Noise removal

All other questions regarding DCMTK

Moderator: Moderator Team

Message
Author
JL
Posts: 33
Joined: Sun, 2004-11-14, 00:05
Location: US

Noise removal

#1 Post by JL » Fri, 2011-09-16, 06:24

Hello All,
I have a series of studies for a person. Most of the image studies were generated by a GE MG workstation. We had terminated the lease of this workstation from GE early this year and have switched to lease another MG workstation from Hologic.

Yesterday, I noticed that some of the studies generated by the GE MG workstation has a milky color gloss over the entire tissue area (however, the dark area remain dark). Other studies from the same GE system from the same person are OK (i.e. without the milky mask and I can see the tissue clearly).

I think this type of noise (the milky mask over the tissue) is probably caused by some misconfiguration of the MG workstation at the time when the images were taken. I have quite a number of patient studies that experiencing similar behavior.

Have 2 questions:
1/ Does anyone has similar experience with this phenomenon? And how you resolve this problem (i.e. removing the milky mask within the image)? Please noted that we don't have the GE MG workstation (nor the GE Viewing station) anymore.
2/ Is there a tool set in DCMTK library that I can use to correct/filter this noise from the original dcm file?

Thanks in advance.
JL

J. Riesmeier
DCMTK Developer
Posts: 2297
Joined: Tue, 2011-05-03, 14:38
Location: Oldenburg, Germany
Contact:

#2 Post by J. Riesmeier » Fri, 2011-09-16, 16:28

Is there a tool set in DCMTK library that I can use to correct/filter this noise from the original dcm file?
This depends on whether the problem is caused by the pixel data itself or by the other attributes from image pixel module (e.g. the various color/grayscale transformations).

JL
Posts: 33
Joined: Sun, 2004-11-14, 00:05
Location: US

#3 Post by JL » Fri, 2011-09-16, 17:44

[quote="J. Riesmeier"]This depends on whether the problem is caused by the pixel data itself [/quote]

I am not too familiar with this terminology. Would you be able to explain a bit more what you mean "caused by the pixel data itself"? Thanks!!
Do you mean that if the milky mask is created by the MG workstation, then this effect is machine related hence "caused by the pixel data"? Correct?
If this is the case ("cause by pixel data"), can you suggest a way to get rid of this milky mask?

[quote="J. Riesmeier"]by the other attributes from image pixel module (e.g. the various color/grayscale transformations).[/quote]

I exported the image study to a bmp file and did some basis filtering. But still cannot get rid of the milky mask. Is there a tool in DCMTK lib that I can use to get rid of this "milky mask"?

Would it help if you can see the actual "milky mask" on the original study?

Thanks in advance.
John

Chung-Yueh Lien
Posts: 61
Joined: Wed, 2010-05-19, 09:27
Location: Taipei, Taiwan

Re: Noise removal

#4 Post by Chung-Yueh Lien » Sat, 2011-09-17, 16:46

JL wrote: 1/ Does anyone has similar experience with this phenomenon? And how you resolve this problem (i.e. removing the milky mask within the image)? Please noted that we don't have the GE MG workstation (nor the GE Viewing station) anymore.
A single histogram for line-segment cross the region can help you to find the problem where caused from (pixel itself or transformation module). Maybe it is not the real noise.

JL
Posts: 33
Joined: Sun, 2004-11-14, 00:05
Location: US

Re: Noise removal

#5 Post by JL » Sun, 2011-09-18, 05:51

[quote="Chung-Yueh Lien"]
A single histogram for line-segment cross the region can help you to find the problem where caused from (pixel itself or transformation module). Maybe it is not the real noise.[/quote]

Hello Chung-Yueh,

Thanks for the feedback. I used KPAC to exact the images (in bmp format) of the study that has this "Milky mask" over the tissue area. I used all the available filters in Photoshop (including histogram) across the image. I cannot remove this Milky mask. However, when I tired various filtering, I did see some outline of the tissues underneath this milky mask.

