3 ms·

Their resolution is 60 x 60 x 90 nm which is far below the 20 um that BigBrain used in 2013 for their whole-brain scan [1]. Especially, axons are 0.2-20 um in

by est31 8y ago

Their resolution is 60 x 60 x 90 nm which is far below the 20 um that BigBrain used in 2013 for their whole-brain scan [1]. Especially, axons are 0.2-20 um in diameter. With a resolution of 20 um you won't really find them, but with 0.01 um, you will certainly be able to identify them. This could lead to new anatomical insights in human brains. I remember recently someone complaining on hn that a new finding about a connection of two brain regions in mice hasn't been confirmed in humans yet [2]. If scans of entire human brains at 0.01 um resolutions were available, you could verify claims like these by just checking the data. Edit: corrected the BigBrain resolution. [1]: https://en.wikipedia.org/wiki/BigBrain https://en.wikipedia.org/wiki/BigBrain [2]: https://news.ycombinator.com/item?id=18943385 https://news.ycombinator.com/item?id=18943385

SubiculumCode 8y ago
The Wikipedia link for BigBrain you provided lists the resolution at 20 µm isotropic 3D voxels. That is ~2.71 µm in each direction.
- est31 8y ago
  Nope. Yes, the Wikipedia link says 20 um isotropic resolution, but that means 20 um per direction. 20 um is a unit of length, while 20 um ^ 3, a unit of volume. If it were about volume, you'd be correct. Also, I just looked at the paper [1] and here it says: > To generate a data set with isotropic resolution, we down-scaled all images to 20 µm by 20 µm to match the section thickness of 20 µm. So that confirms the 20 um per direction, no? Edit: oh no, I just see that I wrote 1 um above, not 20. Not that 2.71 is correct but I guess that's what confused you. Corrected. Thanks! [1]: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.899.2730&rep=rep1&type=pdf http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.899...
  - SubiculumCode 8y ago
    <slaps forehead> You are right about the resolution. Apparently I did not have my coffee because I work with voxels all the time.
    - est31 8y ago
      It's all good :)
nanometer 8y ago
You shouldn’t mix units like that. Everything should be experessed in nanometers only, and avoid micrometers, since expressing nanometer sizes in micrometer units requires confusing decimal points, and an extra cognitive load to translate fractional proportions. Axons are 200 to 2000 nanometers in diameter. Much easier to read and compare to the nanometer resolutions defined in the first paragraph. Anyway, we’d need thousands of desireable human brains to draw conclusions. And my desireable, I mean smart, sociable people mostly. You’ll want total morons in the data set too, but only to distinguish total assholes in the training data, when we use AI to rationalize their extermination. That way, the Hitlers go to the concentration camp, this time.

by est31 8y ago

SubiculumCode 8y ago

The Wikipedia link for BigBrain you provided lists the resolution at 20 µm isotropic 3D voxels. That is ~2.71 µm in each direction.

est31 8y ago
Nope. Yes, the Wikipedia link says 20 um isotropic resolution, but that means 20 um per direction. 20 um is a unit of length, while 20 um ^ 3, a unit of volume. If it were about volume, you'd be correct. Also, I just looked at the paper [1] and here it says: > To generate a data set with isotropic resolution, we down-scaled all images to 20 µm by 20 µm to match the section thickness of 20 µm. So that confirms the 20 um per direction, no? Edit: oh no, I just see that I wrote 1 um above, not 20. Not that 2.71 is correct but I guess that's what confused you. Corrected. Thanks! [1]: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.899.2730&rep=rep1&type=pdf http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.899...
- SubiculumCode 8y ago
  <slaps forehead> You are right about the resolution. Apparently I did not have my coffee because I work with voxels all the time.
  - est31 8y ago
    It's all good :)

nanometer 8y ago

You shouldn’t mix units like that. Everything should be experessed in nanometers only, and avoid micrometers, since expressing nanometer sizes in micrometer units requires confusing decimal points, and an extra cognitive load to translate fractional proportions. Axons are 200 to 2000 nanometers in diameter. Much easier to read and compare to the nanometer resolutions defined in the first paragraph. Anyway, we’d need thousands of desireable human brains to draw conclusions. And my desireable, I mean smart, sociable people mostly. You’ll want total morons in the data set too, but only to distinguish total assholes in the training data, when we use AI to rationalize their extermination. That way, the Hitlers go to the concentration camp, this time.