I am acquiring image sets from both modalities in an attempt to use the resolution of CT and the soft tissue contrast of MRI for more accurate pathological tissue characterization. I am not sure whether I need to limit my FOV in the MRI to the lowest setting of our uCT in order to be able to do meaningful image analysis. Any help would be greatly appreciated.
Given that the image modalities use a totally different type of physics to generate the images you are under no compulsion to limit FOV, resolution or other parameters. Your first task will remain producing an accurate transformation between the to FOV at the appropriate coordinates where the tissue to be correlated has the same volume element.
Structurally we will see more soft-tissues structures with MRI than CT. CT without a contrast does relatively poorly on separating continuous soft tissue.
Fixing the FOV, resolution, etc to be the same size may help at first glance to help co-register the two images. However, since the transformation from one image space to another will not be analytically exact for these values and the MTF for each images will be different you are best served having as much resolution to work with to "fit" the image spaces together before cross-correlation.
Given that the image modalities use a totally different type of physics to generate the images you are under no compulsion to limit FOV, resolution or other parameters. Your first task will remain producing an accurate transformation between the to FOV at the appropriate coordinates where the tissue to be correlated has the same volume element.
Structurally we will see more soft-tissues structures with MRI than CT. CT without a contrast does relatively poorly on separating continuous soft tissue.
Fixing the FOV, resolution, etc to be the same size may help at first glance to help co-register the two images. However, since the transformation from one image space to another will not be analytically exact for these values and the MTF for each images will be different you are best served having as much resolution to work with to "fit" the image spaces together before cross-correlation.
I agree with H. Morris.
Making the grids of the MR and the CT equal, would only speed up the first iteration of a mutual information registration. All subsequent (many!) iterations will rotate and shift one grid, and get interpolated pixelvalues from the fixed modality. So the slight initial gain is immediately lost.
Pure cross-correlation only gets linear relations between the image intensities, so pure cross-correlation is not sufficient for joining CT and MRI images. As Lambert Zijp mentioned, something like mutual information or as a simple approximation a local cross-correlation should be used for comparison.
Now to the FOV. As H. Morris mentioned, FOV and image resolution are driven by the physics of the acquisition technology. For a good and nowadays standard image registration algorithm FOV and image resolution does not really matter. It takes some effort to make it run, but this is standard procedure. A simple rule of thumb is: make sure, that the MRI FOV is a true subdomain of the CT-FOV. Use the MRI as "fixed image" and use for instance mutual information or normalized gradient fields as error measure. Then the alignment can be reached quick and easily.
Normalized mutual information, introduced by Studholme et al. may be a good jumping off point for your future explorations in this area. If you have access to it, Lecture Notes in Computer Science, v 6204, published in 2010, has a nice review chapter by N.D. Cahill.
I agree with H. Morris and Lambert Zijp. Having as much as the same gantry position and volume size may also contribute to bring you the best results for 'fitting' the image spaces.
To add slightly to what H Morris said.
It will be vary very difficult to acquire images on the same subject in two different modalities in the exact same plane unless it is a combined scanner. To my knowledge a combined MRI-CT scanner has not yet been developed, but there is always hope.
A fairly common slice thickness in CT is 5mm. If you acquire both your CT and MRI with the same FOV and slice thickness but do not get the exact same slice placement, a small difference in axial rotation (roll, or pitch) or a z-shift (+/-) will cause huge headaches in the cross registration algorithm. In most cases a high resolution MR image will provide 3 to 5 slices covering the same 5mm slice in the CT providing a much better co-registration.
It is best to have one very high resolution image when doing a cross modalities co-registration.
Matthew Getzin mentioned "uCT" in his question, so he may work with other equipment than used in hospitals, and other slice thickness (I guess "u" stands for "micro"). The comment about "the exact same plane" is of course valid anyway.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing