[Insight-users] mi registration problem

[Dill, John]

The bar does not exist in the CT, and there is no 3D volume wrt the
portfilm, so I can't mask it out.  I don't think replacing the bar with the
background value is a good idea, as it removes the underlying real data
information.

After some more thought, I do have the ability to take a portfilm without a
patient, and hopefully with that, I'll be able to form a intensity model of
the bar and be able to search for it, through some line correlation.  If I
can assume that the error is additive, I can find where the bar is, and then
take the bar intensity model and subtract it out to get the original image
data, and use that in the registration.  It's a little more complicated
perhaps due to differing widths in the projections due to the location of
the bar, but I haven't determined its effect yet.  There is another problem
where the distortion in some images is not just a straight bar, but has more
shape due to the position of the patient on the table.  Does this seem like
a step in the right direction?

John