Hey,

Maybe you can have a look at Marcel Prastawa's answer from January 13th, 2014:
http://www.nitrc.org/forum/forum.php?thread_id=4249&forum_id=2386
In this post, it is suggested to create a 'healthy' MR-image by making a linear combination of the input probabilities and the input textures that you use for generating the tumor image.

Good luck,

Tom