Hello,
The parcellation of grey and white matter appears to be off in several T1-weighted structural scans I have been processing. The effect is observed in the temporal lobes, parietal lobes, and part of the occipital lobes. The frontal lobe does not seem to be affected as much.
For example, the medial white matter is super white, while more distal white matter near the cortex is less white (grey). Grey matter itself is rather dark (black in some cases). If you monkey around with the brightness of the image you can see that, for example, the pial line has actually followed the distal white matter and not the grey matter. Moreover, the white matter line has followed the boundary between medial (super white) white matter and more distal (greyer) white matter. This effect is most pronounced in the temporal and parietal lobes nearest the brightest parts of the white matter across the entire scan. About 3-5 voxels are cut off. For the frontal lobe and caudal occipital lobe, though, it is only cutting off about 1 voxel around the edge of the brain.
I have attached a picture of this problem with the pial and white matter lines on the image. Look at the cortex in the lower left hand part of the image (lateral temporal cortex). For the image called Missing_Voxels.tiff I have adjusted the standard brightness of the image in freesurfer to show how much of the brain we are missing. In the image called Missing_Voxels_Brighter.tiff I have left the brightness levels as they appear when you open freesurfer. The problem appears less severe in this latter image because the standard brightness level has made some of the grey matter appear black and it is easier to miss that it is actually present. Now look at the top of the image and see that the pial line there is pretty accurate, though it does appear to cut off maybe 1 voxel too much.
Extracting the eTIV from one of these scans shows that their intracranial volume is much smaller than other participants whose data were obtained on a different scanner. Also, each lobe volume is reduced. For example, the parietal lobe is about 15 standard deviations smaller than a group of participants scanned on a different scanner. These types of reductions are observed if you include grey matter only and also if you include grey+white matter in the calculation of the lobe volumes.
Thus, the problem appears to be that we are missing the most outer part of the grey matter across a large part of the brain. I should also mention that the skull itself is missing (on the T1 image) near the lateral temporal lobes so there may be some signal dropout there. Though the lateral temporal cortex itself becomes visible when the brightness of the scans is adjusted, the skull in that area never does become visible.
Is there any way to fix this problem at a basic level in one of the earlier steps of freesurfer?
I have tried putting in a bunch of control points in the "real" white matter and it does make the white matter extend out further. However, the pial line is still cutting off ~3-4 voxels of brain in the problematic areas.
Best, Christine
