Transcript MICCAI03_Gerig_HippoStats

Age and treatment related local hippocampal changes in
schizophrenia explained by a novel shape analysis method
1,2G
Gerig, 3K Muller, 3E Kistner, 3Y Chi, 1M Chakos, 2M Styner,1JA Lieberman
Departments of 1Psychiatry, 2Computer Science, 3Biostatistics
University of North Carolina, Chapel Hill,NC 27614, USA
[email protected] / http://www.cs.unc.edu/~gerig
SUMMARY
RESULTS
Volume reduction of the hippocampus observed with MRI is one of the most
consistently described structural abnormalities in patients with schizophrenia.
However, the timing, the association with treatment, and an intuitive explanation
of morphologic changes are not known. This study analyzed subtle changes of
the hippocampal structure in schizophrenics (N=57) as compared to matched
controls (N=26). The effect of age, duration of illness and drug type to local
shape changes was studied with a new shape representation technique and an
exploratory statistical analysis.
The M-rep shape representation model allows a separate analysis of shape
deformation and of local atrophy. The exploratory statistical model was a
repeated measures ANOVA, cast as a general linear multivariate model.
The exploratory nature of the analysis means that the promising results must
be replicated in order to provide full confidence in the conclusions.
Shape Deformation between Schizophrenics and Controls
SUBJECTS
Quantitative 1.5 Tesla MRI examinations of the brain were utilized to
examine a young patient group in the early illness stage of illness (N=34)
and an older group of schizophrenic patients who were chronically ill (N=22).
MRI examinations of the brain were also acquired for 26 right-handed male
controls matched to both patient groups for age and handedness (young
controls N=14, older controls N=12). Patients were characterized with
regards to duration of illness and illness severity utilizing PANSS
assessments. At the time of scan, 17 early illness patients were on typical
antipsychotic medication (haloperidol), and 17 were on atypical antipsychotic
medications (13 olanzapine, 4 on risperidone). For chronic patients, only 5
were on typical medications (3 haloperidol, 1 tri u-perazine and 1
thiothixene), and 18 were on atypical medications (6 olanzapine, 8 clozapine
and 4 risperidone).
Patient-Control Euclidean distance differences (shape deformation or
bending) at age 20, 30 and 40, left to right. Height indicates the deformation
difference between patients and controls in mm. The graph characterizes time
slices of the continuous age range. The differences between patients and
controls increase over time and are located mostly in the tail and to a lesser
extent in the head of the hippocampus.
L/R Asymmetry of Radius Difference
Drug-type versus Control group differences of left
to right radius difference log2(L/R), which can be
interpreted as L to R local width asymmetry. Left:
Typical Drug - Control radius asymmetry difference
at age 20, 30 and 40. Right: Atypical Drug – Control
radius asymmetry difference. the graphs
characterize time slices of the continuous age range
and show decreasing asymmetry of local width with
age, particularly in the tail and head regions.
Comparison: Longitudinal Shape Change of Controls
METHODS
Shape
representation by
sampled medial
mesh (M-rep).
Mesh represents
shape with over
99% overlap.
Summary of Results
Features for Shape Analysis
Statistical analysis
M-rep 3x8 mesh
M-rep 3x8 mesh
Head
Tail
radius
Deformation and width
asymmetry changes due to
aging within controls for 10
years. Left: Distance model
representing shape
deformation, scale in mm.
Right: Radius model
representing left/right
asymmetry of local radius
change (local width), log-scaled
in mm.
Statistical
analysis of
3x8 mesh.
Height
represents
difference.
Radius: Local width
attribute (atrophy,
growth).
CONCLUSIONS
Deformation: Bending,
curvature change.
Statistical Model
• New analysis scheme overcomes problem of feature reduction/selection from
high-dimensional shape representation.
• Analysis does not require correction for multiple tests as in schemes using a
high number of surface-based statistical tests.
• Systematic integration of shape with clinical variables (age, duration of illness
and drug effect)  Leads to intuitive answers in regard to treatment effects
and longitudinal changes.
• Significant hippocampal deformation difference between schizophrenics and
controls (p<0.0091). Deformation increases with age (tail/head)
• Drug treatment groups (Atypical, Typical) show significant differences of Width
Asymmetry measure. Decreasing asymmetry of width with age is found.
• Deformation and width asymmetry changes in controls analyzed over a ten
years interval are not significant.
• Conclusions w.r.t. age changes from cross-sectional study: Longitudinal study.
• Exploratory analysis: Requires confirmatory analysis in independent study.
Supported by NCI grant P01 CA47982 and the UNC-MHNCRC (MH33127).
MICCAI Nov. 2003