Dicom files: how to extract brain/organ from MRI scan?
This is technically not a graphic design question, but it is technically not not a graphic design question either, so here I go.
I have a set of MRI scans of my head. They are 5mm slices in x,y,z (or if you prefer med-speak: saggital, coronal, axial). (I also have some sets in 1.5mm slices, but they are incomplete. I.e. only a section, not the entire brain.). My ultimate goal is to extract the brain only, and have it 3D printed. I am using Osirix (though I have Slicer, but have not tried it yet).
I could hand-draw the outline of the brain in each individual slice, but apart from taking weeks to do properly, I am not sure the resulting slices will add up to anything near good enough (If I knew it would turn out good, I would consider it). But there have to be a better way.
I have tried to make a 3D model of the entire head using 3D surface rendering, and the result even of that is rather disappointing:
I have also tried using the "growing" tool, where, as far as I understand, I should be able to select a colour/density and then removing everything else.
As I see it, there are two ideas, but they both boil down to the same: How to select one organ, when the other organs and tissue close to it are much of the same density? (the inverse approach would be; instead of extracting something spesific, then delete what is not needed).
I realise that if I am ever successful in extracting the brain, there will be the voxel lego-effect (albeit in x, y, z), and I would also need to set some smoothing parameters.
So. In short:
- What settings (for contrast/density/tissue type/bone) do I set, to make the extraction possible?
- Do I use the "grow" tool, and if so, how do I get it to select all voxels in all three series?
If not the grow tool... then what?
If I get it extracted, where and how do I determine smooting options?
The segmentation provides the set of the data with the needed density. Without a segmentation you can use built-in volumetric filters and obtain a rough reconstruction. This is possible in ImageJ too.
Be aware that segmentation is a slow technique and needs a certain time to be properly used.