pouët.net

Volume rendering in radiology: some new videos

category: general [glöplog]
Hi all,

Autumn is a slow month for the demoscene and I've been working on something that many of you with an interest in graphics or medical imaging may find interesting:

http://www.youtube.com/user/ambivu1#p/u

Here you can find some videos of renderings of various medical datasets, including the visible human project and my own made "alien":

http://www.youtube.com/watch?v=WGRu9QEPKhQ

It is an interesting story how this was made, I could elaborate if asked...

In some of the videos you may see types of rendering not seen in any demoscene production that I'm aware of, for example MIP (Maximum intensity projection). Although I think "1995" had one of those.

The website for this software is up and running, and the viewer is/will be free. At the moment we get requests and send to radiologists but the plan is to serve everyone from Jan onwards.
added on the 2009-12-05 10:02:59 by Navis Navis
Navis: looks cool! please elaborate..
added on the 2009-12-05 10:26:26 by magic magic
rendering is cool + really nice that it's free
Go Navis Go!
it's great that you're using your skills to help medicine.
added on the 2009-12-05 11:26:27 by bdk bdk
Looks great!

BB Image
added on the 2009-12-05 12:45:41 by Wade Wade
mm... that's nice, but you really should check this out..

This piece of software blows that out of the water IMO... and its free not just as a viewer, but as a renderer.

http://www.nationalfacility.apac.edu.au/Vizlab/drishti/
(check the google code link for the latest beta versions & beta code)

This is a frame from a fly-through of a bone scanned by CATscan. Rendered by Drishti (open source, free). GPU accelerated realtime Voxel rendering..

BB Image

it can do realtime dissect, clip, voxel warping, transfer function modifications, 1D transfer funcitons, 2D transfer functions, 2D occlusion transfer functions, up to 5 simultaneous models, addresses up to 4gb of GPU ram, full keyframing of all features and interpolation between all scene parameters between keyframes..

this is a vid from version1, version2 is much much better
http://anusf.anu.edu.au/Vizlab/drishti/Movies/torso.mov

windows, mac, linux etc.etc.etc.

check it out :) :)
added on the 2009-12-05 13:03:20 by baldrick baldrick
Thanks baldrick,

very interesting images! great quality, I watched the videos and had a go myself using the software (is it yours?) with my own medical volumes.


I'm aware of all those techniques, and internally there is support for pretty much everything (or is possible to do so in little time). But I think chasing rendering perfection is not in my direct priorities. Decent framerate and interactivity is. I follow what radiologists prefer and streamline the software to that end.

So it is hard to compare the two applications. Ambivu was never designed to be a generic high quality 3D renderer, just like Drishti could never work in a clinical environment (it would require a complete rewrite). But I can see that they have overlapping functionality.


Magic: I'll say the story of the alien at a later post.
added on the 2009-12-05 14:41:20 by Navis Navis
the alien mesh is based on amusic on a bad hair day? ;)
added on the 2009-12-05 15:03:59 by maali maali
Ehm no, it is actually a warped version of a full body scan. Problem is, how do you do it (we are talking about 2x 600+ slices of 512x512 16 bit data).

I did it for fun last year. The sequence was:

1) Write a program to export the dicom slices of a full body patient into jpegs.

2) Make a 3D model of an alien (my good wings3d!)

3) Make a program that "slices" the 3D model of the alien into jpegs, keeping the outline of the 3D mesh.

4) Get somebody to "morph" the CT/PET jpgs (with the bones and all that) into the alien jpegs. This somebody must have ALOT of patience (Thanks Hugh!).

5) Render all back into Dicom format

it was fun. Obviously it can be done better, but back then we had no way of testing the result. So for a completely blind process it is ok!
added on the 2009-12-05 15:42:12 by Navis Navis
that "morphing" can probably be done with the liquify brush? :)
added on the 2009-12-05 15:43:27 by maali maali
Quote:
Dr Kostas Pataridis, Head of R&D: holds a PhD in clinical medicine from the University of Oxford

o_O
added on the 2009-12-05 15:44:33 by Gargaj Gargaj
nah, I learned the trade here in demoscene "the pavement university" as we say in Greece (hope it makes sense).

