CT Image Orientation

Hello,

My earlier "MPR and voxel space" got completely ignored, I'm
shamelessly posting another one, hoping someone will find it
interesting enough to respond to.

1. Like MR...
a. Can a CT series be arbitrarily oriented (via the row and col
direction cosines)? Or, does it always have either all axial or all
coronal or all sagittal slices?

b. Can sets of images *within* a given CT series be arbitrarily
oriented?

2. Does anyone know of any software from any vendor that supports such
3D features as MPR, Volume Rendering, Surface rendering... for
arbitrarily oriented images and series? Essentially, is it
algorithmically even possible to attempt such a thing, and if yes then,
more importantly, is it going to be practically useful to the
radiologist?

Regards,
Harry.

PS: I still believe my earlier post was very much relevant to this
group and none of the queries therein were repeats. So? So, I'm hoping
some will be kind/helpful enough to say at least a few words of advice.

[color=blue]
>
> 1. Like MR...
> a. Can a CT series be arbitrarily oriented (via the row and col
> direction cosines)? Or, does it always have either all axial or all
> coronal or all sagittal slices?[/color]
No, they can be arbitrarily oriented. It is relatively frequent to have
slices inclined from vertical in spine series exams.

[color=blue]
> b. Can sets of images *within* a given CT series be arbitrarily
> oriented?[/color]

Yes they can. Again is is relatively common for a CT series to be
acquired where the slices are oriented perpendicular to the spinal
colum at the point where they intersect the spine - ie the images of
the series change orientation in a varying curve over the series,
[color=blue]
> 2. Does anyone know of any software from any vendor that supports such
> 3D features as MPR, Volume Rendering, Surface rendering... for
> arbitrarily oriented images and series? Essentially, is it
> algorithmically even possible to attempt such a thing, and if yes then,
> more importantly, is it going to be practically useful to the
> radiologist?[/color]

Not off the top of my head. There have been some inquires on the
newsgroup in which people were asking how to take such images load them
into a voxel space. I've certainly had the experience of 3d
visualization vendors complaining about images in a series not being
uniform in their orientation and/or spacing. Clearly it is a much
simpler task to build the voxel space when the slices are uniform; but
is not neccessarily impossible, just a lot harder, when they are not.

I'd put the question to tech support/developers at Volview,
AnalyzeDirect, and Cedera to see what they all had to say. They all
have enough experience in this area, that they more likely than others
to have done it.

[email]eric.goodall@gmail.com[/email] wrote:[color=blue][color=green]
> > 2. Does anyone know of any software from any vendor that supports such
> > 3D features as MPR, Volume Rendering, Surface rendering... for
> > arbitrarily oriented images and series? Essentially, is it
> > algorithmically even possible to attempt such a thing, and if yes then,
> > more importantly, is it going to be practically useful to the
> > radiologist?[/color]
>
> [...] I've certainly had the experience of 3d
> visualization vendors complaining about images in a series not being
> uniform in their orientation and/or spacing. Clearly it is a much
> simpler task to build the voxel space when the slices are uniform; but
> is not neccessarily impossible, just a lot harder, when they are not.[/color]

Yep. Incidentally, when faced with a series that is not uniform, one
common strategy is to split it into "chunks", each of which is uniform.
For example, you might see an abdominal scan with 10 slices at 2mm
separation and 20 more at 1mm and split this into two chunks, each of
which can be visualized.

If the slices are all parallel, a more sophisticated viewer could stich
this back together; though I don't know of one. Unfortunately, I did
encounter a less sophisticated viewer that seemingly ignored the slice
spacing, assumed that all slices were equally spaced, and produced a
very distorted MPR output :-(

Bonjour Harry,

I do have a tool that does 3D reconstruction and (some) volume
rendering from arbitrarily oriented slices. However it does have a
few constraints and (most important!) it is a research tool, not to be
used in clinical applications!

- It will do volume rendering and 3D polygon mesh from the slices if
they are parallel (they do not have to be uniformly spaced).

- If the slices are not parallel, then it will only let you export
contour lines from the desired tissue from each slices, either as line
segments, or point clouds. You can then use a software such as
Geomagic to create surfaces from the point cloud...

Mind you, my volume rendering is not the nicest. I placed more energy
in getting accurate surfaces from which we can get measurements than
on making pretty pictures...

The tool is called sliceOmatic and you can download it from my web
site at TomoVision.com.

Yves

On 3 Nov 2006 13:29:20 -0800, "Steve Robbins" <smr@sumost.ca> wrote:
[color=blue]
>eric.goodall@gmail.com wrote:[color=green][color=darkred]
>> > 2. Does anyone know of any software from any vendor that supports such
>> > 3D features as MPR, Volume Rendering, Surface rendering... for
>> > arbitrarily oriented images and series? Essentially, is it
>> > algorithmically even possible to attempt such a thing, and if yes then,
>> > more importantly, is it going to be practically useful to the
>> > radiologist?[/color]
>>
>> [...] I've certainly had the experience of 3d
>> visualization vendors complaining about images in a series not being
>> uniform in their orientation and/or spacing. Clearly it is a much
>> simpler task to build the voxel space when the slices are uniform; but
>> is not neccessarily impossible, just a lot harder, when they are not.[/color]
>
>Yep. Incidentally, when faced with a series that is not uniform, one
>common strategy is to split it into "chunks", each of which is uniform.
> For example, you might see an abdominal scan with 10 slices at 2mm
>separation and 20 more at 1mm and split this into two chunks, each of
>which can be visualized.
>
>If the slices are all parallel, a more sophisticated viewer could stich
>this back together; though I don't know of one. Unfortunately, I did
>encounter a less sophisticated viewer that seemingly ignored the slice
>spacing, assumed that all slices were equally spaced, and produced a
>very distorted MPR output :-([/color]

