Rescale slope and intercept in MR datasets, what are they used for?

[markww]

Hi all,

What is rescale slope and intercept used for in MR datasets?

Is it ok to get rid of these tags and calculate a real
window and level value for the image based on the raw
pixel data?

Thanks,
Mark

[Somchai K]

> What is rescale slope and intercept used for in MR datasets?

Same as they are used in CT and others modalities, to fit the range
of data within the optimal range of pixel value.

> Is it ok to get rid of these tags and calculate a real
> window and level value for the image based on the raw
> pixel data?

It is ok to always apply window and level values directly to
the raw pixel data as long as the range of the data does
not exceed the allocated size (0028,0100).

Just to be on the safe side, you can always set the slope to 1 and offset to
0.
I believe the tags still have to be there.

-Somchai

"markww" <mar...@gmail.com> wrote in message
news:1130342088.9...@g47g2000cwa.googlegroups.com...

[Dan Konigsbach]

For other modalities, rescale slope and intercept are used to map the
stored pixel values from arbitrary numbers into meaningful units, e.g.
Hounsfield units for CT.

For MR, that's not the case, and when they are present, they normalize
the stored values for some other purpose. The reason I'm most familiar
with is in compressed images, where the pixel values had to be scaled
to get best quality compression, and the rescale is used to get back
the original values, thus hiding the change from the rest of the pixel
processing pipeline. (That use has been controversial, but it's out
there.)

Nevertheless, Rescale Slope and Intercept may not be ignored. You
*MUST* apply the rescale before applying the window/level - window
center and width are based on the post-rescale values. Similarly, if
the image has a Modality LUT instead of Rescale Slope and Intercept,
then the Modality LUT must be applied (except for one case, and that's
a separate story.)

Also, note that the Bits Allocated and Bits Stored only affect how
pixel bits are extracted from the Pixel Data, they should *NOT* be
treated as the range of possible values for intermediate results in
the rest of the pixel processing pipeline.

All of this is described in extreme detail in Part 3 of the spec -
check out section C.11 LOOK UP TABLES. (Perhaps others will suggest
other references, too.)

[David Clunie]

Somchai K wrote:
>>What is rescale slope and intercept used for in MR datasets?
>
> Same as they are used in CT and others modalities, to fit the range
> of data within the optimal range of pixel value.
>
>
>>Is it ok to get rid of these tags and calculate a real
>>window and level value for the image based on the raw
>>pixel data?
>
> It is ok to always apply window and level values directly to
> the raw pixel data as long as the range of the data does
> not exceed the allocated size (0028,0100).

No. The grayscale pipeline in DICOM defines that the window
center/width be applied to the output of the rescale
operation, but ...

> Just to be on the safe side, you can always set the slope to 1 and offset to
> 0.
> I believe the tags still have to be there.

No !

The MR IOD does NOT include Rescale Slope and Intercept, and
not should you when creating objects..

So if you are creating MR IODs, do NOT include them. All
receiving applications will then assume an identity rescaling.

Philips was, for a while, including them, which created a problem
for interpretation - should or should they not be used prior
to applying the window center/width operation ?

Typically, renderers of MR images will apply the rescale
values if (undesirably) present before the window center/width
operation, and usually this works out OK.

The PET IOD makes this much worse - the Rescale Slope and
Intercept are present and important, but all known vendor
implementations create the window center/width values with
the intent that they apply to the stored pixel values, not
the rescaled values (the reason being that in PET, the rescale
slope may result in very small fractional numbers, e.g. for
SUV).

So if the IOD is a PET IOD you should check for this and not
apply the rescale values; alternatively, check for rescale
slope being much less than 1.0.

See also previous threads on this subject:

http://groups.google.com/group/comp.protocols.dicom/browse_frm/thread/45920ed2a68973cb/67eebfc8bf194faa?q=mr+rescale&rnum=4&hl=en#67eebfc8bf194faa
http://groups.google.com/group/comp.protocols.dicom/browse_frm/thread/d048cf01456cd165/685be164f030f739?q=mr+rescale&rnum=2&hl=en#685be164f030f739

David

[markww]

Thank you all - my problem is that I am receiving MR datasets with the
scale
and slope values defined! So I do some processing on these datasets and
create
new images, my idea was to just delete these 2 tags (since they're not
defined for MR)
and then just recalculate an appropriate widow/level for my generated
series based
on the raw pixel data values.

I found I can't ignore the scale/slope when presenting the MR datasets
because
they will look too dark without them being taken into consideration. So
this is
where my confusion comes from,

Thanks,
Mark

[Somchai K]

"David Clunie" <dcl...@dclunie.com> wrote in message
news:4360BC9A...@dclunie.com...

>
> No. The grayscale pipeline in DICOM defines that the window
> center/width be applied to the output of the rescale
> operation, but ...
>

Sorry .. my bad. I was thinking about applying slope and offset
directly to the pixel value, which were asked in the question,
but I typed window and level.

> > Just to be on the safe side, you can always set the slope to 1 and
offset to
> > 0.
> > I believe the tags still have to be there.
>
> No !
>
> The MR IOD does NOT include Rescale Slope and Intercept, and
> not should you when creating objects..
>

Thank you for clearing this one up. I had looked them up in the standard,
but
couldn't find them. Thought they could have been in one of the supplements
that I don't know of since many vendors, specially Philips, use those fields
extensively.

-Somchai

[akma...@gmail.com]

Actually i want process my spect image using Matlab.

But my problems is i dont know what is the pixel value represent? For Pet-Ct image i know, the (Pixel Value x Rescale Slop) we can get concentration (Bq/ml).

But in dicominfo for spect, it did not mentioned. either we can use this formula or not? U = m*SV + b. Can you help me? I give u the dicominfo data.

[akma...@gmail.com]