Mahli: I think the guy used a program on a Mac that would let him morph and export as a .mov. I'm not quite sure what it was called.. Ah I found something

http://hughpryor.blogspot.com/2009/06/medstamp-alien.html

aftereffects
added on the 2009-12-05 15:47:44 by Navis Navis
Quote:
This piece of software blows that out of the water IMO... and its free not just as a viewer, but as a renderer.


how does a renderer differ from a viewer in this case? in most cases, I'd say the two are interchangeable for the same thing..
Cool.

If you ever need a full size Visible Human Male dataset, at 1 cubic milimeter per voxel, I, let say, *know* somebody that have it.
added on the 2009-12-05 16:28:13 by texel texel
heh thanks, I have both male and female full dataset (CT photos etc), all legally. But I think you can get the head online pretty easily.
added on the 2009-12-05 16:39:02 by Navis Navis
Navis, does it is difficult to get a license for the datasets? As far as I know, you need a reason to ask it, and I suppose "I want to play a bit with it" is not a good one.
added on the 2009-12-05 16:52:18 by texel texel
I got the data through university, somebody else did all the paperwork. But that was 5-6 years ago, so it must have changed since then.

have a look at this:
https://mri.radiology.uiowa.edu//VHDicom/
added on the 2009-12-05 16:54:46 by Navis Navis
Great. But those looks as being only the CT scans... I remember years ago, you had to ask for a license for the color images, and it was a very big dataset, I suppose gigabytes of data, they used backup tapes for it... maybe there were not yet DVD-W in that time...
added on the 2009-12-05 16:58:09 by texel texel
Also, I would like to try again to render these with my GTS250. I suppose that even a simple raycasting should be very fast with the GPU, isn't it?

I used to try software rendering for it years ago, but it was slow even for little datasets... afaik, the memory speed was the bottleneck. But probably in the current computers, even in software rendering by cpu it might be fast enough for middle size datasets...
added on the 2009-12-05 17:01:52 by texel texel
Oh, I've just remembered something:

If you try aditive raycasting with all the voxels, it doesn't look very good, it looks just too transparent, but if you use something like "density" it looks much better.

For example, when the color intensity is higher than N, you sum only the next 50 traversed voxels higher than N. It looked great that way.
added on the 2009-12-05 17:05:25 by texel texel
Navis, does the color photos dataset and the CT scan does perfectly overlap? Could you use the CT data for the density?
added on the 2009-12-05 17:06:35 by texel texel
thanks for all this info.

I'm not sure they do overlap, in theory they should. I remember long time ago that I was carrying these color scans on big disks (not dvds, but still optical) that were read on some scary machines at my local hospital. I have the data still somewhere there, but haven't touched them for ages.
I do stuff with the CT dataset but not with the cryo-photographed one.

I'll think about your comment on MIP...
added on the 2009-12-05 17:24:47 by Navis Navis
Quote:
o_O

Gargaj, I guess it makes a difference from the usual demosceners, who when they 'grow up' often become computer / math / engineering types. Although of course Navis still is - just in a different (and interesting) field.
added on the 2009-12-05 21:07:24 by t-zero t-zero
I'm an engineer and despite the title I know jackshit about medicine... For your demoscene medical problems you are better off consulting Adok :-)
added on the 2009-12-05 21:48:21 by Navis Navis
That looks really good Navis. Great rendering, and the application looks really usable.

I'm wondering something though: would it be possible to simulate an operation with this? i.e. mark an incision on the skin, then have the software discard the area around the incision to show the layers beneath? That could be pretty useful I suspect.
added on the 2009-12-06 01:19:44 by psonice psonice

login