Figure 6 .

The agreement between manual and automatic segmentation. The bilateral claustra of 10 subjects were manually segmented by 2 observers (O1 = observer 1; O2 = observer 2; auto = automatic segmentation algorithm described herein). (A) The agreement between the different segmentations was quantified using the Dice coefficient. On average, the greatest agreement was found between the 2 manual segmentations. The agreement between our automatic segmentation and the 2 observers is moderate, as expected by the small size of the claustrum. Dots represent individual subjects. Lines represent medians. (B–C) Axial slices of the right claustrum for the best subject (B) and worst subject (C) when comparing with observer 1, as marked in panel A. Notice that in (C), the segmentation error results mainly from the inaccurate putamen segmentation, so that some of the voxels labeled as associated with the claustrum (in blue) are actually overlapping with the putamen. O1: observer 1 (manual segmentation); O2: observer 2 (manual segmentation).