Selecting a resolution

Choosing an imaging resolution

Choosing the correct imaging resolution is important. If you under-sample you risk losing important information but if you oversample you can quickly generate very large datasets that are unnecessary for the question at hand.

Optical section thickness

Each physical section is imaged at multiple optical planes. For brains, mapping probe tracks and bulk projections is best done at about 20 micron spacing. Going finer than this just generates redundant data because the objects you are tracing are quite large. Counting cell bodies requires a spacing of about 5 microns in Z.

X/Y resolution

Probe tracks can be imaged at about 10 to 4 microns in X/Y (It is even reasonable to trace probe tracks in the Allen 25 micron isotropic space). Bulk projections can be imaged at 8 to 2 microns depending on what you are trying to quantify. If you simply need to know where this there fluorscence, then obviously a lower resolution is just fine. If you need to visually distinguish between fibres of passage and terminations than you will need a higher resolution. Note that in this last case a resolution of, say, 2 by 2 by 20 microns is a good choice: you need good pictures in X/Y but you don't need to sample this very often in Z. For cell body counting a resolution of about 2 microns per pixel works.
X/Y res
Z res
Probe tracks
8 to 4 microns
40 to 50 micron sections, two optical planes
Mapping bulk labeling
8 to 2 microns
40 to 50 micron sections, two optical planes
With 8 kHz scanner average 4 frames if labeling is faint
Cell counting
2 to 2.5 microns
5 to 7 micron optical planes, 40 to 50 micron sections
With an 8 kHz scanner the dwell time is short so if you have sparse/faint signal you might need to average up to 8 frames. For bright labeling, such as strong rabies tracing experiments, you won't need to average frames.
The numbers above assume you want to register data to the atlas. If you know you don't care about this, then it's totally reasonable to take one physical section every 50 microns and no optical planes.
Currently cellfinder requires high resolution in Z as it is designed to find all cells in the brain. In other words, it is not currently designed to cope with the one plane every 50 microns scenario.

Choosing cut thickness

Say you want to image a sample taking planes every 20 microns. That's a little too thin to cut reliably, so you'll have to take thicker sections with multiple optical planes within each section. Cutting thick sections (e.g. 80 microns) with many optical planes will be a little faster but in many tissues the increased scattering from deeper planes will need to a pronounced decrease in resolution and signal strength. As general rule, try to cut as thin as possible to avoid this problem. For PFA-fixed uncleared brains you should be able to cut 40 micron sections at a speed of about 0.35 to 0.5 mm/s.
There is no compelling reason to match your z step size to the resolution of the atlas to which you are registering. i.e. there is no harm acquiring data at a higher z resolution then using a lower resolution atlas since the registration will almost certainly end up tilting the brain in the coronal plane. For instance, you might cut 40 micron sections with a 20 micron spacing between optical planes then register to the 25 micron atlas.
The following image shows optical planes spaced 10 microns apart. Images show autofluorescence from a mouse brain imaged at 920 nm, starting from the very top of the sample (denoted as -30 microns). Here the field of view is large (about 1.6 mm) and the objective exhibits a good deal of field curvature so it takes 40 microns for the whole FOV to fill with sample. We would start imaging about 35 microns below the surface (about where is shown by 0 microns in this image series). Typically we aim for 40 micron sections, which means we'd acquire the images shown at 0, 10, 20, and 30 microns. The last 5 images look substantially fuzzier: the resolution is worse. For this reason we want to cut thin and start imaging as near to the surface as possible. A smaller FOV or an objective with a flatter field would allow us to start imaging higher up and so the whole stack would look better.

Laser power and depth

In the above stack the laser power is being increased with depth to maintain image brightness. Note that this is not sufficient to stop the white matter track becoming darker with the depth. There is nothing you can about that other than cut thinner and start nearer the surface.
You do not need to worry about the images not looking identical in depth. We correct for this in the downsampled image stacks generated by StitchIt in case it might influence image registration. This is shown below:
Orig downsampled stacks
Corrected downsampled stacks
The full-sized stitched images are currently not corrected, but code exists to do this if you need it. So far nobody has asked for this.

Averaging to improve image quality

You may choose to average frames to improve image quality. For resonant scanning, this is achieved by altering the Frame Average number in the ScanImage Image Controls window. You may alter this number once the acquisition started but you have to wait until the sample is being sliced to do so. For linear scanning you should instead alter the pixel dwell time in the ScanImage Configuration Controls. Averaging will not increase the data size but it will slow down the acquisition. Think carefully, therefore, if this is worth it. For example, cellfinder generally copes well without any averaging even if you have a fast resonant scanner. The following image shows sequential physical sections imaged at n=4, n=2, and n=1 frame averaging with an 8 kHz scanner. Pixel size is 2 microns. Neurons are expressing tdTomato. The goal of the experiment is to count somata. The n=4 averaging looks a little smoother but is obviously not needed to count somata.
Effect of averaging a bright tdTomato signal
You might want to average signal if you are trying to see faint sparse fibres. Below are two fields of view showing such data. Each panel shows different values of frame averaging. Data acquired with an 8 kHz resonant scanner.
Two views of faint sparse fibres at different levels of frame averaging
From this we learn that n=16 provides no advantage over n=8. We see that a lot more fine detail is visible compared to n=1 or n=2. However, you would need a mechanism of automatically extracting meaning from the higher level of averaging if it is to be of practical use. Lacking that, you might be better off at n=1 or n=2.
In summary, whether to average and by how much will depend on your question and how you are quantifying features extracted from the images.

Acquisition time

Roughly speaking you can set up four brains at once and image them in under 24 hours at a resolution suitable for tracing bulk projections and electrode tracks. You can image two brains at cell counting resolution in about 18 hours.