[MINC-users] mincbet

[Claude LEPAGE]

Hi Jon,

To answer some of your questions in your recent messages...

> I'm not seeing how one plots the histogram output from mincstats 
> (maybe you use another program or that?) nor how to tweak nu-correct 
> to optimize the normalization? 

mincstats -histo t1.txt -bins 1024 t1.mnc
gnuplot
gnuplot> plot "t1.txt" u 1:2 w li
gnuplot> set grid
gnuplot> set xrange [0:400]        (choose suitable upper limit)
gnuplot> replot

You may want to cut off the first line of t1.txt which is a very
high voxel count for background.

> If mincbet fails to converge, does it not generate a mask if you have asked it to?

It will write out the last available mask, which may not be accurate at all.

> Would anybody know why volumes from the same scanner would yield 
> vastly different mincbet results? On several scans, I wind up with 
> only half the brain masked. I've tried with and without nu_correct 
> first, though mincbet on the nu_corrected file often results in seg 
> faults. Is there any way to compare the image files to see why and how 
> they are different (I'm not sure what the critical parameters might 
> be). Any suggestions would be appreciated.

In general, the masks should be similar. I have to admit though that
mincbet is quite sensitive to data at 3 Tesla. Moreover, to obtain
good non-uniformity correction, a good mask is needed. So there is no
easy solution. I have been trying to improve mincbet to remove noise
in the selection of the thresholds. It would be very useful if you 
could provide a pair of scans - one that works and one that fails -
for testing. I could give you a better answer of the problem and
solution. One thing you may want to try is to crop the image to 
remove the neck, something like:
  mincreshape -start 0,0,0 -count 160,180,240  in.mnc out.mnc
This will ensure that mincbet starts with a good centre of mass, which
does not seem to be that easy at 3 Tesla.

Claude