Alignment quality metrics

Development
#1

I’d like to have some estimation of how much of an outlier each of my landmarks is as I’m working through a 3D CLEM alignment. When I’m doing tilt series alignment for tomography, IMOD (eTomo) gives me a residual error value for the image of a fiducial that I’m tracking through the projections. I then have to correct the large residuals - usually the landmark is not aligned with the fiducial, so you have to move it to make them line up. As you do that, your mean residual error drops. This way, you have values to indicate how good your alignment is (although you have to be careful to not read too much into this).

So, I’m guessing that a Similarity or Affine BigWarp transform would make the LM and EM landmarks align with each other in 3D space less well than a Thin Plate Spline. Would there be a way to put a value on this difference and indicate it in the landmarks table? It would be a sort of equivalent to showing on the image how much distortion is needed to make the datasets fit each other. But with values, you could identify the landmarks that are creating a lot of distortion and try to fix them. Doing a manual FIB SEM CLEM alignment is quite an iterative process, and I’m really making it up as I go along. It would be nice to have a way to guide you through the process with some kind of metric for how good your alignment is.

1 Like
#2

@bogovicj I’ve also added a few thoughts/issues on the BigWarp github pages but I wondered if they’re more appropriate here?

I’m less familiar using image.sc and github since I’m coming from the bio research/microscopy side as opposed to the coding side as @Martin_Jones will tell you!

I actually started a Slack Workgroup for people doing Bio EM (with about 200 people so far), and one of the reasons was to facilitate this sort of interaction between people who write the software and those who use it. Link here if you’d like to join in;
EMofCTO Slack

1 Like
#3

Hi @mattrussell

I think this is a great idea!

Adding a column to the table would be straightforward and pretty easy, but I worry a little about is making the table too “busy.” I’ll probably start with that and see how it goes. Would you be willing to alpha-test?

Thanks for the feedback,
John

1 Like
#4

Hi @bogovicj - Great, I’m happy to alpha test. I take your point about the table getting too busy. Maybe there could be a separate table just listing those values? I think that’s how it works with eTomo, where you can get a list of residuals.

Actually, in IMOD (which I’m assuming everyone knows but I realise that’s not necessarily the case!), eTomo shows the residuals on the images, with an arrow pointing from a ‘landmark’ in the direction of the error.

It’s more involved than a table, but if we can show arrows on the image indicating the direction of the difference at each landmark, that would also be very helpful. Not sure how that could be implemented in 3D, though!

Matt

1 Like