Image normalization

[asmagen]

Is there an option to normalize the ndpi images to account for faint slides and stains in such a way they won't fall below the light threshold, and similarly, darker stains wouldn't fall above the dark tissue fold threshold?

Thanks

[choosehappy]

unfortunately, that is a whole area of research which has not yet come to a conclusion

you can read some of my work on the topic (pay particular attention to the other references, some of which have open-source code available): https://www.ncbi.nlm.nih.gov/pubmed/27373749

as well as newer stain gan type approaches: https://arxiv.org/abs/1804.01601

On Thu, Apr 16, 2020 at 5:05 PM Assaf Magen notifications@github.com wrote:

Is there an option to normalize the ndpi images to account for faint slides and stains in such a way they won't fall below the light threshold, and similarly, darker stains wouldn't fall above the dark tissue fold threshold?

Thanks

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/choosehappy/HistoQC/issues/173, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACJ3XTGX6JIGDYQ2JF6M3ATRM4NC7ANCNFSM4MJYWOUQ .

[asmagen]

Thanks for the references. I am inclining to not use any normalization since it could bias my quantification of % positive pixels for a given molecule staining, because in lowly expressing stains the normalization would render more than expected pixels to appear positive relative to a highly expressing stain. Do you think this concern applies to the gan type normalizations?

[choosehappy]

yes, i think it especially applies to gan normalization approaches. given their blackbox nature, it will be very hard to predict what the output would be in all situations. i would take a more hand-crafted approach where at least one could start to envision what the edge cases might be are

On Sun, Apr 19, 2020 at 7:31 PM Assaf Magen notifications@github.com wrote:

Thanks for the references. I am inclining to not use any normalization since it could bias my quantification of % positive pixels for a given molecule staining, because in lowly expressing stains the normalization would render more than expected pixels to appear positive relative to a highly expressing stain. Do you think this concern applies to the gan type normalizations?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/choosehappy/HistoQC/issues/173#issuecomment-616190537, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACJ3XTEZWAQ3VKD4TLAW6G3RNMYPNANCNFSM4MJYWOUQ .