-
Notifications
You must be signed in to change notification settings - Fork 1.1k
Pathology WG Meeting Notes 06 December 2022
# | Owner | Title | Time (min) |
---|---|---|---|
1 | Sachi | HoverNet in Monai Label demo | 0 |
2 | All | Priorities for 2023 | 0 |
David Manthey, Sachidanand Alle, Michael Boone, Jonny Hancox, Gigon Bae, Gregory Lee, Lee Cooper, Jeff Baumes
Sachi provided a demo of the latest version of Monai Label with QuPath (not with Hovernet yet). Initial experiments show that HoverNet is significantly slower for larger ROIs - especially on lower-end GPUs, but this is to be expected in this first release. Jonny suggested that the next pipeline that we focus on should be Tissue/Lesion Segmentation, which would complement the more computationally challenging nuclei segmentation. Reasons for this would be:
- Providing scope to apply some of the 2D techniques already developed for 2D radiology use-cases (e.g. DeepEdit)
- Complementary to the existing nuclei segmentation pipeline
- Much faster to compute - so better for GUI interactivity We should identify some clinically relevant use-cases rather than just providing an algorithm.
Lee & Jeff concurred with this direction.
Lee suggested that MONAI could offer pretrained unsupervised task weights - perhaps trained on TCGA (need to check licence restrictions).
Lee mentioned that there is an issue with CuCIM (#370). Gregory & Gigon are somewhat time-bound with other commitments until the new year.
PIC | Item |
---|---|