Hi Isabel,

if dura is left around you are *much* better off than if brain is stripped. In the latter case we will never recover, while if there is some dura we generally can avoid including it in the surfaces. Note that in the screenshots you attached that the dura was labeled as cortex by the aseg, but that is irrelevant as we don't use the aseg for cortical analysis, but rather the ?h.white and ?h.pial surfaces. Avoiding dura in general is an extremely difficult, and in fact unsolved problem, as it can be physically adjacent to the cortex and has similar contrast properties on most T1-weighted images. In locations where there is some spatial separation we usually avoid it, but if not you can manually erase it in the brainmask.mgz.

cheers, Bruce

On Mon, 17 Jul 2006, Dziobek, Isabel wrote:

Hi Everybody,

I have a question in regard to the skull stripping and subsequent grey and white matter segmentation. The skull stripping process is faulty with almost all my subjects, leaving large parts of the dura and sagittal sinus. I tried changing the watershed parameters but that did not work (it seems that once parts of the dura are removed, other parts of the more inferior aspects of the brain are removed too). It would be good to have the option to e.g. only eat up some pixels from the top, is that possible? Also, on what is the subsequent segmentation in gray and white matter dependent? It seems that in some cases, although dura and sagittal sinus are still thee after the skull strip, they are correctly not segmented into gray (or white) matter. So I am wondering if the skull strip does not have to be perfect but instead ohter subsequent steps are more important. Is it possible to influence these steps? I am attaching a snapshot of a slide, in which the dura is misclassified as gray matter and the sagittal sinus is not classified at all.

Thanks in advance for your help!! Best,

Isabel

PS. Are there any example scans (other than bert) where I can see how a very good (or good enough) skull strip and segmentation look like?