The Human Connectome Project multimodal cortical parcellation: new avenues for brain research.
1. The Human Connectome Project
multimodal cortical parcellation:
new avenues for brain research.
Dr Emma. C. Robinson
Biomedical Engineering, Kings College
emma.robinson@kcl.ac.uk,
emmarobinson01.com
Twitter: @emrobSci
3. Overview
1. The Human Connectome Project (HCP)
Goals
Neuroimaging approach (surface-based processing)
4. Overview
1. The Human Connectome Project (HCP)
Goals
Neuroimaging approach (surface-based processing)
2.The HCP v1 multi-modal parcellation:
Prospects and Future Challenges
5. Overview
1. The Human Connectome Project (HCP)
Goals
Neuroimaging approach (surface-based processing)
2.The HCP v1 multi-modal parcellation:
Prospects and Future Challenges
3.Translating HCP processing to developing data:
The dHCP project
6. The Human Connectome Project (HCP)
Goal: Build the most accurate model to date
of the adult* human Connectome
Capture high spatial and temporal
resolution functional, diffusion
and structural MRI
Deliver enhanced image
processing pipelines and methods
Improve understanding of the
functional organisation of the
human brain
7. The Human Connectome Project (HCP)
Goal: Build the most accurate model to date
of the adult* human Connectome
Capture high spatial and temporal
resolution functional, diffusion
and structural MRI
Deliver enhanced image
processing pipelines and methods
Improve understanding of the
functional organisation of the
human brain
8. The Human Connectome Project (HCP)
Goal: Build the most accurate model to date
of the adult* human Connectome
Capture high spatial and temporal
resolution functional, diffusion
and structural MRI
Deliver enhanced image
processing pipelines and methods
Improve understanding of the
functional organisation of the
human brain
* the HCP is now acquiring new data sets across the lifespan
9. The Human Connectome Project (HCP)
HCP young adult study:
1206 healthy adult subjects (aged 22-35yrs)
Twins and non-twin siblings
Acquisitions:
0.7mm T1 and T2
1hr resting state functional MRI (rfMRI)
