Issues with J2K for 16-bit CT/MR

    There seems to be an issue compressing 16-bit CT and MRI images, at least from GE equipment where I have tested 5 different machines and Toshiba CT where I have tested two equipment.


    The outcome of subjecting the images to J2K lossy compression is images with artefacts that appear like boxes of different sizes. It does not matter the quality factor.


    I have attached before and after screen-shots of the images as seen on a viewer.


    I have also attached before and after dicom images themselves (anonymized)


    Is there a known issue with conquest in this regard?


    Thanks.

    I have been looking at the code for the JasPer library interface (used to convert to J2K). First, JasPer does not use a percent, but instead a compression ratio with 1 being a one to one and .01 producing a 100:1 compressed file size. Jasper also changes modes when from lossless (int) to lossy (real). With this said, if you run the lossy compression with a debug mode of 3, a line in the log file should say "JPEG2K Lossy Quality = ". This number can range from 95% to 1%. Does it change when you change your lossy compression?
    Bruce

    Thank you Bruce for your response.


    Yes, the value on lossy quality in logs changes from 1 to 99 in response to changes in the ini file.
    These "square boxes" artefacts are present all the way from 1 to 99.
    With very high compression ratios (e.g 3), other expected artefacts appear together with the squares.
    With quality value of 100 and above, the logs say J2K lossless and the artefacts disappear.


    I have done further tests on 3 16-bit generating CT machines and one MR.


    It seems the artefacts are not consistent:


    On one of the CT machines, all the CT brains I checked had the artefacts. Other studies I checked like Spine, abdomen did not show the artefacts.
    On the second CT machine, The reconstructed images had artefacts (e.g coronal images) while the original (axial) images did not have any problem.
    On one of the CTs and the MR, the few studies I tested did not show any artefacts (Maybe they changed the protocols on the equipment?)


    I usually do the J2K lossy using Pacsone. Compression using Pacsone does not result in such artefacts.


    Thanks.


    Mukoya.

    Hi,


    I am going to assume you are on 1417d. If not, upgrade to that version. If you are, the fact that the number changes means the problem is in the JasPer library. This is strange because the JasPer library is used by many programs, but kit is also poorly documented. It may be fixed by switching libraries. The default build is to use JasPer (HAVE_LIBJASPER) for JPEG2000, we can ask Marcel for a dgate build with the openJPEG library (HAVE_LIBOPENJPEG) and see if that fixes it. If he can not, I will try to setup a system to build it on.


    Bruce

    Thanks Bruce.


    I am actually using version 1.4.19 alpha.


    I also experienced this issue with 1.4.17d.


    Anyway, for me, I can live with the little inconvenience caused by this issue in case there is no easy fix.


    Please let me know in case you come up with something you would like me to test.


    Regards.

    Interesting finding: When the images are compressed by conquest as j2k and then decompressed by conquest, they turn out fine.


    For example:


    1. Send study to Conq1 as Lossless > Conq1 compresses as j2K and sends to viewer > Artefacts present.


    2. Send study to Conq1 as Lossless > Conq1 forwards to Conq2 as J2k > Conq2 Forwards to Viewer as uncompressed/lossless > No artefacts


    I am wondering if there might be an issue between the specific viewer (MediView Web) and conquest.


    I will try find another j2k capable viewer to test this. Seems inbuilt Conquest viewer does not support j2k.


    Thanks.

    Just confirmed with two different viewers. Same images that show artefacts on MediView Web are OK when viewed with the other viewers.


    I guess it could be a combination of how conquest compresses the images and mediview's decoding. This is because when the same images are compressed as j2k by pacsone, such artefacts do not appear when viewing in Mediview.


    Thanks.

Participate now!

Don’t have an account yet? Register yourself now and be a part of our community!

Register Yourself Login