Evaluate the accuracy of DIR

[Esther]

There are several ways to evaluate the accuracy of DIR. One of the
most used ways is dice similarity coefficient. In DIRART, you can
compare two structures under ART Tools. However, it does not make
sense to compare the contour on the CT of the treatment day (moving
image) with the deformed contour on the CT of the treatment day. I
believe the better approach should be to compare the deformed contour
on the CT of the treatment day with the contour on the planning CT
(fixed image). DIRART cannot do this comparison. Is it possible can
you add this function? Or can you let me know specific m-file that I
can modify it?

Thank you very much for your help.

Esther

[Yang, Deshan]

Esther

You can use DIRART to register the images and then to deform associate
contours in either direction, if you know how to so. So DIRART does
support contour comparison in the way you suggested. You just need to
deform the contour in the reverse way. Please pay attention to the
features "inverse DVF", or "inverse consistency".

BTW, verification of DIR is a very big topic. Dice coefficient by itself
won't be sufficient.

Deshan

Esther
The materials in this message are private and may contain Protected Healthcare Information. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail.

[Esther]

Dear Deshan,

I did notice the features of inverse DVF and inverse consistency.
After the deformable registration done, I can "compute the inverse
DVF" and I can also "compute DVF from inverse DVF" after I have
inverse DVF.

For the structure comparison, I can only compare deformed structures
on "Moving Image set" with structures on "Moving Image set." Or I
deformed the structures on Fixed Image set, then I can compare
deformed structures on "Fixed Image set" with structures on "Fixed
Image set." Do I do something wrong? Can you please tell me your
steps to get the comparison of structures on Fixed Image and deformed
structures on Moving Image?


Another questions is regarding the figure generated from inverse
consistency analysis. The x-axis is in unit of pixel and y-axis is
log10? I don't quite understand what this mean exactly. Let say
x=0.00507 and y=4.43. Can you tell me what these values mean? There
is also a values shown on the commend line:
==================================================
Reverse consistency analysis results:
==================================================
Mean = 6.570465e-004, std = 1.055049e-003
Median = 2.931104e-004, max = 3.382152e-002
WO boundary: Mean = 6.706252e-004, std = 1.063051e-003
WO boundary: Median = 3.037901e-004, max = 3.382152e-002

What do these values mean?


Last question I have now is if you do a phantom study using DIRART.
If so, do you have any publications to share with me? If not, how do
you get the ground truth since DIRART offers the analysis of "motion
field error" with ground truth?

Thank you so much for your effort!!

Esther

[Yang, Deshan]

I think you can deform the structures either way, e.g. to use the DVF to deform structures from the moving image onto the fixed image, and use the IDVF to deform the structures on the fixed image onto the moving image. After you deform a structure, you should be able to assign this structure onto either image, and DIRART will be the dirty work for you by reslicing the structures.

To do what you want, you need to reassign the deformed structures on the moving image onto the fixed image, then you can compare them.

For your second question, as I remember, the y is the log of the number of pixels, so the plot is the semi-log plot of the histogram. Please read the code to understand it.

#3, you should try to obtain the lung CT images from www.dir-lab.com. They have both images and landmark point data. Ground truth on patient data is very difficult to come by. You can of course prepare an artificial DVF field, and use the DVF to deform a 3D image. Then you can use the artificial DVF as the ground truth, and this was what I did in one of my paper.

[Esther]

Thank you so much!! I did find the way to compare the structure
similarity. However, I did some phantom study, but I don't know how
to use your implemented tool "Motion Field Error Analysis with Ground
Truth." What kind of information should be contained in the MAT
file? I got a error message "ERROR: Motion field is not available in
the file." Also, I found that there is "Land Mark Analysis" m-file
under dirart_programs, but it seems not to be seen at the GUI
interface. Is this function available to be used?

I also got the image from DIR lab, but won't be able to get images
loaded into AIRART. Have you ever done those image sets before? If
so, can you please provide me some guides?

Cheers,
Esther