7 tasks fMRI experiments including: language,
emotional, relational, gambling, motor, social
cognition, working memory
HARDI diffusion data
***COMPLETED***
•
10. The HCP’s neuroimaging approach
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
11. The HCP’s neuroimaging approach
7 Tenets:
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
12. The HCP’s neuroimaging approach
7 Tenets:
1. Collect multimodal imaging data from many subjects
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
13. The HCP’s neuroimaging approach
7 Tenets:
1. Collect multimodal imaging data from many subjects
2. Acquire data at high spatial and temporal resolution
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
14. The HCP’s neuroimaging approach
7 Tenets:
1. Collect multimodal imaging data from many subjects
2. Acquire data at high spatial and temporal resolution
3. Preprocess data to minimize distortions, blurring and temporal
artifacts
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
15. The HCP’s neuroimaging approach
7 Tenets:
1. Collect multimodal imaging data from many subjects
2. Acquire data at high spatial and temporal resolution
3. Preprocess data to minimize distortions, blurring and temporal
artifacts
4. Perform cortical surface-based processing
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
16. The HCP’s neuroimaging approach
7 Tenets:
1. Collect multimodal imaging data from many subjects
2. Acquire data at high spatial and temporal resolution
3. Preprocess data to minimize distortions, blurring and temporal
artifacts
4. Perform cortical surface-based processing
5. Accurately align corresponding brain areas across subjects and
studies - surface-based alignment
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
17. The HCP’s neuroimaging approach
7 Tenets:
1. Collect multimodal imaging data from many subjects
2. Acquire data at high spatial and temporal resolution
3. Preprocess data to minimize distortions, blurring and temporal
artifacts
4. Perform cortical surface-based processing
5. Accurately align corresponding brain areas across subjects and
studies - surface-based alignment
6. Analyze data using neurobiologically accurate brain
parcellations
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
18. The HCP’s neuroimaging approach
7 Tenets:
1. Collect multimodal imaging data from many subjects
2. Acquire data at high spatial and temporal resolution
3. Preprocess data to minimize distortions, blurring and temporal
artifacts
4. Perform cortical surface-based processing
5. Accurately align corresponding brain areas across subjects and
studies - surface-based alignment
6. Analyze data using neurobiologically accurate brain
parcellations
7. Open data-sharing via user-friendly databases.
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
19. The HCP’s neuroimaging approach
7 Tenets:
1. Collect multimodal imaging data from many subjects
2. Acquire data at high spatial and temporal resolution
3. Preprocess data to minimize distortions, blurring and temporal
artifacts
4. Perform cortical surface-based processing
5. Accurately align corresponding brain areas across subjects and
studies - surface-based alignment
6. Analyze data using neurobiologically accurate brain
parcellations
7. Open data-sharing via user-friendly databases.
Glasser, Matthew F., et al. "The human connectome project's neuroimaging approach." Nature Neuroscience 19.9 (2016): 1175-1187.
21. Cortical Surface-based image
processing
• HCP cortical data is projected to the cortical surface for
two reasons:
Volumetric smoothing mixes
signals
Surface-constrained smoothing
averages only GM signals
1. Surface based smoothing improves SNR
22. Cortical Surface-based image
processing
• HCP cortical data is projected to the cortical surface for
two reasons:
1. Surface based smoothing improves SNR
2. Surface-based registration improves alignment of cortical
folds
Small mis-registraions in 3D can have a large
impact on the alignment of the cortical sheet
25. Surface extraction
The HCP uses a refinement of the FreeSurfer Pipeline
• Higher resolution
• Pial surface
extraction uses both
T1 and T2
26. Surface extraction
The HCP uses a refinement of the FreeSurfer Pipeline
TISSUE
SEGMENTATION
MESH
EXTRACTION
FEATURE
PROJECTION*
*Using partial volume
weighted ribbon-constrained
volume to surface mapping
27. Multimodal cortical features from the HCP
Curvature
Task fMRI
Myelin
Structural Connectivity
Sotiropoulos et al
NeuroImage (In press)
Functional Connectivity
Glasser, 2011. J.
Neurosci, 31
11597-11616
28. CIFTI: A new file format
for surface imaging data
29. CIFTI: A new file format
for surface imaging data
• Surface AND Volume data
“Grayordinates”
30. CIFTI: A new file format
for surface imaging data
• Surface AND Volume data
“Grayordinates”
• Compact representation:
Left/right hemispheres
(without medial wall)
Deep grey-matter only
Natural sub-space for
representation of fMRI
32. Limitations of morphological
alignment
• Cortical folding patterns
are highly variable across
subjects
• Example: Cingulate (blue
arrows)
• Some subjects have one
fold where others have two
• Courtesy of Van Essen,
NeuroImage 28 (2005) 635
– 662
33. Limitations of morphological
alignment
• Alignment by cortical folds can lead to high residual
variability of functional regions across subjects
Nenning et al, Neuroimage 2017
34. Limitations of morphological
alignment
• Alignment by cortical folds can lead to high residual
variability of functional regions across subjects
Nenning et al, Neuroimage 2017
Solution - drive surface alignment with ‘areal’ features
such as rfMRI/tfMRI, cortical myelination
35. Surface-based alignment: MSM
• Spherical framework for cortical surface registration
• Use low resolution control point grids to constrain the
deformation
• Optimised using discrete methods
• Modular
36. Surface-based alignment: MSM
• Spherical framework for cortical surface registration
• Use low resolution control point grids to constrain the
deformation
• Optimised using discrete methods
• Modular
37. Surface-based alignment: MSM
• Spherical framework for cortical surface registration
• Use low resolution control point grids to constrain the
deformation
• Optimised using discrete methods
• Modular
data cost: i.e. correlation, MNI, SSD
Regularisation cost to encourage smoother warp
39. • Alignment driven
multivariate feature
vectors
myelin (M)
rfMRI (R)
visuotopic (V))
M
R {
V {
MSM: improves alignment of areal
features
40. • Alignment driven
multivariate feature
vectors
myelin (M)
rfMRI (R)
visuotopic (V))
• Improves alignment
of task activations
and correspondence
across functional
networks
Smith, Stephen M., et al. "Functional connectomics from resting-state fMRI." Trends in cognitive sciences 17.12 (2013): 666-682.
