May 24, 2022 (Dev Meeting)

Agenda:

#	Owner	Title	Time (min)
1	Sachi and David	Update on MONAI Label	20
2	David and Lee	Update on expected inference time	5
3	Behrooz	Update on MONAI Pathology Core	10
4	Behrooz	Feedback on MONAI Microscopy	10
5	Johnny	QA for histopathology	10

Attendees

Notes

• David has timed inference of MONAI Label on the whole slide images up to size of 126kx200k (16 GB GPU Mem, 64 CPU MEM), processing on the foreground, and has initial timing but will have the improved ones soon.
• MONAI Label uses openslide but Sachi will switch to WSIReader from MONAI
• Challenges for Microscopy:
  • Image reading: LargeImage supports many formats in microscopy, cucim does not have immediate feature to support them, Gigon suggests AICSImageIO
  • Image dimensions: 3D images, time dimension, more channels (hyper spectral), stage position, etc. In microscopy image dimension can be generalized as STCZYX [sites, time, channel(or wavelength), z, y, x]. S - Sites or multiposition locations.
  • More viz opportunities
• WSIReader should be modified to not use cucim cache during training and use it only for inference. Here is a notebook discussing cucim cache.

Action items:

PIC	Item
David	share the inference time results with the WG