[Insight-users] Best way to detect 'bouncing' level set solution?

[Zachary Pincus]

Hello,

Does anyone have any suggestions as to the best way to detect when the 
solution to a level set segmentation starts "bouncing" -- that is, 
oscillating back and forth around the true solution? Or to prevent it 
from bouncing in the first place?

It seems like one way would be to look for a short cycle in the values 
of the RMS change from iteration to iteration -- but I'm not sure that 
this would catch all bouncing solutions and wouldn't wrongly flag some 
non-bouncing solutions. Another option would be to actually keep a 
record of the level sets at the past few iterations and at each step 
compare the current level set to the previous few to detect a 'bounce' 
-- though that seems extremely expensive.

Are there any standard ways of dealing with this that anyone knows of?

Thanks,

Zach Pincus

Department of Biochemistry and Program in Biomedical Informatics
Stanford University School of Medicine