Karan Bajaj wrote:

> My dilemma is that if a CR image contains both the tags for pixel
> spacing and imager pixel spacing, and they are not the same value,
what
> should I use.

Quite a dilemma ! I can't see any reason why they would be
different, unless someone has manually "calibrated" the image
using some software tool. A question you will need to ask the
vendor. Who is it, by the way ?

I did a quick survey of CR and DX images lying around, and in almost
every case (Fuji, Agfa, Siemens, GE, Canon, Philips, Kodak, SwissRay),
if both attributes were present, they had the same values.

There were a handful of exceptions, and they were all Fuji, which
had different values for Imager Pixel Spacing and Pixel Spacing,
though not always - indeed most Fuji images had only Pixel Spacing
and not Imager Pixel Spacing. I could not narrow this down to any
particular model number.

I will mention this to someone at Fuji to see if I can get an
explanation.

David

I think they could be different since Pixel Spacing is related to the
physical distance in the patient while Imager Pixel Spacing is related
to the physical distance on the detector housing.
In case of an optical magnification (object not in contact with the
detector) they could be very different.

In my opinion for measurements you should use Pixel Spacing (since it
should be related to the patient) and then, if it is not present,
Imager Pixel Spacing.

Valerio

Valerio Salomoni wrote:
> I think they could be different since Pixel Spacing is related to the
> physical distance in the patient while Imager Pixel Spacing is
related
> to the physical distance on the detector housing.
> In case of an optical magnification (object not in contact with the
> detector) they could be very different.
>
> In my opinion for measurements you should use Pixel Spacing (since it
> should be related to the patient) and then, if it is not present,
> Imager Pixel Spacing.

No, no, no.

The presence of Pixel Spacing in vendor's implementations is a
historical accident.

There is no Pixel Spacing attribute in the DICOM CR IOD.

Because there was nothing present in the standard in 1993,
vendors took it upon themselves to extend the standard in
their implementations, and reused an attribute that was
intended only for cross-sectional modalities like CT and
MR, which do know how big a pixel is inside the patient.

Projection radiographic modalities with a diverging X-ray
beam are subject to geometric magnification, therefore
a device cannot know how big a pixel is inside the patient
because it changes depending on the depth within the patient
along the course of the X-ray beam.

Therefore the standard was extended to add Imager Pixel
Spacing to the CR object, with a definition that is
appropriate to the mechanism of the modality. This was
done in CP 66
(ftp://medical.nema.org/medical/dicom/final/cp066_ft.pdf).

However, some vendors have been slower than others in terms
of getting with the program and adopting this now standard
but optional attribute and not sending Pixel Spacing as a
poor man's substitute. This includes PACS vendors who do
not look for Imager Pixel Spacing and use it in preference
to the Pixel Spacing that they learned from experience was
sometimes present.

Regardless, if Pixel Spacing is present in a CR image it is
non-standard, and its meaning likely has nothing to do with
"physical distance in the patient" whatsoever, regardless
of the words used to define it in CT and MR.

It remains to be determined what the particular vendor who
is sending both attributes with different values intends
this to mean; experience would suggest that this is a bug.
I am making inquiries.

David

I have some images where they are different and clearly
erroneous/problem images. The images came from a CR vendor stitching
software where two different original CR images have been combined into
a composite image for procedures like a Scoliosis survey or hip to
ankle study.

It appears the authors of the software convinced themselves that they
didn't need to update imager pixel spacing attribute in the combined
image, since it wasn't generated directly from a physical phosphor
plate. In the resulting image, the imager pixel spacing contained the
value "inherited" from one of the source images whereas the pixel
spacing was based on the characteristics of the new derived images.

Not all other vendor software packages receiving these images used the
correct value, causing distortion in the displayed image (change in
aspect ratio).

