Mammography CAD SR, for Processing v.s. for Presentation - DICOM

This is a discussion on Mammography CAD SR, for Processing v.s. for Presentation - DICOM ; Hello, When a Mammography CAD SR "points" to the "For Presentation" mammography image it's an easy task to load the referenced image and bind the SR to it for showing the CAD findings. But, I understand that some CAD products ...

+ Reply to Thread
Results 1 to 8 of 8

Thread: Mammography CAD SR, for Processing v.s. for Presentation

  1. Mammography CAD SR, for Processing v.s. for Presentation


    Hello,

    When a Mammography CAD SR "points" to the "For Presentation"
    mammography image it's an easy task to load the referenced image and
    bind the SR to it for showing the CAD findings.

    But, I understand that some CAD products require input images to
    contain raw, unprocessed (FOR PROCESSING) pixel data.

    in that case:
    1. is it possible (and dicom legal) that the Mammo CAD SR will refer
    the FOR PROCESSING images.
    2. how should the rendering device work - does it have to find the "FOR
    PRESENTATION" image that matches the "FOR PROCESSING" one pointed by
    the SR.
    3. if so, how can the rendering device get to the for presentation
    image.

    Thanks,
    Roi


  2. Re: Mammography CAD SR, for Processing v.s. for Presentation


    r...@algotec.co.il wrote:
    > But, I understand that some CAD products require input images to
    > contain raw, unprocessed (FOR PROCESSING) pixel data.
    >
    > in that case:
    > 1. is it possible (and dicom legal) that the Mammo CAD SR will refer
    > the FOR PROCESSING images.


    I share your concern about this issue. Regrettably I have no answers.
    For me it seems very strange to generate SR objects for raw (FOR
    PROCESSING) images. To my knowledge some of these CAD products are FDA
    approved.

    The image viewing system will have severe problems in showing the SR
    together with the FOR PRESENTATION image. How can one be sure that the
    image is not shifted, flipped etc when it is transformed from a FOR
    PROCESSING image to an FOR PRESENTATION image? Frame of Reference is
    not mandatory for MG images.


  3. Re: Mammography CAD SR, for Processing v.s. for Presentation


    Thanks,

    I can understand why a CAD product needs "for processing" data but I
    don't really know how the rendering device should match the SR with the
    "FOR PRESENTATION" data.

    I was suggested to look for an image with Source Image Sequence
    (0008,2112) equal to the one referenced by the SR
    (the FOR PRESENTATION image is derived from the FOR PROCESSING one and
    should have the Source Image Sequence pointing to it)
    but I can't be sure the archive will provide a query that allows me
    retrieve this info.

    I also agree with you that I can't be sure that the pixel coordination
    of the SR findings will fit the location of the "FOR PRESENTATION"
    image if it was created according to the FOR PROCESSING one.
    what do you think ?

    Roi


  4. Re: Mammography CAD SR, for Processing v.s. for Presentation

    >From having discussed the issue with a development engineer at a mammo
    cad company, I understand thid issue may be that some digital mammo
    vendors change more than the pixel values in the FOR PROCESSING to FOR
    PRESENTATION transform.
    At least one vendor (I don't know who) is also doing a geometric
    transform, rearranging location of pixel values in the for presentation
    images. Not sure why they would be doing that, but possibly it is to
    counteract distortion induced by parallax or xray path.My source at the
    CAD company said they do it to "make the images prettier for the
    radiologist reading the image". At any rate, since the CAD algorithms
    already account for these distortions in their computation of where
    significant image features are located in the breast, these fixes to
    the distortions break the CAD algorith. Hence the CAD vendors require
    the FOR PROCESSING. Seems counter intuitive since the CAD SR has to be
    displayed with the image in order for the radiologist to use the
    result. One immediate idea to fix the problem would be for the CAD SR
    to reference the FOR PROCESSING along with a GSPS which would transform
    the image into a displayable state. However there are a couple problems
    with that approach. Current spatial reference doesn't support that kind
    of double reference - point to an object with the provided display
    transform (HL7 Imaging SIG/DICOM WG20 is addressing that very thing in
    CDA at the HL7 conference this week) Also, even if you could do the
    dual object and presentation reference in the SR , the fact that the
    Digital Mammo devuce is introducing geometric distortions in the For
    Presentation image (or de-distortions if you please) is problematic: a
    spatial reference to a location in the For Processing Image doesn't
    neccessarily point to the same anatomic location in the For
    Presentation image. Since GSPS does not support a warping function as
    one of its transforms a GSPS cannot capture that distortion either.
    Perhaps the new DICOM spatial registration object could be used to
    capture the (de-distorting) warp of the image; but then you have the
    problem in triplicate. The SR makes a reference to a location in the
    for processing image that must be displayed after being transformed
    both by a greyscale transformation in the GSPS and geometric distortion
    in a spatial registration object. I'll agree, the process is broken. I
    think this has to go to WG6 for a fix.
    This was supposed to be one of the discussion topics in the
    workshop/seminar at SCAR last june re: Issues with CR to DX
    Transition. Don't know if it ever got more than a passing mention
    because I didn't make it to SCAR


