I would think a series of morphological closing operations with larger and larger structuring elements would do the work to some degree. Else you will have to come up with a local thickness measure to control your method.

Just a quick idea: you could perhaps (haven't tried it) use a 3D erosion followed by a dilation to remove the thin structures; then calculate the difference to the original and apply a dilation to this difference; finally "add" the result to the original mask.

Thank you all for the suggestions. I look forward to trying them on our data.

Gary,

I come from the remote sensing community (satellite, airborne, and acoustic shipborne imaging) with about 4 decades of 'experience'.  I've had an interest in the use of some of the image processing and analysis methods used in remote sensing within the medical imaging field; I realize that I know very little about the current state of the art within the medical imaging field so I want to say sorry ahead of time if I state the 'obvious' to you and others. 

What I would do is apply a small High Pass Spatial Filter (HPSF), stretch all the pixel values in the original image that are within 1.5 to 2 std dev in the HPSF to zero (which would leave the 'thinner lines / features in the original image, and then thicken the lines/pixels that are left, and then merge it back with the original.  In the remote sensing field we are use to working with images that have 4000 by 4000 or more pixels which I think is not the case in the medical imaging field, so I would think that a filter window size of 3 by 3 to 5 by 5 might be a good starting point (I would have to see an example and know how many pixels are in the image).

Again, sorry if I 'missed' the mark because of not knowing what can and can not be done at this staage in your field.......... I am willing to learn.

Pat

Try some of the filters in the Leung Malik filter bank