Do you have further suggestions? I can email you one of the above bitmap image if you want.

Again, thanks for the feedback and much appreciated.
Regards,
John

Chung-Yueh Lien
Posts: 61
Joined: Wed, 2010-05-19, 09:27
Location: Taipei, Taiwan

Re: Noise removal

#6 Post by Chung-Yueh Lien » Sun, 2011-09-18, 08:06

JL wrote: I used KPAC to exact the images (in bmp format) of the study that has this "Milky mask" over the tissue area. I used all the available filters in Photoshop (including histogram) across the image. I cannot remove this Milky mask. However, when I tired various filtering, I did see some outline of the tissues underneath this milky mask.
Do you have further suggestions?
The point is that images are created by Hologic and then sent to GE Workstation. Perhaps this the main reason why you get the "Milky mask" over the image. That is the meaning of "caused by the pixel data" written by J. Riesmeier.
I guess that GE workstation cannot "adjust" the presentation of image of Hologic. As I knew that Hologic provides 3D mammography and the transformation module could be different than 2D mammography or deal with according to Hologic's private tags that GE Workstation cannot deal with 3D image. You can check the conference statement of modality Hologic to find out which transformations, and dump the DICOM attributes used for display of image.

The other possibility is the reason of simultaneous lighten contrast which is one of phenomenon of human visualization system.

Just few comments! :)

JL
Posts: 33
Joined: Sun, 2004-11-14, 00:05
Location: US

Re: Noise removal

#7 Post by JL » Sun, 2011-09-18, 16:36

[quote="Chung-Yueh Lien"]
The point is that images are created by Hologic and then sent to GE Workstation. Perhaps this the main reason why you get the "Milky mask" over the image. That is the meaning of "caused by the pixel data" written by J. Riesmeier.[/quote]

OK. I got that now. Thank you. Actually, it is the other way around. Most of the prior studies were generated on a GE workstation. The newest study and all current studies are generated from a Hologic workstation.

[quote="Chung-Yueh Lien"]
I guess that GE workstation cannot "adjust" the presentation of image of Hologic. As I knew that Hologic provides 3D mammography and the transformation module could be different than 2D mammography or deal with according to Hologic's private tags that GE Workstation cannot deal with 3D image. [/quote]

But this does not explain why one of the prior studies of this patient (generated by GE) can be viewed very clearly on the Barco viewing station and this same prior study can be view clearly on KPACS at on my laptop.

[quote="Chung-Yueh Lien"]
You can check the conference statement of modality Hologic to find out which transformations, and dump the DICOM attributes used for display of image.
[/quote]

OK. Thank you. So the key conformance statement is from GE not Hologic. For the new Hologic system, I know the viewing station is from Barco (packaged with the Hologic MG workstation). So I have to go through the Barco conformance statement and the Hologic MG workstation conformance statement. The same thing to the previous GE MG workstation and viewing station. Correct?

[quote="Chung-Yueh Lien"]
The other possibility is the reason of simultaneous lighten contrast which is one of phenomenon of human visualization system.
[/quote]

I had tried the dark/light contract (with photoshop) on the bmp image extracted from the milky image. No improvement at all.

If I cannot pin point to what transformation is missing or lacking between the GE and Hologic/Barco, what other methods such as image filtering, DCMTK toolset can I use to tackle this problem (i.e. to remove the milky mask over the tissue)?

