Doubts about linear and Hounsfield measures in DICOM images

[davidec]

Hello!
I'm currently developing part of an application which allows to
perform linear measures on an image and Hounsfield measures on an area
or a single pixel.
In general, I have doubts on how to display the computed value,
though: The application can zoom images, so a scale factor is
introduced. I think I can't display an exact value, but I should use a
tolerance instead.

In particular:
1. I'm using the PixelSpacing tag to find the relation between image
pixel and millimiters: Does it make sense to speak about tolerance for
PixelSpacing? If so, how can it be determined? Is there a tag or is it
related to the number of decimal digits used in the value or ...?
2. How does PixelAspectRatio influence measure and tolerance?
3. If the measured image is scaled, one pixel of the screen does not
match only one pixel of the image: Should I consider the error in the
measure or is it a habit to consider exact values? In the first case,
how does scaling affect the error on the measure?

Similar considerations apply to Hounsfield measures, I think, so
having a reply to these questions should point me in the right
direction...

I realize some of the questions are a bit naive, but I think the
subject is sensitive, so I'd like not to skip basic concepts... ;)

Thank you in advance!

[jmilet]

Hey David

Here are some of ideas regarding your post.

1. Let's start from the beginning. There are at least two images
involved in this process. The first one, the dicom image and the
second one, the display image ( the one that the user see, with zoom,
pan, clips , etc ).

2. Let's say that the dicom image has 16 bits . In this case, the
intensity of each pixel represent Hounsfield units.

3. When the display image is produced (starting from the dicom image )
at lookup table is applied. Thus, the display image has at most 255
levels (0 to 255) , while the 'dicom image ' has at most 2 ^ 16
different values (with possible negative values ).

4. Let's say the user selects a particular region of interest ROI in
the display image , say a rectangle. And the task is to give
provide the user with info regarding the size of the ROI and some info
regarding the mean and variance of the Hounsfield measures. So the
question is how to compute these values ?

5. Regarding the size of the ROI (in mm) I will suggest to transform
the points x1,y1, x2,y2 of the display image into points x1', y1',
x2', y2' at the dicom image (Here you might have to deal with some
rounding issues). Once you have the points x1', y1', x2' and y2' use
info such as distance between x and y pixels to compute the ROI's
size.

6. To compute the info regarding the hounsfield measures ,use
coordinates x1', y1', x2', y2' in the dicom image. Having these
coordinates allows the computation of values such as mean, variance,
histograms , etc.

7. In summary, use the display image to capture the user ROI
coordinates, transform these coordinates in to dicom image
coordinates. Use the ROI at the dicom image to compute the info
required. Notice that the the display image contains a set of gray
scale values while the dicom image a set of hounsfield values.

Hope this help

Jose

[davidec]

Thanks for your suggestion and your help, but my problem is a bit
different.
For Hounsfield measures I'm wondering whether it makes sense to
provide tolerances to measures (which are indirect).
In these measures, what participates to the computation are pixels'
gray values and other values such as RescaleSlope and
RescaleIntercept. Provided measures originate form a computation of
these values.
The question is whether I need to specify an error (a tolerance) for
these starting values, and if so, of what magnitude?

Thanks in advance!

[V]

Hi,

I was investigating this myself last week and I'm still researching.
Please let me know if you receive a satisfactory respopnse.

Thanks,

Vince