How can I make compelling 3D images and videos from SBF-SEM stacks?
There are two types of 3D images:
There are volume-based views of the stack (cube-like views) which you can make in ImageJ 3DViewer or Amira simply using the stack.
There are also mesh-based views in which you separate out the things you care about by manual tracing or automated segmentation methods. You can make a simple mesh by tracing your object of interest in a program like Reconstruct.exe (free) or using surfaces generated from the ImageJ/FIJI 3DViewer. ImageJ and Reconstruct can provide quite powerful 3D views that you can export immediatelly. Meshes can also be imported into Blender, Amira or Imaris. (see links coming)
How can I do analysis and obtain statistical data?
For some questions, there are already automated and semi-automated (expedited) approaches for annotating the dataset and extracting measurements. ImageJ will measure areas and lengths, for example. If you are counting objects, this can be as simple as taking the counts into a spreadsheet. Where there are a lot of image sets, however, a little programing (or collaboration) allows you to greatly expedite numerical data extraction. Using ImageJ scripts, R (or prism) for data assembly, statistical treatment and graphing.
Can I use deep learning and machine learning to analyse my data?
Absolutely. It takes a fair bit of data for training as system yourself. As time goes by, however, people are providing solutions that you can take off the shelf and use. Obviously, though, collaborating with someone that knows keras/tensorflow or can help generate the original training data by tracing things is worthwhile. (link to posts on the topic coming).