Thanks in advance.
Regards,
John

Chung-Yueh Lien
Posts: 61
Joined: Wed, 2010-05-19, 09:27
Location: Taipei, Taiwan

Re: Noise removal

#8 Post by Chung-Yueh Lien » Sun, 2011-09-18, 17:01

JL wrote: But this does not explain why one of the prior studies of this patient (generated by GE) can be viewed very clearly on the Barco viewing station and this same prior study can be view clearly on KPACS at on my laptop.
You can also try to display image by other DICOM Viewers. if all of them are seen clear, this is not a critical problem.
JL wrote: OK. Thank you. So the key conformance statement is from GE not Hologic. For the new Hologic system, I know the viewing station is from Barco (packaged with the Hologic MG workstation). So I have to go through the Barco conformance statement and the Hologic MG workstation conformance statement. The same thing to the previous GE MG workstation and viewing station. Correct?
Correct! More precisely: to check the information related to display image.
JL wrote: I had tried the dark/light contract (with photoshop) on the bmp image extracted from the milky image. No improvement at all.
Try to use the "pixel data", instead of bmp image.
JL wrote: If I cannot pin point to what transformation is missing or lacking between the GE and Hologic/Barco, what other methods such as image filtering, DCMTK toolset can I use to tackle this problem (i.e. to remove the milky mask over the tissue)?
How about dcm2pnm? Some options can be used for display such as:
modality LUT transformation
VOI LUT transformation
presentation LUT transformation
display LUT transformation

:)

JL
Posts: 33
Joined: Sun, 2004-11-14, 00:05
Location: US

Re: Noise removal

#9 Post by JL » Mon, 2011-09-19, 19:21

[quote="Chung-Yueh Lien"]
You can also try to display image by other DICOM Viewers. if all of them are seen clear, this is not a critical problem.
[/quote]
Umm... Sorry to confuse you. In the patient study, there were 5 priors. One generated by Hologic (from last week). The remaining 4 prior studies were generated by GE dated back for a few years. One of the 4 prior studies from GE as well as the latest study from Hologic (for the same patient) has a clear and vivid display of the tissue. I have the same normal clear view on at least 2 other DICOM viewers (KPACS and Digital Jacket) on my laptop (as well as on the Barco viewing station). For the other 4 prior studies (all from GE Senograph DS), they all have a milky mask over the tissue area - no matter which viewer I used (Barco viewing station or viewers on my laptop).

[quote="Chung-Yueh Lien"]
Try to use the "pixel data", instead of bmp image.
[/quote]

Would you be kind enough to tell me how to extract the pixel data from the original dicom image please? If I use dcm2pnm app, to extract the image in pnm format, is this the pixel data you talk about?

[quote="Chung-Yueh Lien"]
How about dcm2pnm? Some options can be used for display such as:
modality LUT transformation
VOI LUT transformation
presentation LUT transformation
display LUT transformation

:)[/quote]

That is good info. Thank you. I will look into the DCMTK lib to see which app allows me to modify these transformations on the image study.

Regards,
John

Chung-Yueh Lien
Posts: 61
Joined: Wed, 2010-05-19, 09:27
Location: Taipei, Taiwan

Re: Noise removal

#10 Post by Chung-Yueh Lien » Tue, 2011-09-20, 08:48

JL wrote:
Chung-Yueh Lien wrote: You can also try to display image by other DICOM Viewers. if all of them are seen clear, this is not a critical problem.
For the other 4 prior studies (all from GE Senograph DS), they all have a milky mask over the tissue area - no matter which viewer I used (Barco viewing station or viewers on my laptop).
Oh! this is a issue of medical imaging process. You may try to choose proper parameters to display image. But those milky masks may not be removed well, if the major problems was caused by the pixel data itself.
JL wrote:
Chung-Yueh Lien wrote: Try to use the "pixel data", instead of bmp image.
Would you be kind enough to tell me how to extract the pixel data from the original dicom image please? If I use dcm2pnm app, to extract the image in pnm format, is this the pixel data you talk about?
You're doing it. The steps is:
1. Get pixeldata from DICOM image.
viewtopic.php?t=356&highlight=findandgetuint16array
2. Choose a transformation module and parameters for presentation.
3. save pixel value as a common image type i.e. bmp, jpg, tiff, etc for display.

