← Back to context

Comment by amluto

5 hours ago

re: imaging red blood cells

The super-resolution trick as they’ve done it is highly reliant on the sparseness of the bubbles. If you imagine a point or a very sparse set of points at low resolution, you can fit for the locations of those points even though you don’t see them clearly. This is a common technique in radio astronomy and (I assume although I don’t have personal knowledge) astrometry, and compressed sensing was an extremely hot field a while back.

But RBCs are weird squishy things, and they fill the bloodstream quite densely, and ChatGPT estimates that they’re spaced about 20µm apart and that, when confined to a capillary, they’re about 7µm long. (And that sounds at least plausibly correct to me.)

So, even ignoring the much worse scattering properties of RBCs, they not nearly as sparse. You mostly lose a whole dimension of sparseness and up trying to resolve the entire capillary. Which seems possible but much harder. Unfortunately, brain capillaries are about 40µm apart, so the result might be a mess.

The article did not say what wavelength they’re using or what their native (wavelength/2) resolution is.

6 comments

amluto

Reply
Aurornis  4 hours ago

Showing us a technique that is entirely reliant on sparseness and then saying they hope to employ it on something that isn’t sparse at all (blood cells) does feel misleading.

I’m filing this in the category of technologies I wish could be true, but for which no plausible path to overcoming the obvious limitations has been provided.

  • nick238  3 hours ago

    From the bubble center plot, I'm guessing that the bubbles are separated on average about a few mm apart? Taking the other comment's guess at face value, you're going from about 2 mm to 20 um, so 2 orders of magnitude. Air (technically SF6 in the article) and water (RBC is close enough) have acoustic impedances that differ by 3.5 orders of magnitude.

    My assumptions here are *extremely generous*, i.e. favorable to the "oh, we'll just make it work with natural contrast", and even then, they can't hand wave 5-6 orders of magnitude of improvement. Furthermore, because of the use of super resolution, I'm guessing there's some exponential factor in there, i.e. double the density of bubbles/tracking points past some critical limit, then you need 8x the data to reconstruct things.

    • amluto  2 hours ago

      As another way to estimate this, here’s a data sheet for some microbubbles:

      https://pdf.benchchem.com/1673/Application_Notes_and_Protoco...

      So 1-5e8 bubbles per mL, and let’s suppose you inject 5mL. (I have no idea what the human dose is, but that’s what’s in this particular kit.)

      You apparently have 5e9 or so RBCs per mL of blood:

      https://en.wikipedia.org/wiki/Complete_blood_count

      You have about 5L of blood, so that’s three orders of magnitude more volume than the contrast, and RBCs are 10x-50x as concentrated as the microbubbles in the syringe, so about 4 orders of magnitude concentration difference.

      It’s basically changing this from a 0D problem to a 1D problem.

    • Aurornis  3 hours ago

      > From the bubble center plot, I'm guessing that the bubbles are separated on average about a few mm apart?

      The page is vague so I can't tell. I think the images they're showing are actually a composite of many bubbles tracked through the vasculature.

      They say this:

      > As bubbles flow through the vasculature, we accumulate millions of these positions and stack them into a single image with detail finer than the wavelength.

      And the rendering showing the bubble centers they're tracking only shows a few small points moving at a time.

      I think that the amazing animation they produced at the top is actually a composite of many different trackings, not an actual representation of what they capture in real-time.

sheepscreek  5 hours ago

I’m a complete layman to this field, but what the article did say was they’re hopeful that AI/ML can help develop a model that can pull out information such as the scattering caused by RBCs (which is present in the large volume of data gathered by the probe but is too weak to be used for manual techniques) and turn that into meaningful visuals. That’s gonna require a ton of data and that is exactly what they are trying to gather now with what they have built so far.