How many cubes has the dataset?

#1

Hi HQ,

crazyman4865 and I tried to figure out how many cubes there are in your dataset. That is the information we have:
  • dataset is 350×300×60 µm3 large (micrometers)
  • a single voxel is 16.5×16.5×23 nm3
  • a cube consists of 256x256x256 voxel
  • cubes overlap each other about 32 pixel
So… calculating:

(1/224 * 350 µm / 16.5 nm) * (1/224 * 300 µm / 16.5 nm) * (1/224 * 60 µm / 23 nm) = 89500
Actually a bit less since you don’t have “half cubes”, I guess.

But 89,500 cubes seems to be a bit low… I think I already saw something like “volumeID” > 100,000. (Not to be mistaken with the cubeID you can see in the debug-window! Multiple cubeIDs can share the same cube volume, so there have to be a lot more cubeIDs than 89500…)
#2

For any particular neuron traced so far the cube numbers have been sequential but of course the volume of space occupied by that neuron is very small: therefore the unoccupied cubes between dendrites of the neuron are not (yet) numbered; only those cubes with a path traced through them are numbered.

 

I can't remember what the total cubes are per neuron (1200?) and we've traced about 10 neurons I guess so there would be a year or twos worth of life in the data set yet. Actually more than that -  I guess that on average there are a least 100 dendrites passing through each cube so each cube would have to be traced 100 times to get the full picture.

#3

The current volume count in the DB is 95828.  However, we list channel and segmentation as separate volumes so that would be 2 for each cube.  That leaves about 48k.  That’s a rough estimate, but it’s probably about right.  I don’t see anything all that wrong with your calculations though.  Remember that we only have cubes around the cells that we are studying.  We haven’t created all the cubes especially around the edges of the volume.  We continue to add cubes as needed and filling out the full volume.

#4

@matt so what would be the final number, if there were cubes for the entire volume? Is @nkem right? This is a good stat!



#5

Like I said, I don’t see any problem with the math.  Would be worth double checking the numbers w/Jinseop.