3d MRI imaging

John Goble goblej at brain.neuro.virginia.edu
Sun Jul 10 14:33:23 EST 1994


The lack of availability comes from the problem size:  a typical 3D MR data 
set consists of 
		    128 "slices"
		      x
                  256x256 pixels/slice
		      x
                  2 bytes per pixel
		  ------------------
		  about 16 Mbytes of data...                     

And that is just the raw data - it doesn't include any representations of the
type of "stuff" in each voxel, etc.  Now add to that a similarly sized data
set from the functional mapping, and you can see why  you need a substantial
piece of hardware to work in this area.  Furthermore, the algorithms for
delineating various tissues in the brain are not for the faint of heart.
 
Since much of the segmentation/image understanding is floating point, rather
than graphics bound, we use HP 9000 series UNIX workstations for this work. 
We use 24bit graphics, but otherwise, the graphics requirements are fairly
mundane.  Polygonal graphics engines are not what you need here.  What you
*do* need are: piles of fp92 specmarks, large, fast disks, and plenty of
memory. Most of our 755's have no less than 192 Mbytes.

We've developed a number of tools for automatic segmentation of the cortical
surface, depth electrode implantations, stereotactic biopsies, volumetric 
analysis of temporal lobectomies, etc.  We also integrate MRA, SPECT, CT, 
etc into MRI when it is required.  
 
This software is in routine clinical use at U.Va., and we've occasionally let
it out the door to people in related research areas if the right sort of
collaboration can be established.  Drop me a note or call me if you'd like
to explore this further.  But don't hold your breath for a PC version...
-----------------------------------------------------------------------------
John Goble                           phone: (804) 243-0221
Neurosurgical Visualization Lab      fax:   (804) 243-0332
U. Va. Neurosurgery





More information about the Neur-sci mailing list