Chung-Yueh

Michael Onken
DCMTK Developer
Posts: 1815
Joined: Fri, 2004-11-05, 13:47
Location: Oldenburg, Germany
Contact:

#11 Post by Michael Onken » Tue, 2011-09-20, 13:06

Hi JL,

for the terminology: The pixel data itself is stored in the DICOM attribute Pixel Data. However, before displaying it, a number of transformations can be applied that are (can be) stored within the mammography image. Those transformations map input values from Pixel Data and to new output values which are then used for display on the screen. Thus, the question Jörg asked aims to find out whether the problem is within Pixel Data itself, or comes from the transformations defined within the file.

You might want to expirement enabling and disabling, or even re-computing some of the transformations. In order to do this, you may use the dcm2pnm command coming with DCMTK which permits converting DICOM images to general image formats like BMP. E.g. try

Code: Select all

dcm2pnm --write-bmp yourfile.dcm output.bmp 
in order to get a BMP image.

Now, play with some options of dcm2pnm, i.e. especially try the options given under " modality LUT transformation" and "VOI LUT transformation", e.g.

Code: Select all

dcm2pnm --write-bmp yourfile.dcm output.bmp  --min-max-window-n
. (If the call leads to an error, use additionally option "-d" to find out what goes wrong. Not all options may be applicable to your image.)

Then, look at the output file each time and find out whether the milky impression is gone. If yes, then you have an idea what the problem might be in the image and we can tell you how to solve it.

Good luck,
Michael

JL
Posts: 33
Joined: Sun, 2004-11-14, 00:05
Location: US

#12 Post by JL » Thu, 2011-09-22, 05:59

[quote="Michael Onken"]

Now, play with some [url=http://support.dcmtk.org/docs/dcm2pnm.html]options of dcm2pnm[/url], i.e. especially try the options given under " modality LUT transformation" and "VOI LUT transformation", ...

... Then, look at the output file each time and find out whether the milky impression is gone. If yes, then you have an idea what the problem might be in the image and we can tell you how to solve it.
[/quote]

Hello Michael, Chung-Yueh:
Thanks for your feedback.

I did a series of tests with all the available options for the app dcm2pnm.
I found a unexpected outcome and probably a clue why I have a milky washout appearance on the tissue area.
Overall, I cannot get rid of the milky mask over the tissue. Please note, the black area of the study is pitch black and the milky mask is on the tissue area only.

Unexpected outcome:
Command: +Piv
Observation: I was expecting the color would be inverted. But the image remained the same. As a control experiment. I repeated the same test with a normal dicom file (meaning, nothing was wrong with the image, there was no noise, no contamination etc. on the image) from the same patient on the same date and time (from the same GE MG workstation). The image "color" was inverted - so white became black and vice versa (like a negative).

Intersting observations:
Command: +Pid
Observation: The description for this feature on the document is not clear. I have no idea what "Identity" means. But for the study that I have problem with (the one with milky mask), I had the "negative" conversion effect I was expecting from +Piv. For the control experiment (with the same good image used above), I observered some lightening of tone, particularly on the dark area (non-tissue section). Definitely not the dramatic color inversion.

Command: +Pod
Observation: I think I can see the outline of the tissue very very slightly better than the original bmp image extracted from the dicom file (same amount of miniscule image improvement over the outcome of the bitmaps from all other command prompts for dcm2pnm)

Problems: Some commands (+Dp, +Dn, +Dx, -te, -tb, -ti) gave me error messages and all except +Dp did not generate bmp output file as a result of the errors. I think this has to do with my improper usage of the command line due to the lack of understanding.

Troublesome: I had tested various parameters changes with these commands (+Da, +Di, +O, +Osf, +Ost, +Wh, +Ww). No observable visual differences can be detected among the bmp generated due to input parameter changes. I had not use bit diff check to test the difference between the digital content of these output images with the original.

