I am reaching out to the wisdom of the crowd. This is my problem. I need to scan this whole piece of tissue. I have over 100 slides to scan. All pieces are in the same place within the slide and roughly have same size. Has anyone done something similar? Thanks.
This should be exactly the sort of problem the openflexure microscope is good for I wonder if I could break down the problem slightly:
- How can you position the slide on the microscope so the sample is in the right place (that probably means defining what “the same place within the slide” means)
- What field of view do you use/how big is the sample
- I guess the samples are not a regular shape?
@JohemianKnapsody has done some preliminary thinking about how to scan samples that aren’t square, but so far I think has only tested it in simulation. @j.stirling made a version of the “slide riser” for the v7 microscope that should position the sample more repeatably, and perhaps you could customise that to put the slide in the right place?
At this point I’m really wishing we had a nice “home” feature - but provided you don’t switch the microscope off in between times, you should be able to repeatably return to roughly the same place just by moving to the same set of coordinates. So if you remember where you started on the last slide, you ought to be able to scan it, swap in the next slide, and return to the same place to find the next sample.
Thanks so much for your insight Richard!. The problem is not much with homing the samples as it is for focusing them. I made this little adapter to place the samples at the right spot each time.
Lets say my samples fit in a 6x5 grid like in this picture.
I start by finding the right focus and then move the stage to the start point and zero the coordinates. Since the samples is not square it will try to focus on a blank space and move on. If the second space is also blank it will focus again. When it reaches the tissue it will be so far off that it cannot focus anymore. I could start my scan on 3rd square from the start but then I will be missing part of the tissue from the 3rd and 4th row.
I see the problem! Hopefully @JohemianKnapsody will get around this with his improved scan algorithm - I’ll leave him to chime in with details once he’s ready. The trick will be to notice when the focus goes crazy, and recover in a sensible way.
Just for the record, I think Joe’s approach so far (following a suggestion from @j.stirling) is to do a snake scan as you have above, but instead of always going to the end of the row, it will turn around as soon as it realises it’s left the sample. The plan is that we’ll use the autofocus results to figure out when the sample has gone, which I hope should work farily well, and we can remember the last Z position that was nicely focused, so it doesn’t “wander off”. This ought to work pretty well for samples like the one you show, if you start in a corner and work up. It won’t work for really crazy shapes, but I think it’s the most sensible place to start - the trade-off is that anything that works for really crazy shapes results in a way more complicated series of moves…
If there any merit in checking the image before autofocussing? I am just thinking that if you are in focus, you then move 1/2 a FOV in x or y, and you can now see nothing. It is very unlikley that the problem is focus.
I picked on focus because it’s probably quite a good way of detecting an empty sample in a content-independent way; if your sharpness vs Z curve suddenly goes from nicely peaked to totally flat, it’s probably a good way to detect that the sample’s vanished. If you are lucky enough to have a sample that just goes white when you’ve passed the edge, you’re right that there may well be an easier metric that detects when there is nothing there.
Fair point I suppose there are two parts to be solved independently.
- How to decide if anything is there
- What to do if nothing is there
If we solve #2 then you there can be many ways to do #1 some which may work best for different samples.
@dgrosen this is the exact problem I’ve been thinking about recently. As @j.stirling says, there’s a few options to identify if you’re looking at the sample or the background, and ideally after identifying the background we’ll have an algorithm to avoid capturing empty images. Currently I’ve written something that prunes each column when it detects background, but this doesn’t guarantee capturing the entire sample if it has an unusual shape.
Tomorrow I’ll have a look at improving the scan pattern and ways to detect background, and update here
Thanks so much! Never thought to draw so much attention. I think this is a fundamental problem in all whole slide scanners. Some commercial scanner use a low power 1x-2x low resolution scan to identify the tissue within the slide and others use background detection. It will be very cool if the system can detect nothing is there and turn around. Another option is that if nothing is there, remember the last focal point and move to the next field. That way it will not be so far off when it encounters tissue again.