it is an overkill for many situations. As you have all functionality already present, you just need to allow possibility to work with multiple datasets. When the view hits a certain area, you identify which dataset should be rendered and after that identify the exact position inside the dataset. I think that would be really nice feature to have.

Visualizing a montage of multiple ROIs is already supported in webKnossos. It can be achieved by be assembling a dataset with multiple “layers”. Each layer can have its own bounding box. The layers can be overlapping and will be blended as such. In order to assemble such a dataset, the coordinates need to be known.

There is no GUI function for creating a dataset like this at present, but it can be done with a Python script. I am happy to create a sample script, if this is of interest :slight_smile:

1 Like