MSM: improves alignment of areal
features
41. • Alignment driven
multivariate feature
vectors
myelin (M)
rfMRI (R)
visuotopic (V))
• Improves alignment
of task activations
and correspondence
across functional
networks
MSM: improves alignment of areal
features
Robinson, Emma C., et al. "MSM: a new flexible framework for multimodal surface matching." Neuroimage 100 (2014): 414-426.
43. • Recent improvements to MSM
regularisation:
Robinson, Emma C., et al. "Multimodal
surface matching with higher-order
smoothness constraints."
NeuroImage (2017)
MSM: improves alignment of areal
features
44. • Recent improvements to MSM
regularisation:
Robinson, Emma C., et al. "Multimodal
surface matching with higher-order
smoothness constraints."
NeuroImage (2017)
• Even greater improvements in
alignment 0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
%improvementoverFS
FS
SD
MSMSulc
MSMAll sMSMP AIR
MSMAll sMSMST R
MSM: improves alignment of areal
features
45. • Recent improvements to MSM
regularisation:
Robinson, Emma C., et al. "Multimodal
surface matching with higher-order
smoothness constraints."
NeuroImage (2017)
• Even greater improvements in
alignment 0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
%improvementoverFS
FS
SD
MSMSulc
MSMAll sMSMP AIR
MSMAll sMSMST R
MSM: improves alignment of areal
features
• Lower peak
distortions
46. • Recent improvements to MSM
regularisation:
Robinson, Emma C., et al. "Multimodal
surface matching with higher-order
smoothness constraints."
NeuroImage (2017)
• Even greater improvements in
alignment 0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
%improvementoverFS
FS
SD
MSMSulc
MSMAll sMSMP AIR
MSMAll sMSMST R
MSM: improves alignment of areal
features
47. • Recent improvements to MSM
regularisation:
Robinson, Emma C., et al. "Multimodal
surface matching with higher-order
smoothness constraints."
NeuroImage (2017)
• Even greater improvements in
alignment 0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
%improvementoverFS
FS
SD
MSMSulc
MSMAll sMSMP AIR
MSMAll sMSMST R
MSM: improves alignment of areal
features
Full method behind HCP multimodal parcellation
Increased robustness in the face of (sometimes
extreme) variation in structural and function
organisation across subjects
49. Topological variation in the human
brain
Evidence from folding
Van Essen, David C. "A population-average, landmark-and surface-
based (PALS) atlas of human cerebral cortex." Neuroimage 28.3
(2005): 635-662.
50. Topological variation in the human
brain
Topological variability in the
human brain
A
A
C
C
B
B
Group 1
Group 2
Evidence from folding
Van Essen, David C. "A population-average, landmark-and surface-
based (PALS) atlas of human cerebral cortex." Neuroimage 28.3
(2005): 635-662.
And function
Amunts, K., A. Schleicher, and K. Zilles. "Cytoarchitecture of the
cerebral cortex—more than localization." Neuroimage 37.4 (2007):
1061-1065.
Wang D, Buckner RL, Fox MD, Holt DJ, Holmes AJ, Stoecklein S,
Langs G, Pan R, Qian T, Li K, Baker JT. Parcellating cortical
functional networks in individuals. Nature neuroscience.
2015;18(12):1853.
Gordon EM, Laumann TO, Adeyemo B, Petersen SE. Individual
variability of the system-level organization of the human brain.
Cerebral Cortex 2017;27(1):386-99.
Glasser, Matthew F., et al. "A multi-modal parcellation of human
cerebral cortex." Nature (2016).
51. Topological variation in the human
brain
Topological variability in the
human brain
A
A
C
C
B
B
Group 1
Group 2
Evidence from folding
Van Essen, David C. "A population-average, landmark-and surface-
based (PALS) atlas of human cerebral cortex." Neuroimage 28.3
(2005): 635-662.
And function
Amunts, K., A. Schleicher, and K. Zilles. "Cytoarchitecture of the
cerebral cortex—more than localization." Neuroimage 37.4 (2007):
1061-1065.
Wang D, Buckner RL, Fox MD, Holt DJ, Holmes AJ, Stoecklein S,
Langs G, Pan R, Qian T, Li K, Baker JT. Parcellating cortical
functional networks in individuals. Nature neuroscience.