Turns out the problem was known and had been corrected by the CR
vendor, but the facility using the CR system either had not received
the notice of the problem with the software, or had failed to act on
getting their system updated until the orthopaedic surgeons started
complaining about the way the images displayed on the new PACS
workstations (the CR vendor workstations displayed the images
correctly).

dclunie@dclunie.com wrote:
> Karan Bajaj wrote:
>
> > My dilemma is that if a CR image contains both the tags for pixel
> > spacing and imager pixel spacing, and they are not the same value,
> what
> > should I use.
>
> Quite a dilemma ! I can't see any reason why they would be
> different, unless someone has manually "calibrated" the image
> using some software tool. A question you will need to ask the
> vendor. Who is it, by the way ?
>
> I did a quick survey of CR and DX images lying around, and in almost
> every case (Fuji, Agfa, Siemens, GE, Canon, Philips, Kodak,
SwissRay),
> if both attributes were present, they had the same values.
>
> There were a handful of exceptions, and they were all Fuji, which
> had different values for Imager Pixel Spacing and Pixel Spacing,
> though not always - indeed most Fuji images had only Pixel Spacing
> and not Imager Pixel Spacing. I could not narrow this down to any
> particular model number.
>
> I will mention this to someone at Fuji to see if I can get an
> explanation.
>
> David

You are right, David.
I made a mistake when looking to the Image Plane module that is
included in CT but not included in CR.
I'm sorry if I created further confusion on this topic.

Valerio

dclu...@dclunie.com wrote:
> No, no, no.
>
> The presence of Pixel Spacing in vendor's implementations is a
> historical accident.
>
> There is no Pixel Spacing attribute in the DICOM CR IOD.
>
> Because there was nothing present in the standard in 1993,
> vendors took it upon themselves to extend the standard in
> their implementations, and reused an attribute that was
> intended only for cross-sectional modalities like CT and
> MR, which do know how big a pixel is inside the patient.
>
> Projection radiographic modalities with a diverging X-ray
> beam are subject to geometric magnification, therefore
> a device cannot know how big a pixel is inside the patient
> because it changes depending on the depth within the patient
> along the course of the X-ray beam.
>
> Therefore the standard was extended to add Imager Pixel
> Spacing to the CR object, with a definition that is
> appropriate to the mechanism of the modality. This was
> done in CP 66
> (ftp://medical.nema.org/medical/dicom/final/cp066_ft.pdf).
>
> However, some vendors have been slower than others in terms
> of getting with the program and adopting this now standard
> but optional attribute and not sending Pixel Spacing as a
> poor man's substitute. This includes PACS vendors who do
> not look for Imager Pixel Spacing and use it in preference
> to the Pixel Spacing that they learned from experience was
> sometimes present.
>
> Regardless, if Pixel Spacing is present in a CR image it is
> non-standard, and its meaning likely has nothing to do with
> "physical distance in the patient" whatsoever, regardless
> of the words used to define it in CT and MR.
>
> It remains to be determined what the particular vendor who
> is sending both attributes with different values intends
> this to mean; experience would suggest that this is a bug.
> I am making inquiries.
>
> David

eric.good...@gmail.com wrote:

> I have some images where they are different and clearly
> erroneous/problem images. The images came from a CR vendor stitching
> software where two different original CR images have been combined
into
> a composite image for procedures like a Scoliosis survey or hip to
> ankle study.

The Fuji examples I have are not stitched, they are ordinary chest
x-rays
performed on the normal 14 by 17 inch (35 by 43 cm) CR casettes, which
with a pixel matrix of 1760 by 2140 corresponds to a spacing of about
0.2 mm; the value in Pixel Spacing (0028,0030) is 0.20\0.20, whereas
in Imager Pixel Spacing (0018,1164) it is 0.10\0.10, clearly wrong.

So there seems to be a greater problem in the field than just with
psot-processing stitching applications.

David

Thank you for replying David.

We are seeing the issue with a few more modalities, particularly a
Kodak DR(it might be the 850), the Philips Digital Diagnost and Easy
Vision Rad from Philips.
In the easy rad conformance statement, it clearly mentions for the CR
image pixel module.
"Pixel Spacing 0028, 0030 Contains pixel size in patient if image is
calibrated, contains pixel size in detector plane otherwise."

