August 30, 2022 (Dev Meeting)

Agenda:

#	Owner	Title	Time (min)
1	Behrooz	Pathology Patch MetaTensor	30
2	Lee, Behrooz	Read WSI at magnification/mpp instead of level	30

Attendees

Behrooz Hashemian, Lee Cooper, Ziyue Xu, Jonny Hancox, Prerna Dogra, David Manthey, Gigon Bae, John Pocock, Gregory Lee

Notes

• We discussed about the MetaTensor details for MONAI pathology, and specifically what are the important metadata to be included for patches extracted from histopathology whole slide images.
  • The importance of using such metadata in pathology has many aspects. One is to be able to use this metadata to extract the original patch or know the location of each patch in the whole slide image for many post-processing applications. Besides post-processing, it also provides a way to distinguish the patches related to a specific WSI and know if the entire slide is processed for usages like inference to save the output and free up the memory for the following processes.
  • We realized that there can be two notion of patch metadata, one that gives us the required information to extract the patch again, and another one that describe the information about the transformed patch image. We agree to focus on the former for the v1.0 release and continue discussion about the later.
  • We agreed on the patch metadata to be included in the next release:
    • location
    • size
    • level
    • mpp
    • downsample
    • magnification
    • path
• Currently in MONAI we are using the convention of HxW (y,x) for the tensors, so we want to keep this for digital pathology. Although it might be confusing for the people coming from computational pathology where the convention of (x, y) to define locations. We agreed that it is better to be consistent with the rest of MONAI and take care of this discrepancy in WSIReader. Lee suggested to have a clear documentation on this to avoid any confusion.
• We also discussed about reading at a specific magnification/mpp (which does not necessarily exist as a level of pyramid in the file) in more details.
  • Behrooz explained two ways to achieve this in MONAI either inside WSIReader (which results in a simpler API) or as combination with a special transform (which provide more flexibility and traceability).
  • Lee and David were advising to use a tolerance on mpp too.
  • David were suggesting to make it a relative tolerance
  • John mentioned that in his experience magnification would be more useful than mpp
  • More discussion on the implementation details is needed.

Action items:

PIC	Item
Behrooz	Check on license of CoNIC with Daguang