Pathology WG Meeting Notes 21 June 2022

Agenda:

#	Owner	Title	Time (min)
1	JH/DM	Nuclick in DSA	10
2	JH/SA	MONAI v0.9	10

Attendees: Jonny H, Lee C, Shan R, Jeff B, David M, Gregory L, Gigon B, Michael B, Sachi A

Notes:

• David M provided explanation of the mechanism that DSA uses to allow third-party algorithms to be run. This involves the spawning of a container with associated latency, which is fine for many types of longer running use-cases, but for rapid point clicking, in the case of something like NuClick, this is not practical. David has proposed a solution that would allow Python processes to be launched for this type of task.
• Sachi provided a detailed overview of the performance of Monai Label and how it has reduced the total time to result to a level that is certainly within the realms of feasibility for interactive WSI inference. Obviously the time is very dependent on the size of the WSI and the amount of foreground tissue, but we seem to be in the region of 60-100s per WSI (1 x 32GB v100 GPU). Performance of WSI Inference (MONAI Label)
• We also had a general discussion about the prioritisation of image formats for cuCIM and whether the writing of image data should be a priority. Because of CUCIMs ties with Skimage, there are options for resaving using some of its third-party libraries such as tifffile. Zarr-based formats seem to be the favoured option for writing data back out (e.g. tiles or segmentation maps).
• Some discussion around the use of DICOM. Shan R mentioned that it does seem to be gaining traction - from the community at ECDP recently.

Action items:

PIC	Item
Sachi	David M & Sachi to sync on the best way forward for NuClick integration into DSA