Yves Martel wrote:
[color=blue]
> Mind you, my volume rendering is not the nicest. I placed more energy
> in getting accurate surfaces from which we can get measurements than
> on making pretty pictures...[/color]

The precise ray-casting volume rendering with medium-high opacity
setting provides an accuracy of iso-surface representation unachievable
by polygonal substitution. In addition, VR gives subsurface scattering
input which is really makes VR images look so realistic and authentic
to the original object. Video-game driven development is designed to
make artificial emulations look realistic and it is really hard to make
it work with real things like CT data; it is probably the reason of
such common misconception that VR is not good for surface
representation since the practical VR techniques utilized in Video
Cards does not work well with medium-high opacity settings.

Bonjour Stefan,

I agree with you, volume rendering is wonderful for displaying the
data. It make for great diagnostic tools (I guess).

However, you can not place markers on the surfaces (or can you? I may
be out-of-date on this...). My tools are used in body composition and
to create finite element mesh of the anatomical structures. This is
why I am still using the archaic polygons :o)

I meant no disrespect to the Volume rendering crowd! I am (like
everybody else) in awe at what can be done nowadays with a 64 slice CT
and volume rendering (though volume rendering does not seem to work so
well with MR data sets, anybody has comments on this subject?)

Yves

On 9 Nov 2006 10:26:46 -0800, [email]stefanbanev@yahoo.com[/email] wrote:
[color=blue]
>
>Yves Martel wrote:
>[color=green]
>> Mind you, my volume rendering is not the nicest. I placed more energy
>> in getting accurate surfaces from which we can get measurements than
>> on making pretty pictures...[/color]
>
>The precise ray-casting volume rendering with medium-high opacity
>setting provides an accuracy of iso-surface representation unachievable
>by polygonal substitution. In addition, VR gives subsurface scattering
>input which is really makes VR images look so realistic and authentic
>to the original object. Video-game driven development is designed to
>make artificial emulations look realistic and it is really hard to make
>it work with real things like CT data; it is probably the reason of
>such common misconception that VR is not good for surface
>representation since the practical VR techniques utilized in Video
>Cards does not work well with medium-high opacity settings.[/color]

[color=blue]
>Yves Martel wrote:[/color]
[color=blue]
> However, you can not place markers on the surfaces (or can you? I may
> be out-of-date on this...).[/color]

It is quite natural procedure for ray-casting.
There are two ways:
1) Multiple depth maps: each map presents the alpha levels (or ray
energy levels); the blending with polygonal 3D object may be done by
OpenGL means on video card. Depth-maps obtaining is virtually free for
ray-caster (at least for few of them).
2) Shooting of probe ray to the marker location position works fine for
simple markers and it does not require video card involvement. For the
composite marker the procedure is repeated for each marker's point.
The ray-energy level of probe-ray at the marker position is basically a
marker "visibility" if ray crosses the targeted point at all.
[color=blue]
> This is why I am still using the archaic polygons :o)[/color]

It is not archaic it is currently a mainstream of 3D. The mesh is
probably the only feasible way (today) to build dynamic model of some
3D structure since the interactive modification of 3D scalar field is
far beyond of today PC capabilities. However implanting of the static
object through scalar field modification gives the most accurate
representation of the final result.
[color=blue]
> though volume rendering does not seem to work so
> well with MR data sets, anybody has comments on this subject?)[/color]

Volume rendering is not one specific procedure; the only kind of VR
which involves gradients does not work well with MRI due to ill-defined
gradients in MRI data. MIP,Fading-MIP,non-shaded VR - works fine.

[color=blue]
> Bonjour Stefan,
>
> I agree with you, volume rendering is wonderful for displaying the
> data. It make for great diagnostic tools (I guess).
>
> However, you can not place markers on the surfaces (or can you? I may
> be out-of-date on this...). My tools are used in body composition and
> to create finite element mesh of the anatomical structures. This is
> why I am still using the archaic polygons :o)
>
> I meant no disrespect to the Volume rendering crowd! I am (like
> everybody else) in awe at what can be done nowadays with a 64 slice CT
> and volume rendering (though volume rendering does not seem to work so
> well with MR data sets, anybody has comments on this subject?)
>
> Yves
>
> On 9 Nov 2006 10:26:46 -0800, [email]stefanbanev@yahoo.com[/email] wrote:
>[color=green]
> >
> >Yves Martel wrote:
> >[color=darkred]
> >> Mind you, my volume rendering is not the nicest. I placed more energy
> >> in getting accurate surfaces from which we can get measurements than
> >> on making pretty pictures...[/color]
> >
> >The precise ray-casting volume rendering with medium-high opacity
> >setting provides an accuracy of iso-surface representation unachievable
> >by polygonal substitution. In addition, VR gives subsurface scattering
> >input which is really makes VR images look so realistic and authentic
> >to the original object. Video-game driven development is designed to
> >make artificial emulations look realistic and it is really hard to make
> >it work with real things like CT data; it is probably the reason of
> >such common misconception that VR is not good for surface
> >representation since the practical VR techniques utilized in Video
> >Cards does not work well with medium-high opacity settings.[/color][/color]