Information Technology Reference
In-Depth Information
Fig. 6. Evaluation of FBM-TS on a real 3T brain scan (a). For comparison the tissue
segmentation obtained with FBM-T is shown in (b). The tissue segmentation obtained
with FBM-TS is given in (c). Major differences between tissue segmentations (images
(b) and (c)) are pointed out using arrows. Image (d) shows the structure segmentation
with FBM-TS
the gain obtained with tissue and structure cooperation is particularly clear for
the putamens and thalamus.
We also computed via STAPLE [26] a structure gold standard using three
manual expert segmentations of BrainWeb images. We considered the left cau-
date, left putamen and left thalamus which are of special interest in various
neuroanatomical studies and compared the results with LOCUS-TS [4] and
FreeSurfer (see Table 2). FBM-TS lower results for the caudate were due to
a bad registration of the atlas in this region. For the putamen and thalamus
the improvement is respectively of 14.7% and 20.3%. Due to high computa-
tional times, only the 5% noise, 40% nonuniformity image was considered with
Freesurfer. For this image, results obtained for FBM-TS were respectively 74%,
Table 2. FBM-T. Mean Dice metric and mean computational time (M.C.T) values
on BrainWeb over 8 experiments for different values of noise (3%, 5%, 7%, 9%) and
nonuniformity (20%, 40% )
 
Search WWH ::




Custom Search