Demo optimise - unpaired abdo CT

[agrimwood]

Improve training/performance of unpaired abdo CT model

This issue is an extension of 543

Can we improve the demo so that the final trained model has practical utility for users' research?
tagging @YipengHu @ebonmati for now, feel free to assign yourself and others (can't find SB!)

The easiest recommendation I can think of is to add more CT data, as previously discussed.
If there are any more suggestions regarding hyperparameter tuning and/or model modifications please feel free to make recommendations.
To begin, I will try to implement a simple improvement(s) before end of Jan.
From February onwards, happy to introduce more complex improvements and even work towards transfer learning ticket.

[YipengHu]

@agrimwood - @ me know when you had a chance to have the data looked/ready.

[YipengHu]

@agrimwood - @ me know when you had a chance to have the data looked/ready.

@agrimwood I am going to start to look at the data now, but do let me know if any progress from your side when you see this.

[agrimwood]

Hi @YipengHu , I had a look at the following at the start of December:

• Synapse data (requires signing up to d/l): 30 abdo CTs with 13 anatomical labels per volume.
• Medical Decathlon: various image volumes organised by anatomical site, but can't find any labels.
• Cancer Imaging Archive: 82 patient CTs with pancreas segmentations (contrast enhanced, non-pathalogical volunteers).

[mathpluscode]

Hi, @agrimwood please let me know before you start training any models? Please be aware that PR #602 will change network structures and it's better to do any big training after the merge.

[agrimwood]

Hi, @agrimwood please let me know before you start training any models? Please be aware that PR #602 will change network structures and it's better to do any big training after the merge.

Will do!

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.