View Full Version : Building brain from MRI DATA

Walter Pollard
10-29-2004, 01:59 PM

I was wondering if anyone had any experience building3d models from MRI data.

I am beginning to work on a project where I need to model the brain, and skull from of a human head. I currently have numerous slices of MRI images, and was wondering the best way to tackle this.

Is there a way to automatically build geometry from grayscale levels in the MRI images? How about alpha channels?

Would one build the brain model in slices, or build one exterior volume than cut into slices to model interior?


10-29-2004, 03:40 PM
i dont think its availble yet but i think there was a plugin that could do this or somethign similar - perhaps the guy making it would let you beta test it or somethign if you asked nicely

otherwise no idea =/

hmmmmm unfortunatly i cant even remember the guys forum handle - but perhaps someone else will know

10-29-2004, 10:46 PM
What format does the data come in? You could use image slices in modeler as templates to build cross section splines from so that you can patch them or skin/bridge them.

10-30-2004, 01:35 AM
Like most real anatomy the internal structure of the brain is too complex to accurately model from templates or primitives with any degree of realism. This, however, looks incredibly promising


The plugin promises to take raw DICOMM (or other formats) and build your model !



10-30-2004, 07:33 AM
Yah mon,
Elmar Moezler's company. He's around the periphery of LW dev, or deeper, not sure. But it has been said that we may be graced with his genius in future LW incarnations.

Walter, get in touch with him because it is a commercial product or planned to be. Chuckle, you may help to set the price for it. Tell us what comes of it.

10-30-2004, 10:35 AM
Elmars on the Dev team. That plugin is pretty cool. I saw it awhile ago. :)

10-31-2004, 01:20 AM
Couldn't agree more. The plugin is more than cool it has implications for NewTek that I'm sure Elmar is aware of. For the rest of us, imagine the benefits to NT of penetration into medical imaging, research and education. Current 3D medical imaging uses mostly extremely expensive dedicated scanners (MRI, CT, ultrasound). The computing is usually expensive unix based workstations which most of us medics can't relate to as we are arrested at the windows stage of development. With the forthcoming 64 bit boxes and OS's we could see this on our desktops. (tiger for me when some philanthropist donates a G5). Think of firing up modeler with an image sequence of an MRI and then virtually probing the structure of the organ. Then there is virtual surgery for training (or rehersal prior to the real thing).

Meanwhile back at Walter's problem. You can buy models of the brain. I have attached a quick and dirty render of one of viewpoints models (viewpoint now taken over by digimation). This was using VP18976 and despite the exorbitant cost of the model and the high poly count you can see from the split brain that the internal stucture is extremely limited in its detail. Now I suppose that you could get around that by uv mapping scanned images from standard anatomy textbooks but it depends on your project (and you run into copyright problems).

Here's hoping the plugin is not too expensive when it comes. i can't even aford G2.


Elmar Moelzer
11-01-2004, 03:27 PM
Hello everyone!
Volumedic is still under development. Currently we are very busy working on LightWave- development, so Volumedic has to wait a little. I am sorry.

11-03-2004, 12:13 PM
That's even better! We can wait for volumedic...

Oh, here is an idea for LW dev. Be able to use the exsisting gradient routines and tie them into other things... One being Light Falloff/Intensity.


11-03-2004, 07:36 PM
yeah, I would start off in the same direction - greyscale and alpha channels, if you could get the MRI slices to convert well into alpha masks...

Apply every nth MRI slice as a transparancy texture to a block of polys layered up like leaves in a book, jiggle them with a bit of motion blur, could look good.

There was two threads here that might be relevant:
Clip maps are funny to play with

and a useful script by Matt that names surfaces sequentially: