Hi Richard and All, At Formulatrix we haven't yet seen an algorithm that can come even close to what the human brain can do for drop scoring. Crystals come in too many shapes, sizes, plates, and confusing backgrounds making automated detection extraordinarily difficult. Because of this, we have found users will not trust or use an automated scoring algorithm for visible light. Plus, these algorithms usually only reliably pick out the easy crystals which a human can usually do in a flash by just looking at 96 thumbnails. If there is a an algorithm out there that is successful with visible light crystal detection, we would love to hear about it.
Indeed, as Jose and Zhijie points out, when you factor in additional detection technologies like UV and SHG (Second Harmonic Generation a.k.a. SONICC) then image processing can start to be helpful. In this case, however, Formulatrix only uses image processing to resort the images to put drops most likely to contain crystals at the top of the list. We call this auto scoring but it's really auto sorting. We never pitch this as infallibly able to score drops, because it certainly isn’t. Interestingly, the additional information from UV or SHG combined with auto scoring doesn't necessarily reduce the time looking at drops. Yes, you can find the easy crystals much faster with these techniques. But, what we see instead is we are giving users even more information to look through and the advantage isn’t a time savings for looking at drops. Instead, you find smaller crystals and find crystals you otherwise would have missed. This gives you more hits that you can optimize sooner. We would love to hear further opinions, especially contrarian, from the community on this topic.