Possible clue:
I used the command -im to do a screen dump of the image properties.
Observation: The problematic image (the one with milky mask) had a bit depth of 14 and the presentation shape was marked as inverse.
The control image (the one which was taken on the same day, about the same time, on the same machine - used above) had a bit depth of 12 and the presentation shape was marked as Identity.
Based on these observation, I cannot conclude the person who took these images may have mis-configured the GE workstation wrong when she took the problematic image. Good images and bad images from the same patient were taken at the same time. There was probably no time lapse between shots to allow a change of setting on the GE MG workstation.

The following were the result of the problematic image (the one with milky mask) and the control image (the normal image study used above)


Problem Image:
I: transfer syntax : Little Endian Explicit
I: SOP class : DigitalMammographyXRayImageStorageForProcessing
I: SOP instance UID : 1.2.826.0.2.202387.0150650331400592060290100302210520
14146224065
I: columns x rows : 1914 x 2294
I: bits per sample : 14
I: color model : MONOCHROME1
I: pixel aspect ratio : 1.00
I: number of frames : 1
I: VOI LUT function : <default>
I: VOI windows in file : 0
I: VOI LUTs in file : 0
I: presentation shape : INVERSE
I: overlays in file : 0
I: maximum pixel value : 16383
I: minimum pixel value : 626

Control Image:
I: transfer syntax : Little Endian Explicit
I: SOP class : DigitalMammographyXRayImageStorageForPresentation
I: SOP instance UID : 1.2.826.0.2.202387.1441770472332361702221191022100030
11086254033
I: columns x rows : 1914 x 2294
I: bits per sample : 12
I: color model : MONOCHROME2
I: pixel aspect ratio : 1.00
I: number of frames : 1
I: VOI LUT function : SIGMOID
I: VOI windows in file : 4
I: - NORMAL
I: - HARDER
I: - SOFTER
I: - USER
I: VOI LUTs in file : 8
I: - NORMAL
I: - HARDER
I: - SOFTER
I: - USER
I: - NORMAL
I: - HARDER
I: - SOFTER
I: - USER
I: presentation shape : IDENTITY
I: overlays in file : 0
I: maximum pixel value : 3648
I: minimum pixel value : 90


Question: We have discovered this past week that we may have more than 100 similar cases that are stored in our archive. Is there a method, I can remove the milky mask of the problematic image? If there is, what kind of tools, I can use? Is there such a tool set available in DCMTK?

Thanks in advance.

Regards,
John

Michael Onken
DCMTK Developer
Posts: 1815
Joined: Fri, 2004-11-05, 13:47
Location: Oldenburg, Germany
Contact:

#13 Post by Michael Onken » Thu, 2011-09-22, 09:09

Hi John,

thanks for your valuable investigations. Some of the errors popped up since the corresponding options are just not applicable to that specific image. Overall, my impression is that the pixel data itself is a little flawed any maybe your old workstation was able to filter the image to look better (since it knows the images need to be adapted).

If you really need image processing tools for applying different kind of images, then we cannot help you. This is something done by projects like Slicer3D or VTK, or in the future, CTK.

If you want us to look at the image, you may send a sample to dicom at offis dot de.

Michael

J. Riesmeier
DCMTK Developer
Posts: 2297
Joined: Tue, 2011-05-03, 14:38
Location: Oldenburg, Germany
Contact:

#14 Post by J. Riesmeier » Thu, 2011-09-22, 09:15

Maybe, it's too obvious but images of the SOP Class DigitalMammographyXRayImageStorageForProcessing are not meant to be displayed. This is the reason why there is another SOP Class called DigitalMammographyXRayImageStorageForPresentation. The latter (your "Control Image") is displayed correctly as you said.

Michael Onken
DCMTK Developer
Posts: 1815
Joined: Fri, 2004-11-05, 13:47
Location: Oldenburg, Germany
Contact:

#15 Post by Michael Onken » Thu, 2011-09-22, 09:35

Yes, good point :-)

Post Reply

Who is online

Users browsing this forum: Bing [Bot] and 1 guest