3-D Software for 'Virtual Surgery'

Roland Piquepaille writes "Computer scientists at Brigham Young University (BYU) have developed a new software tool to perform 'virtual surgery'. This tool, dubbed 'Live Surface,' will allow surgeons to visualize in 3-D any part of a patient's anatomy with just a few clicks of a mouse. Similar software already exists, but according to the Deseret Morning News, Live Surface is interactive and fast. This software can be used for better diagnosis by physicians, but it might even suppress the need for some exploratory surgeries. The researchers add that Live Surface might even been used for special-effects in movies or games by extracting an actor's performance from a video clip."

Comments

Re:BYU press release link with more media and info

[grammar fascist]

Can't seem to find any papers or articles about the process, though I noticed it's being patented so there may not be a lot available (?)

Having taken a vision class from Dr. Barrett (CS 750 at BYU), I can fill in some details. I might be able to dig up the paper later. I think you can find it in the latest SIGGRAPH proceedings - dunno if Citeseer has indexed it yet.

It's a segmentation algorithm that works well and fast in 3D images. It uses a graph-cut algorithm to classify voxels as inside or outside whatever you're trying to isolate. You (the doctor) lay down "seed" voxels with a mouse, clicky-clicky, and a few seconds later, the algorithm has isolated the structure. For example, say you want to isolate bone. Hold down the mouse button and move it over the bone. Hold down the other and move it over non-bone. If the algorithm makes a mistake, make some more seed voxels.

This is nothing new so far - the CV folks have been segmenting with graph-cut for ages. The problem is that it's very, very slow - minutes for a single segmentation. Barrett and Armstrong have developed a hierarchical version of the algorithm that uses watershed regions to presegment, and merges them as it runs. Doing graph-cut on large regions is a lot faster than doing it on single voxels. Their stuff is the first interactive speed, seeded 3D segmentation algorithm that produces quality results.

I saw the demo in class. It was really rather impressive, if you're familiar with the subject area.