Hello,
I'm using a home-grown GCA constructed from five acallosal subjects to try to improve the automatic labeling and segmentation of volumes from this population. However, the labeling was quite bad when performed on a representative subject, and the ventricles were only partially filled and crossed the IHF in areas:
Left_Lateral_Ventricle (4): linear fit = 1.00x + 0.0 (713 voxels, overlap=0.480) Left_Lateral_Ventricle (4): linear fit = 1.00x + 0.0 (713 voxels, peak = 38), gca=37.8 Right_Lateral_Ventricle (43): linear fit = 1.00x + 0.0 (370 voxels, overlap=0.292) Right_Lateral_Ventricle (43): linear fit = 1.00x + 0.0 (370 voxels, peak=38)
I also noticed that the 152-subject MNI atlas that came with FS had ventricles that were much "darker" (when converted to NIFTI and displayed in fslview) than my own atlas: the mean ventricle intensities were about 15 and 30, respectively. We suffer from a lack of data, too. Is it possible to improve the GCA and labeling with such limited data?
Jason