Clinical validation of realistic annotation protocols

Validate the proposed whole-slide image annotation protocols directly with pathologists to establish whether they accurately reflect clinical annotation practices.

Background

The paper introduces annotation protocols based on localized clicks and scribbles, including representative, error-driven, and human-in-the-loop interactions. These protocols are evaluated through automatically generated annotations derived from dense segmentation masks, rather than through direct use by practicing pathologists. The authors identify direct clinical validation as an unresolved next step needed to confirm the realism of the simulated annotation procedures before routine clinical deployment.