Why aren't we using machines instead of humans to catch stalls?

Since the topic has been rekindled on the chat (mainly by @Plenum – thank you!), sharing this blog post about the topic here: https://blog.eyesonalz.com/machines-or-people/

Plus, as an update to the above (from @pietro, copied from chat today) – we are developing machine-based methods to help catch stalls in three ways:

  1. to automatically trace the blood vessel networks, identify vessel segments, and draw outlines around them

  2. to learn to find and remove “bad” vessels (edge vessels, fat vessels, and 10 other flavors of “bad” vessels)

  3. to annotate the very easiest vessels and save the more challenging ones for catchers! (our current system can detect and annotate the 10% easiest vessels - but we are still validating that method so it is not yet in the SC pipeline)

over time, improvements to each of these three machine-based interventions will hopefully improve the catching experience and help focus human cognition where it’s needed most

2 Likes

There is now a machine learning challenge aimed at finding the best algorithms for analyzing Stall Catchers data! Unlikely any of them will replace the keen human eye any time soon, but they could label the “low hanging fruit” - easiest vessels, leaving the more difficult ones for us, and speeding up the research even more.

Read all about it & find the link to the challenge on the blog: The machines are coming! (but the humans are staying) 🤖🤝🧔

1 Like