Transcript Tatiana Foroud and Peter Hammond

Craniofacial Dysmorphism &
Fetal Alcohol Exposure
Tatiana Foroud & Peter Hammond (PIs)
Leah Wetherill, Mike Suttie
Update on Image Collection
Site
3D Images
(# subjects)
Plans for 2013
San Diego
UCLA/USC
Atlanta
Minneapolis
237 (215)
52 (52)
147 (147)
50 (50)
Camera on location
Camera brought to site
Camera brought to site
Camera brought to site
Ukraine
47 (34)
Camera to be put at location
(training March/April 2013)
Study visit scheduled
(Oct 1-12, 2013)
No longer a site
Camera on location
South Africa
223 (223)
(Jacobson)
South Africa (May)
37 (37)
South Africa (PASS) 1,738 (1,260)
Update on Image Numbers
160
140
Caucasian
African American
Other
Unknown
# individuals
120
100
80
60
40
20
0
control
exposed
fas
unknown
Coordinating Image Collection
• For sites without a dedicated camera, will
subjects be brought in for
Dysmorphology?
– Do we need to develop alternate plans to
ensure image collection?
– Can we collect longitudinal images at these
sites?
Update on Saliva Collection
Site
Saliva
Plans for 2013
San Diego
UCLA/USC
Atlanta
131
45
28
Continue to collect
Continue to collect
Continue to collect
Minneapolis
Ukraine
28
0 (no approval)
Continue to collect
No plans to initiate
South Africa
225
(Jacobson)
South Africa (PASS) 0 (collected as part of
parent study)
Only collect if new
subjects
No need to initiate
Saliva Collection
• Collect saliva from all subjects
– Obtain additional DNA for future studies
– Appropriate since amount of DNA obtained has varied
Use of DNA for Research
• GWAS completed in 240 individuals
– 700,000 SNPs genotyped
• Variable alcohol exposure and phenotype
– Data used for candidate gene studies
– Genotype x alcohol interaction
Zebrafish Model: Gene Pathways
• Collaboration with Johann Eberhart (Pilot
Study)
- Platelet-Derived Growth Factor (PDGF)
 5 genes
- Fibroblast Growth Factor (FGF)
 4 genes
- Bone Morphogenetic Protein (BMP)
 2 genes
• Examine 5 facial phenotypes related to
craniofacial abnormalities seen in the
zebrafish model
PDGFRB and midfacial depth
Use of DNA for Research
• DNA will be made available to Dipak
Sarkar (Pilot Study)
• Plans in place to perform another GWAS
or similar technology in later years of the
grant
• Remain in contact with PASS investigators
(Ingrid Holm) to ensure coordination with
genetic studies
Collaboration with PASS
• Submitted request for data from DM-STAT
for PASS subjects with images
• Received growth measurements which
will be used in initial analyses
• Cannot receive alcohol exposure data for
several years
PROGRESS ON 4 OTHER OBJECTIVES
•
Develop a screening tool that would utilize the data from the 3D facial
images and could be widely used to accurately identify individuals with
a high likelihood of alcohol exposure
•
Recruit and analyze facial imaging data from very young populations to
develop a screening tool that accurately identifies high risk individuals
for future intervention
•
Combine face images, neurobehavioral data and brain images to
identify common pathways and hence improve diagnosis of prenatal
alcohol exposure
•
Extend existing and develop novel techniques and associated software
to cope with demands of larger datasets and more diverse comparison
of controls, alcohol exposed and other developmentally delayed
subjects while accommodating multiple anatomical images per subject
PROGRESS ON 4 OTHER OBJECTIVES
•
Develop ascreening tool that would utilize the data from the 3D facial
images and could be widely used to help
identify individuals
with a high likelihood of alcohol exposure
•
Recruit and analyze facial imaging data from very young populations
to develop a screeniand tool that accurately identifyie high risk
individuals for future intervention (PASS/UKRAINE)
•
Combine face images, neurobehavioral data and brain datas to identify
common pathways and hence improve diagnosis of prenatal alcohol
exposure
