Information Technology Reference
In-Depth Information
Fig. 13 Corresponding segmented cuts from MRI scans, along with 3D polygonal mesh
superimposed with 3D reconstructions obtained from the proposed method
group. The proposed non-linear manifold embedding algorithm presented in Sect. 4
was able to reduce the high-dimensionality of the 3D data, using statistical prop-
erties of neighbouring spine models to infer a global representation of the sub-
population.
Four clusters were detected from the low-dimensional manifold of 3D models
based on inter- and intra-cluster measures. The result from this classified embedding
is presented in Fig. 14 . The
first group consisted in 37 patients with normal thoracic
kyphosis pro
les with hyper-lordosis and the highest levels in main thoracic Cobb
angles for all groups. The second group had 55 patients, mostly non-surgical (minor
curves), with low thoracic kyphosis and normal lumbar lordosis values, but with
the highest degree of rotation of the PMC from the sagittal plane. The third group
had 21 cases with hypokyphotic and hyper-lordotic pro
les. Finally, the fourth and
last group included 57 patients with hyper-kyphotic thoracic pro
les, with major
surgical curves and demonstrating very high axial rotations in the apical vertebrae.
These results show an additional group in contrast to the study by Sangole et al.
which found hypo-kyphotic subgroups in a 3D analysis which used measures such
as planes of maximal curvature and kyphosis as in
fl
uential parameters that split the
cases. Coronal and sagittal spine pro
les, as well as the da Vinci schemas [ 55 ] of the
representative cases that were identi
ed as the cluster centers are illustrated in
Fig. 15 . The manifold representation can potentially be useful for classi
cation of
3D spinal pathologies such as AIS and serve as a tool for understanding the pro-
gression of deformities in longitudinal studies.
Search WWH ::




Custom Search