Adversarial Platipy #168
Comments
|
Hi @SimonBiggs, this all sounds really exciting. Very keen to progress things on this front. Since this is shaping up to be a research project (or sub-project) I would suggest setting up a kick-off meeting with everyone interested as well as @LoisHolloway. We'd need to break down the tasks here so that we can properly split up the tasks and make sure we have the appropriate resources to be able to do this. It's also always a good idea to agree on authorship (in particular first authorship) upfront. I think there is a lot in this so there may even be opportunities for more than one paper depending on how things progress. So I think a meeting would be a good next step to brainstorm and start to formulate all of this. I imagine Shrikant would also be interested to be involved in this. |
|
Sounds good, I'm coming to Sydney next week and could potentially come to Liverpool in person on Thursday. Might that work? |
|
Hi Simon,
Sorry I’m flat out on Thursday with a workshop, would probably have time to say hello but not much more than that. Next week is a bit of a nightmare as everyone was away this week. I’m not sure if Phil might have capacity? Lois
From: Simon Biggs ***@***.***>
Date: Friday, 18 November 2022 at 3:52 pm
To: pyplati/platipy ***@***.***>
Cc: Lois Holloway (South Western Sydney LHD) ***@***.***>, Mention ***@***.***>
Subject: Re: [pyplati/platipy] Adversarial Platipy (Issue #168)
Sounds good, I'm coming to Sydney next week and could potentially come to Liverpool in person on Thursday. Might that work?
—
Reply to this email directly, view it on GitHub<#168 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AVTDR4RZQGGGOQUV4ZC35WLWI4DPDANCNFSM6AAAAAASD5QAWI>.
You are receiving this because you were mentioned.Message ID: ***@***.***>
This message is intended for the addressee named and may contain confidential information. If you are not the intended recipient, please delete it and notify the sender.
Views expressed in this message are those of the individual sender, and are not necessarily the views of NSW Health or any of its entities.
|
|
All good. That makes sense. Let's just keep it to video call then and go with another time. Friday's are a preference for me. |
If it helps, I personally am not going to be vying for first authorship, and I suspect either you @pchlap, or @rnfinnegan should be taking that position depending on the level of commitment being given. And @pchlap, it seems to align very well with your PhD. So you might naturally be doing the most work on it, and therefore potentially it should be your first authorship? Anyway, that's much more between you and @rnfinnegan, so I'll leave that for you. If others are happy with it, I would appreciate the final author position if that's something other's aren't looking to be in. |
|
Hi @SimonBiggs, I'll be around on Thursday so you're welcome to come by for a visit. But I think in regards to this project a zoom call to capture everyone (potentially) involved would be important to kick this off. My main concern is ensuring we have the appropriate resources and capacity to progress this in a timely fashion (or map out when we might have that capacity). I can arrange an appropriate time for a call on our end, a Friday should be manageable (although it won't be this Friday). So we'll try to get:
in the call. Did I miss anyone? |
|
Rob :)
…On Tue, 22 Nov 2022, 8:26 am Phil Chlap, ***@***.***> wrote:
Hi @SimonBiggs <https://github.com/SimonBiggs>, I'll be around on
Thursday so you're welcome to come by for a visit. But I think in regards
to this project a zoom call to capture everyone (potentially) involved
would be important to kick this off. My main concern is ensuring we have
the appropriate resources and capacity to progress this in a timely fashion
(or map out when we might have that capacity).
I can arrange an appropriate time for a call on our end, a Friday should
be manageable (although it won't be this Friday). So we'll try to get:
- Simon
- Phil
- Jake
- Brani
- Daniel
- Shrikant
- Lois
in the call. Did I miss anyone?
—
Reply to this email directly, view it on GitHub
<#168 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ABSBK62PGZHYMXKZILEOLATWJPSJ3ANCNFSM6AAAAAASD5QAWI>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
|
haha of course, he's kind of important. OK I'll try to set something up. Phil |
@rnfinnegan, that was a great talk on the work you and @pchlap have been doing in Platipy undergoing approximately anatomically relevant deformations.
As discussed at the conference I had a thought that it might be quite helpful to utilise that functionality within an autosegmentation testing suite. A deformation can be undergone, and it can be used to detect whether or not the segmentation has got worse. The difference in the segmentation quality can then be combined with https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.basinhopping.html to then search for a "worst performing deformation". This can then be used to give feedback to the model developer, in their search for failing cases. It might also be able to be a very useful metric in reporting one aspect of the generisability of a model. Essentially one could have an "Adversarial Dice" metric and an "Adversarial Hausdorff Distance" metric, and others.
If everyone was interested it would be quite nice to pull in @JakeKendrick1, Brani, and Daniel and create the implementation and write a paper together.
As a disclosure from my end, it would be very helpful for me and my product, as having a range of robust autosegmentation testing tools that I can include within my GitHub CI suite would help me make a significantly more robust product. This would also be a nice component to be able to show to the TGA through my regulatory process.
Keen to hear what the next steps in the collaboration are seen to be.
The text was updated successfully, but these errors were encountered: