As old mate said: "the mind is not a vessel to be filled, but a fire to be kindled". Glad to see it burning brightly, & likely to spread further.

This is also a lot more than an article for me. @felkohdigital.bsky.social is a PhD student in my team, & these are ideas he & I & we as a team discuss a lot; not so much the analogies (though sometimes). It's a beautiful thing to read it in his words, & to see his passion for the ideas.

This is the result of a super fun collaboration with Andrew Gunawan from @locusj.bsky.social and Erik Meijering's groups at UNSW (there aren't many collaborators I'd brave a 9-11 hour time difference for, but these guys are great 🦘) and @micoxscopy.bsky.social a bit closer to home! (5/5)

For giggles, we also discovered that you can deliberately force images to be structurally wacky while still getting high scores on quality metrics (please don't try this at home) (4/5)

Most alarmingly, it turns out that sometimes these metrics tell you that your image has got better after image processing, even when your downstream biological analysis gets worse as a result ☹️ (3/5)

We show that PSNR and SSIM fundamentally suck (technical scientific term there) for fluorescence microscopy in general. They also love processed images - even if those processed images don't look all that great. (2/5)

Follow-up stories coming soon on our next-generation beyond #Extensible_Immunofluorescence (i.e. microscopy image data augmentation with Generative AI) & on #EMT definition + dynamics & on E/M state-dependent computation across diverse stimuli + signalling. Watch this #Systems_Microscopy space 🔬👍

For those interested in the regression / ML-based #Computational_Multiplexing strategy used by @felkohdigital.bsky.social here, it builds on DL-based #Extensible_Immunofluorescence we demo in www.nature.com/articles/s41.... You can build multiplexed IF data from standard IF with either method.