I found some normal and abnormal RVGs for use with your class.  It sounds like
an interesting project.  You taught me most of what I know about structured
programming - I learned by watching over your shoulder!

The first 512-byte block is ascii header, the second and fourth are unused, the
third is real*4 zcount.  The image data begin in the fifth block, they are
2-byte integers.  For RVGs, there are 32 frames, 64x64 pixels each.  Thus, 8kb
per image frame.  The data are in DECstation format; if you use a SPARC, you
will probably need to switch the bigendian/littleendian byte order for the
integer*2 images - I have a  FORTRAN subroutine for that if you need it.  The
zcount is messier, with differing floating point formats between the 2
machines.  You could just skip that.

----

The rvgs are on your computer now.  The image data have been byte-switched; if
they don't look right, I can send the original, but you probably need the
switched data.  The files are Wiener filtered, using your filter!  I believe
all patient identifiers have been removed, e.g. in block 1 (ASCII).  Could you
double check?

rvg001 - left ventricular enlargement due to coronary artery disease.
rvg002 - normal
rvg003 - normal heart in a patient pre-lung transplantation.  Note the reduced
activity in the right lung base due to lack of lung tissue.

---

I have now transferred the unfiltered versions of the 3 rvgs to your
workstation.  For each patient, there are 6 files:  *.x0* are the unfiltered
for the 3 views (approx. 35deg LAO, anterior,lateral) and *.xs* are the
corresponding filtered files.

Your programs, written in 1983 and 84, used a fixed Wiener filter and a
temporal smoothing.  In 1984, Ed Rollins modified these to use an adaptive
I have now transferred the unfiltered versions of the 3 rvgs to your
workstation.  For each patient, there are 6 files:  *.x0* are the unfiltered
for the 3 views (approx. 35deg LAO, anterior,lateral) and *.xs* are the
corresponding filtered files.

Your programs, written in 1983 and 84, used a fixed Wiener filter and a
temporal smoothing.  In 1984, Ed Rollins modified these to use an adaptive
filter, smoother in low-count areas and stronger in high-count regions.

Your original filter C32 (the one we liked best) is an 11x11 filter with 6
unique coefficients.  Starting with h(0):

0.9469
0.4811
-0.0917
-0.2345
-0.0997
0.0043

The length 7 temporal smooth is:

0.44407
0.33614
0.12500
-0.08614

A possible filter exercise for your students would be to write a program to
compute the frequency response; as you know, the filters should always be
thought of in terms of their frequency response even if they are computed in
the spatial domain.

As you know, filters are fun for me, and I enjoy recalling our work together. 
If you need more help, let me know.

Tom
filter, smoother in low-count areas and stronger in high-count regions.

Your original filter C32 (the one we liked best) is an 11x11 filter with 6
unique coefficients.  Starting with h(0):

0.9469
0.4811
-0.0917
-0.2345
-0.0997
0.0043

The length 7 temporal smooth is:

0.44407
0.33614
0.12500
-0.08614

A possible filter exercise for your students would be to write a program to
compute the frequency response; as you know, the filters should always be
thought of in terms of their frequency response even if they are computed in
the spatial domain.

As you know, filters are fun for me, and I enjoy recalling our work together. 
If you need more help, let me know.

Tom