MRI imaging data.
Multiple cardiac images of the same heart often vary in appearance due
to a small number of a variables. This web page considers the
analysis of the following sample data:
Sample Cardiac MRI Data
(Thanks to Nikos Tsekos from the Washington University Medical Center).
A 2 dimensional embedding for this data was found using Isomap. (see
my wumap page
for details. The discovered embedding is a map from each original image to a
point on a 2d plane. These points are shown in white below. Once all
the images load, then moving your mouse over the left image displays
the corresponding original image on the right.
MOUSE OVER BELOW IMAGE.
Current research is investigating further processing to make the axes
of the embedding represent meaningful quantities. This has been done
on this sample data to order the images by the image deformation (and
not the contrast changes). The following video shows the images in
this order. The original image number is still visible on the right
image shown. The left image shows the (Gaussian) movig weighted
average of the original images. This gives denoising of the MRI image
with minimal spatial blurring, because the averaged images had very
similar (but unknown!) deformation parameters.
I am actively seeking collaborations on this project. If you are
interested please contact me directly at pless at cs.wustl.edu or
by phone 314-935-7546. I expect to release code to perform this
embedding and visualization (in matlab and on the web) soon. Until
then, I will be happy to test code on your sample data and return the results.
