Gamma dose comparison should allow sub-voxel search

Question

Gamma dose comparison should allow sub-voxel search

Closed this issue 5 years ago · 13 comments

The gamma dose comparison should allow sub-voxel search, and that choice should be enabled by default. The necessary line of code to enable sub-voxel search is:

gamma->set_interp_search (true);

Answer 1 · 2019-07-11T14:32:44.000Z

Thanks for adding this ticket!

Could you please provide me some text that I can as the tooltip? I think it's important to describe exactly what it interpolates, especially that the same checkbox is repurposed.
Thanks!

Answer 2 · 2019-07-11T16:13:20.000Z

Thanks Greg and Csaba!

I think the 'Use linear interpolation' checkbox (calling resample-nn) should be eliminated all together.
For the new checkbox to enable the sub-voxel search, maybe we should use the same terminology as the Ju et al 2008 paper. Checkbox can be labelled 'Use geometric gamma calculation' or something like that.

The tooltip could read:
"By checking this box, gamma will be calculated according to Ju et al 2008, which finds the point with the minimum gamma value by using the normal vector between the two candidate points."

Answer 3 · 2019-07-11T16:15:10.000Z

Thanks, @kmalexander5 !
I agree, I repurposed the old linear interpolation checkbox for the interp_search function.
So we don't want to mention interpolation at all? It seems odd considering the name of the Plastimatch function.

Answer 4 · 2019-07-11T16:20:02.000Z

True @cpinter, but the paper is actually titled 'Geometric interpretation of the gamma dose distribution comparison technique: Interpolation-free calculation'. So adding in the word interpolation to the check box might seem odd!

Answer 5 · 2019-07-11T16:22:24.000Z

OK so it seems that the Plastimatch function name is misleading a bit.

Anyway, I'll make these changes then. Thanks!

Answer 6 · 2019-07-11T16:23:33.000Z

What do you think @gregsharp?

Answer 7 · 2019-07-11T16:26:55.000Z

Should it be on by default?

Answer 8 · 2019-07-11T17:58:53.000Z

I added the option

It is on by default.

Please test it in tomorrow's preview build and close ticket if works as expected.

Answer 9 · 2019-07-11T18:13:29.000Z

Awesome - just seeing these now. Thanks @cpinter!

Either way - unchecked or checked is fine. I'd tend to suggest to leave it off by default so the user can make a conscious decision to use it.

I imagine I won't be able to do a proper test until the back-end that is resampling the evaluated volume in Plastimatch is eliminated (@gregsharp). Regardless - I'll check it out! Thanks.

Answer 10 · 2019-07-11T18:24:41.000Z

It actually makes a big difference. I just tried the standard IGRT test data and checking it increased pass rate from 80% to 94% (as I remember). So I suggest you do try it tomorrow. That said I should update the slicelets too. I'll do that now

Answer 11 · 2019-07-16T17:51:13.000Z

Are things working as expected? Can I close this issue?

Answer 12 · 2019-07-16T18:11:33.000Z

Seems to improve the comparison and produce more reasonable values. I'd say this can be closed, as the back-end Plastimatch issues will be corrected on gitlab. Will do more thorough testing when Plastimatch is updated. Thanks @cpinter !

Answer 13 · 2019-07-16T18:14:10.000Z

Excellent, thanks! Yes, further changes need to happen at their respective places. Since this particular issue seems to be solved, I'm closing the ticket.