Published work has identified that tissue staining variability can negatively impact on AI performance. The work presented in this talk builds on our previous study that quantified stain variability using animal control tissue from 247 labs who were participants in the UK NEQAS CPT external quality assurance scheme. That analysis was conducted using colour deconvolution from the control tissue compared with expert assessors scores. The results showed a wide landscape of stain variation but also implied a weak correlation of haematoxylin to eosin ratio with assessors score. This new study repeats the previous work to determine if the landscape has improved since the feedback from the first study but also now includes companion results from the stain assessment tool Tango, developed in the UK National Pathology Imaging Cooperative (NPIC ,Leeds,UK).The results will show whether the feedback from stain variation has improved stain consistency in the participating laboratories and also demonstrate the utility of the Tango stain assessment tool as an adjunct to control tissue.Ultimately this data, along-side the results from other studies in NPIC looking at stain variability using the Tango tool, and scanner QC tools, builds the evidence for considering standards for digital pathology. In turn this would provide a more consistent data set which could underpin more generalisable AI with shorter development times. Suggestions for standards will be presented to provoke discussion of the need.

Learning Objectives: 

1. To quantify the breadth of stain variability from a large cohort of international laboratories.
2. To present the results of a companion tool, Tango, used alongside control tissue to objectively quantify staining.
3. To provoke discussion of the needs for standards in image production for digital pathology.