Dear all, Prof. Joshi recommended this new forum to me and I am posting this thread to see if there are anybody also interested in, and sorry to bother others!


our research group has been using the BioImageSuite
Software to do multi-modalities medical image(from CT to MRI)
co-registration. We have tried different datasets to test the
performance of the software and found out several questions regarding
the brain registration facility within 3.0.1 version of BIS.





Basically we have 159 slices of CT images(size of 515kb each single
DICOM file) as reference dataset(Info Tag in IrfanView: Dimension:
512*512*8 bbp, Bits Allocated: 16, Bits stored: 12, Pixel Spacing: 0.489
0.489 1), and for the transform dataset we have several stacks of MRI
images(size of 520kb each single DICOM file) with different number of
images and Resolution(Resolutions include 520*520*8 / 320*320*8 bbp,
Bits Allocated: 16, Bits stored: 12, Pixel Spacing: 0.75 0.75 1).





And there are 3 question that we encountered, we have grabbed the manual
from your webpage and followed the instruction inside, but
unfortunately things could not work out:





 1.  We tried both of rigid and affine algorithms , with or
without initial transform seeded. While it turned out well on some MRI
sets of small volume(no larger than 60 slices) as we visually judge the
match in VolView. But when it came with larger size data sets, the
software will crash with an error during auto save. And it is highly
difficult, if not impossible to do the registration twice during one
activation of brain registration utility, with the same error. Our OS
configuration is Dell Precision T3500 with Intel Xeon CPU
E5507@2.27GHz,
and 3.0GB of RAM. I wonder if it's because of the system memory is not
sufficient. We are planning to transfer the project to a more powerful
station to see the performance but we need your supervision to judge
whether memory is an issue?



 2.  With some of the datasets , we have seen incorrect distortion imposed on MRI images, or the BIS
somehow calculated the wrong scale factors on the result transform
images that when the two volumes are merged, it looked like the MRIs are
just proportional to the geometry of the reference CT images. But since
we could not manually register the image set after the automatic
algorithm computation speculated as memory problem in above mentioned
point 1. Is there any measure taken to guarantee the precision of the
registration?



 3.  A quick question about the interpolation between the two different
datasets. Since it is the common case that we have different volumes
from the patients' brain anatomical images, sometimes the registration
result takes on wrong interpolation as a still image of the first file
of the dataset.





Above all, I believe the BIS software is an
advanced software which includes the most powerful and up-to-date
algorithms to solve the Co-registration issue, and it did provide
effective solution to part of our project. But we really appreciate for
your help for troubleshooting at this key point and improve the
performance BIS if possible. Thanks in advance for your attention!

RAM may be an issue. An easy way to find out is to reduce the Resolution of the images prior to registration. Try using the Resample tool under Image Processing/ Crop/Resample/ Resample Image. I would try to make the images isotropic 1x1x1 mm.

If the images are from the same subject you should just try rigid mapping, affine will add shear/scale/distortion (to try to improve the similarity) but in the case of multimodal images this is erroneous.

Not sure what you mean by "3" below

XP