I guess then, if some does figure out calibrated Pixel spacing, where
could they enter this information? and if there any provisions in the
standard to use this information.

Karan

dclunie@dclunie.com wrote:
>
> The Fuji examples I have are not stitched, they are ordinary chest
> x-rays
> performed on the normal 14 by 17 inch (35 by 43 cm) CR casettes, which
> with a pixel matrix of 1760 by 2140 corresponds to a spacing of about
> 0.2 mm; the value in Pixel Spacing (0028,0030) is 0.20\0.20, whereas
> in Imager Pixel Spacing (0018,1164) it is 0.10\0.10, clearly wrong.

A little further information on this, based on some preliminary
information from Fuji.

Apparently someone in Fuji decided in a particular software revision
that "Imager Pixel Spacing would be used to state the pixel spacing
values that represent the acquired resolution of the reader, not
necessarily the composition of the output image."

It is a shame that they did not use the appropriate attribute from
the DX object, Detector Element Spacing (0018,7022) for this purpose.

The implication is that DICOM compliant workstation software that
uses Imager Pixel Spacing (0018,1164) to measure things will give an
incorrect result in some cases.

Apparently, "Fuji FSEs have the ability to map DICOM attributes to others
using a configuration file, and they have access to an internal Integration
Bulletin that explains how to address the disparity between the two
values by setting them both to the Pixel Spacing value (the output image
pixel spacing)."

I suggest that everyone with a Fuji CR check to see if this problem
is apparent at their site, and to have their FSE make the necessary fix
if it is. You may not be seeing the problem because your software may
already just use the non-standard Pixel Spacing, or you may not measure
things, but other sites that receive your CR images on CD may have this
problem.

David

Karan Bajaj wrote:

> We are seeing the issue with a few more modalities, particularly a
> Kodak DR(it might be the 850), the Philips Digital Diagnost and Easy
> Vision Rad from Philips.
> In the easy rad conformance statement, it clearly mentions for the CR
> image pixel module.
> "Pixel Spacing 0028, 0030 Contains pixel size in patient if image is
> calibrated, contains pixel size in detector plane otherwise."

Can you send me a dump of the relevant DICOM attributes that include
Pixel Spacing, Imager Pixel Spacing and the manufacturer/model
information so that I can follow up on this with those vendors as
well ?

> I guess then, if some does figure out calibrated Pixel spacing, where
> could they enter this information? and if there any provisions in the
> standard to use this information.

The issue of encoding calibrated pixel size information is non-trivial
in a general sense, if one wants to define where within the subject
the calibrated pixel spacing is accurate.

WG 2 is doing some work on this in the new Enhanced XA supplement 83,
and you might want to look, for example, at:

"Object Pixel Spacing in Center of Beam (0018,x404)

Physical distance within the anatomic region of interest in the center of
the beam and perpendicular to the beam between the center of each pixel,
specified by a numeric pair adjacent row spacing (delimiter) adjacent
column spacing in mm."

but if you look in the supplement, there is a lot more to this.

Clearly, it would be possible to retrofit some of these concepts back
to CR, DX and scanned film in secondary capture objects.

One question is whether to derive values from known beam geometry and
assumptions about the location of region of interest within the subject,
or to do this by measuring an object of known size and computing or
entering a calibration correction.

In the latter case, would one want to re-save an entire image pixel
data set, that was presumably created by the acquisition device prior
to a human operator identifying and measuring a known sized object ?

Perhaps the non-geometrically derived scenario would be better handled
by a "calibration SOP class" in DICOM, which like a presentation state
or real world value mapping, encoded information about the image
that was determine retrospectively.

Indeed, one could envisage that there might be different regions at
different depths in the subject, whose calibration would be different
even though they were in the same image. E.g. a ruler on the front
and back surfaces of a very obese subject with a relatively short
source to image receptor distance and source to object distance
would give quite different calibration results.

David