Single Field Winston Lutz Idea

[ethanio12345]

Hi there,

Had an idea for the single field WL test and just wanted to soundboard it as I'm not really sure how to code it within the Pylinac framework. Y'all may have already thought of it and have reasons against it. I'm fairly new to working on big projects (and this is my first suggestion ever...am a medphys reggie), so am welcome to feedback.

If we had a large, symmetric, CAX centred, MLC defined field, we should be able to simulate smaller sub-fields and not lose significant accuracy in the positioning? Assuming the MLCs are visible in an image, and we know the nominal positions, then we should be able to calculate, fairly accurately, the positions of other points on an EPID? I.e., if we know/trust EPID readouts from the DICOM (offsets/roll/etc), then I'm wondering if could we apply a mask around where we expect each BB to be in the image, which breaks the problem down into a ``multi field'' problem, which you've already solved with the multi target implementation?

Here's my algorithm thoughts

• take large MLC defined field to give us field centre.
• apply a binary mask (or just crop) near the ideal BB position (arbitrary field size, could also use Gaussian filter if we want to simulate field edge)
• Apply WL searching algorithm to search for single BB within sub field
• Repeat for all combinations of BB positions and gantry/col/couch
• Calculate idealised BB projection and compare measured position to expected position
• Summarise and work tech magic.

I've got some very basic code working for an image of MPC where I read the XIM, crop, thresh, openCV to find contour as a proof of concept. You can use a simple one-liner for the BB projections, as they are just an Euler rotation using extrinsic coords, which you can pull from ``scipy.spatial.transform import Rotation''. This allows calculating gantry, coll and couch pretty easily...you just need to make sure the rotations match (e.g. rotation of phantom is inverse of gantry)

Just an idea/test concept. Love this project and thanks for all the work!

[jrkerns]

Hey, thanks for bringing this up.

To be honest, I'm a bit confused about what you're actually proposing. Are you wanting to do something like MPC where the displacement from the ideal positions is reversed to get the gantry/coll/couch positional error? If so I don't think you'll be able to get away with a single image for that!

If you are referring more to an MPC-like analysis where there are multiple BBs but one large field vs the multiple subfields of the SNC multi-target WL, it's in our queue 😉.

I was not aware of the scipy.spatial.transform module. I might be able to simplify some of my code with that. I'm not very good with transforms so I'll have to do some research.

If you restate your problem and what output you want I might be able to get more specific.

Thanks

[ethanio12345]

Only for the purpose of making a dad joke (and not to be pushy), you should ``git'' onto it. :P

[jrkerns]

push-y 🙃

[ethanio12345]

https://media.tenor.com/UladBrWhuJ8AAAAM/bowing-thank-you.gif

[jrkerns]

Okay yeah I'm stupid, I can get it in no problem. I'm not quite ready to have it replace the existing functions though before I write some tests. If you revert the changes inside _nominal_point (i.e. just the function exists w/ no usages) and make a PR, then I'll merge it. That way the function is there and ready when I get to it and you'll also get git blame 😄

[ethanio12345]

Ok. Think I've done it. Made a dumb, tried to correct the dumb, realised it wasn't really a dumb and I was just confusing myself. PR made.