Caudate Shape Discrimination in Schizophrenia

1
Caudate Shape Discrimination in
Schizophrenia Using Template-free
Non-parametric Tests Y. Sampath K. Vetsa1, Martin
Styner1, Stephen M. Pizer1, Jeffrey A.
Lieberman2, Guido Gerig1,2 1 Department of
Computer Science, 2 Department of
Psychiatry, University of North Carolina, Chapel
Hill, NC 27599, USA
We describe the shape analysis of the caudate
nucleus structure in a large schizophrenia study.
Left and right caudates were segmented from high
resolution MRI using a reliable, semi-automated
technique. The caudate shapes were parameterized
by spherical harmonic surface description,
aligned and finally represented as medial mesh
structures (m-reps). Schizophrenia patients were
categorized based on treatment, so we could test
size and shape differences between normals,
typically and atypically treated subjects.
Statistical shape analysis was done using
permutation tests on shapes represented by medial
representations. Significant size and shape
differences were found, not only between
treatment groups and controls, but also between
the treatment groups.
VOLUME DIFFERENCES
For the volume tests, the signed difference
between the average volumes of the groups is used
as the distance measure, with a a two-sided
permutation test. We used 100,000 permutations.
Statistics for caudate volumes, with correction
for ICV (intracranial volume). The table lists
p-values. All comparisons are significant.
GLOBAL SHAPE DIFFERENCES
The tables below show the m-rep statistics for
deformation and width, integrated over the whole
shape. The tables list the p-values, with
significant results marked with . We used
10,000 permutation tests with one sided tests.
The patient study is cross-sectional in design.
Early illness (age 16-30, N34) and chronic (age
31-60, N22) patients have been matched to a
young and an older control group (N26). All
patients and comparison subjects were
right-handed males. At the time of the scan, 17
early illness patients and 5 chronic patients
were on typical antipsychotic medications, and 17
early illness and 18 chronic patients were on
atypical antipsychotic medications.
M-rep statistics for the mesh deformation
measure Mesh deformation distance (L1)
calculated as the integrated deformation over the
whole mesh.
Caudate structures were segmented using a
semi-automated technique. Following automatic
tissue classification, a trained user masked the
gray matter structure of the caudate by
overpainting. After segmentation, the shapes were
processed by surface extraction and
parameterization using spherical harmonics
(SPHARM). The shapes were then aligned using
first degree spherical harmonics, which also
gives the correspondence. An m-rep model is
generated with the optimal minimum mesh
sampling, and this m-rep model is used to
generate the m-reps for the individual objects by
deformation into each object.
M-rep statistics for the local width measure
Width (radius) change measured as the integrated
absolute radius differences over the whole mesh.
LOCAL SHAPE DIFFERENCES
The two sets of figures below show the local
shape difference statistics. In the left figure,
the locations on the mesh with p-values more than
0.05 are suppressed, with lower p-value indicated
by darker shading. In the right figure, the
locations on the mesh with p-values more than
0.05 are suppressed, with the size and color of
the ball at each node indicating the significance.
Caudate segmentation
SPHARM representation
PDM M-rep
M-rep Radius
Implied surface
Local deformation statistics for Typ-Atyp
group tests for the left caudate.
The problem with shape analysis is High Dimension
Low Sample Size (HDLSS).
Possible Approach

• Feature Selection
• Feature reduction (e.g. Principal Component
  Analysis and selection of a few modes)
• Multivariate statistical analysis

Our Approach Permutation test

• Random permutation of shapes
• Calculate average shapes
• Scalar shape distance measure between shapes
• Test significance of the original group average
  difference

Local width statistics for Typ-Atyp
group tests for the left caudate.
Surfaces implied from m-reps The figures below
show the average m-rep surfaces for the average
Atyp (solid) and the average Atyp ?(Typ
Atyp)radius overlaid (mesh). Please note that
this display differs from an overlay of average
shapes and illustrates significant changes (here
radius only) between the groups.
The figure below shows an example of a
permutation test. The number of permutations are
on the x-axis, and the scalar shape difference is
plotted on the y-axis. The distance between the
original groups is plotted as the thick
horizontal line. In this case, the difference
between the groups is significant.
Side View
Top View
This work was supported by the NCI grant P01
CA47982. The Stanley Medical Research Institute
is acknowledged for funding the MRI schizophrenia
study providing the caudate structures.
Miranda Chakos and Scott Schobel are
acknowledged for providing the caudate
segmentations.