Are Voxels constructed after image acquision through a 3D-rendering algorithm such a marching cubes? Or are they obtained by direct 3D-image acquision?
If both are possible: What is the current method usually used for obtaining 3D images from MRI scans?
How are Voxels generated from 2D MRI scans?
Generating voxels from MRI Scans was actually the original motivation for marching cubes. The basic concept is to think of the MRI scans as slices. If these slices lie in the XY plane at different Z values, the pixels of the scans make up the nodes of the cubes. Figure 2 in the white paper shows how this is used to construct the cubes. http://academy.cba.mit.edu/classes/scanning_printing/MarchingCubes.pdf
The lightness/darkness of the pixels is then used to decide whether the node is considered "on" or "off". With a cube of 8 nodes and each node can be either 0 or 1 then there are 2^8 = 256 possible configurations. Through symmetry the authors show that there are really only 15 unique cases that need to be considered.
However, I would point you to this white paper as well that shows that the 15 original cases can actually be handled in different ways leading to 33 different cases: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.56.7139&rep=rep1&type=pdf
Now on a personal note: having implemented both of these algorithms myself, if you are looking for an algorithm to reconstruct 3d surfaces I would strongly recommend using marching tetrahedrons as the number of cases is much lower and much easier to implement.