  5. Re: Mammography CAD SR, for Processing v.s. for Presentation

    DICOM could say more about this.
    Here is some history that may help:

    The "For Presentation" and "For Processing" dichotomy originates
    from the fact that the early digital radiography (CR, DX) manufacturers
    created new viewing flavors" (such as "unsharped masking") to make
    clinically relevant features more conspicuous.
    Such processing became popular with radiologists but was problematic
    to products or institutions that apply different processing variants.
    Hence, there is a DICOM element to distinguish these two types.

    You should be able to display either image type without ill effect, other
    than that some radiologists prefer a more familiar processed image.
    If there are exceptions or other issues, I'd appreciate hearing them.

    - Doug

    wrote in message news:1106215671.012880.138380@f14g2000cwb.googlegr oups.com...
    >
    > Hello,
    >
    > When a Mammography CAD SR "points" to the "For Presentation"
    > mammography image it's an easy task to load the referenced image and
    > bind the SR to it for showing the CAD findings.
    >
    > But, I understand that some CAD products require input images to
    > contain raw, unprocessed (FOR PROCESSING) pixel data.
    >
    > in that case:
    > 1. is it possible (and dicom legal) that the Mammo CAD SR will refer
    > the FOR PROCESSING images.
    > 2. how should the rendering device work - does it have to find the "FOR
    > PRESENTATION" image that matches the "FOR PROCESSING" one pointed by
    > the SR.
    > 3. if so, how can the rendering device get to the for presentation
    > image.
    >
    > Thanks,
    > Roi
    >




  6. Re: Mammography CAD SR, for Processing v.s. for Presentation

    Another solution would be to require the CAD system to deliver a for
    presentation image beside the SR, and refer to this image instead of
    the raw image.

    There is a note in the DICOM standard section A.35.5.3.1.3 that which
    SOP Classes a Mammo CAD SR may refer to is documented in the
    Conformance Statement for an application. Shouldn't DICOM restrict this
    reference only to viewable SOP Classes? Or maybe there should be SR
    that are for presentation and for processing?


  7. Re: Mammography CAD SR, for Processing v.s. for Presentation

    >You should be able to display either image type without ill effect,
    other
    >than that some radiologists prefer a more familiar processed image.
    >If there are exceptions or other issues, I'd appreciate hearing them.


    Chapter 3 section C.8.11.1.1.1 is very clear that the intention of
    the for processing images is not for viewing. However the use of
    the word "intention" hints that they CAN be used for viewing?


  8. Re: Mammography CAD SR, for Processing v.s. for Presentation

    Either image is displayable. The FOR PRESENTATION image is
    display optimized, whereas the FOR PROCESSING image is not.
    See Figure C.8-13. FOR PROCESSING really means the image is
    not optimized for human interpretation, and therefore, it is NOT
    "FOR PRESENTATION" (may not have VOI LUT, or be scaled
    for P-Value mapping etc. ).

    It is perfectably reasonable for a product to display FOR PROCESSING
    images, but such a product would probably offer "intelligent" features for
    grayscale windowing and spatial filtering.

    Automated detection products (CAD) may not work well with images
    with spatial, intensity and window processing applied for human benefit.
    These products probably do not attempt to use FOR PRESENTATION images.

    - Doug

    "kr-val" wrote in message
    news:1106313479.713090.7510@z14g2000cwz.googlegrou ps.com...
    > >You should be able to display either image type without ill effect,

    > other
    >>than that some radiologists prefer a more familiar processed image.
    >>If there are exceptions or other issues, I'd appreciate hearing them.

    >
    > Chapter 3 section C.8.11.1.1.1 is very clear that the intention of
    > the for processing images is not for viewing. However the use of
    > the word "intention" hints that they CAN be used for viewing?
    >




+ Reply to Thread