•
Extend existing and develop novel techniques and associated software
to cope with demands of larger datasets and more diverse comparison
of controls, alcohol exposed and other developmentally delayed
subjects while accommodating multiple anatomical images per subject
VISUALISING INDIVIDUAL
FACIAL DYSMORPHISM
DYNAMIC MORPH
between individual &
average of matched
controls
highlights facial
dysmorphism
(Suttie et al, 2013: Pediatrics, in press)
FACE SIGNATURE
quanitifies facial
dysmorphism
Redcontracted
Blueexpanded
Greencoincident
FACIAL DYSMORPHISM ACROSS FASD
(Suttie et al, 2013: Pediatrics, in press)
EXAMPLES OF UPPER LIP SIGNATURES
FAS
PFAS
±1.0
SD
±1.5
SD
±2.0
SD
(Suttie et al, 2013: Pediatrics, in press)
RECOGNITION OF FAS FACIAL FEATURES
probability of correctly classifying two individuals, one taken from
each of the two groups being compared
HC vs FAS
HC vs FAS+PFAS
CM
LDA
SVM
CM
LDA
SVM
Face
0.967
0.967
1.00
0.892
0.909
0.909
Periorbit
0.983
0.917
0.967
0.892
0.900
0.892
Perioral
0.850
0.850
0.884
0.883
0.883
0.883
Perinasal
0.833
0.850
0.934
0.825
0.825
0.817
Profile
0.933
0.933
0.917
0.925
0.933
0.917
(Suttie et al, 2013: Pediatrics, in press)
FACE SIGNATURE GRAPH
clusters individuals with similar face signatures
Children with FAS
clustered together
(boxed faces)
Children without FAS
clustered together
Redcontracted
Blueexpanded
Greencoincident
(Suttie et al, 2013: Pediatrics, in press)
DETECTION OF HE FACIAL DIFFERENCES
presence/absence of FAS like facial differences in HE group
concur with neurobehavioral differences
FAS+PFAS
HE1 – FAS/PFAS like facial differences
HE2 – control like facial differences
FAS
PFAS
HE1
HE2
HC
HE1 vs
HE2 (t)
WISC
IQ
65.4
63.0
65.5
73.3
73.3
-1.80†
CVLT-C
test1
42.7
41.5
40.0
47.3
45.8
-2.02*
CVLT-C
test2
88.5
88.3
84.3
93.7
93.2
-1.89†
†p
(Suttie et al, 2013: Pediatrics, in press)
< 0.08
*p
<0.05
PRELIMINARY ANALYSIS OF
US CAUCASIAN COHORT
Group
HC
HE
FAS/PFAS
US
Caucasian
n mean age
74
11.8
42
12.3
17
13.6
SA Cape
coloured
n mean age
69
10.1
75
10.4
48
10.3
FAS/HE FACE SIGNATURES
FAS
SMS54
SMS130
ESL10991
HE
CAUCASIAN FACIAL GROWTH (PC1)
Facial Growth (PC1)
3
2
1
FAS
Control
HE
SMS130
ESL10991
SMS54
0
-1
-2
AGE
-3
8
9
10
11
12
13
14
15
16
17
18
CAUCASIAN HC VS FAS
unblinded closest mean classification
9
FAS
HC
8
SMS54
7
6
ESL10991
5
4
-3
-2
-1
0
1
2
3
CAUCASIAN HC VS FAS
blinded classification of 3 FAS
9
SMS130
FAS
HC
8.5
8
7.5
SMS54
7
6.5
6
ESL10991
5.5
5
4.5
4
-3
-2
-1
0
1
2
3
DISCRIMINATION TESTING
blinded drop-1-out
FAS vs HC
Sens
Spec
Accuracy
Reduced face
76.5
76.9
76.8
Thin Profile
70.6
84.6
81.2
Mouth
70.6
94.2
88.4
SIGNATURE GRAPHS FOR FAS AND HE
ESL10991
SMS54
SMS130
ESL10991
SMS54
SMS130
CONTRIBUTION TO MOUSE PROJECT
Lipinski et al, 2012: PLoS ONE 7(8)
MULTIPLE COMPONENT MODELLING
-For every image:
Landmarks
Separated regions in 3D space
-For every region we have at least 1
landmark
1.
2.
3.
4.
5.
Select landmark
Use landmark as seed point
Analyse connectivity
Warping (TPS) algorithm now
fed individual 3D region
Components put back together
for PCA
Multi Atlas Segmentation applied to
in vivo mouse brain MRI
•
•
•
•
Manual segmentation
Labor intensive
Requires training and prior knowledge
Inter-operater variability
Multi Atlas Segmentation applied to
in vivo mouse brain MRI
Da Ma1,2, M. Jorge Cardoso1, Marc Modat1, Nick Powell1,2, Holly Holmes2,
Mark Lythgoe2, and Sébastien Ourselin1,3
1
Centre for Medical Imaging Computing, 2 Centre for Advanced Biomedical Imaging,
3 Dementia Research Centre, University College London, UK
AUTOMATIC STRUCTURAL
PARCELLATION PIPELINE
Non-uniform intensity normalization (N3) [1]
Create brain mask (MAPS) [2]
Image registrations
Single-Atlas
segmentation propagation
Multi-Atlas
Segmentation Propagation
[1] Sled et al. 1998 IEEE transactions on medical imaging
[2] Leung et al. 2011 NeuroImage
Parameter
optimisation
Face screening tools –
more details tomorrow