2015;18(12):1853.
Gordon EM, Laumann TO, Adeyemo B, Petersen SE. Individual
variability of the system-level organization of the human brain.
Cerebral Cortex 2017;27(1):386-99.
Glasser, Matthew F., et al. "A multi-modal parcellation of human
cerebral cortex." Nature (2016).
52. The HCP Multi-modal Parcellation
• Regional boundaries found by looking for imaging
gradients in group average data
• Looking for patterns common across multiple modalities
• Informed by the literature where available
53. The HCP Multi-modal Parcellation
• Regional boundaries found by looking for imaging
gradients in group average data
• Looking for patterns common across multiple modalities
• Informed by the literature where available
54. The HCP Multi-modal Parcellation
• Expert manual annotations of 180 functionally specialised
regions (per hemisphere) on group average data
• 97 entirely new areas
• 83 areas previously reported by histological studies
55. The HCP Multi-modal Parcellation
• Expert manual annotations of 180 functionally specialised
regions (per hemisphere) on group average data
• 97 entirely new areas
• 83 areas previously reported by histological studies
56. The HCP Multi-modal Parcellation:
propagating the result to individuals
• Single subject parcellations were then obtained by training
MLP classifiers
• Binary classifications
• Group average data propagated to training subjects
Hacker, Carl D., et al. "Resting state network estimation in individual subjects." Neuroimage 82 (2013): 616-633.
used to train classifier ONLY where subject data
closely agrees with group
58. • Training data = 110 D
feature vectors
• Cortical thickness
• Cortical myelin
• Cortical curvature
• 20 task ICA + mean
• 77 rest ICA
• 5 hand engineered
‘visuotopic’ features
Features
59. • Training data = 110 D
feature vectors
• Cortical thickness
• Cortical myelin
• Cortical curvature
• 20 task ICA + mean
• 77 rest ICA
• 5 hand engineered
‘visuotopic’ features
Features
60. • Training data = 110 D
feature vectors
• Cortical thickness
• Cortical myelin
• Cortical curvature
• 20 task ICA + mean
• 77 rest ICA
• 5 hand engineered
‘visuotopic’ features
Features
61. • Training data = 110 D
feature vectors
• Cortical thickness
• Cortical myelin
• Cortical curvature
• 20 task ICA + mean
• 77 rest ICA
• 5 hand engineered
‘visuotopic’ features
Features
62. • Training data = 110 D
feature vectors
• Cortical thickness
• Cortical myelin
• Cortical curvature
• 20 task ICA + mean
• 77 rest ICA
• 5 hand engineered
‘visuotopic’ features
• Trained on 210 subject
training set
Features
63. • Training data = 110 D
feature vectors
• Cortical thickness
• Cortical myelin
• Cortical curvature
• 20 task ICA + mean
• 77 rest ICA
• 5 hand engineered
‘visuotopic’ features
• Trained on 210 subject
training set
• Validated on 210 subject
validation set
Features
64. The HCP Multi-modal Parcellation:
propagating the result to individuals
• Output from Classifier for 4 example datasets
Group Average
Classifier results for 4 subjects
65. The HCP Multi-modal Parcellation:
propagating the result to individuals
• Output from Classifier for 4 example datasets
Group Average
Classifier results for 4 subjects
67. The HCP Multi-modal Parcellation:
Accurate detection of regions across validation subjects
Top = Training Set; Bottom = Test Set
Darker orange indicates regions that were not detected in all subjects
(or were detected by with very low surface areas)
68. The HCP Multi-modal Parcellation:
• high consistency in group average parcellation
between training and test sets
Top = manual annotation; Bottom = overlap of training and test set classifier results
Blue borders= Train set; Red borders= Test set; Purple=overlap
69. The HCP Multi-modal Parcellation:
Prospects
Improves statistical significance
70. The HCP Multi-modal Parcellation:
Prospects
Improves statistical significance
71. The HCP Multi-modal Parcellation:
Prospects
1. A more robust and accurate reference space
for general adult imaging studies
Enhances the sensitivity of statistical comparisons
Consistent with known patterns of cellular organisation
(cyto-architecture)
Consistent with known patterns of functional organisation
Generalisable to new subjects
Requires only MSM alignment and application of pre-
trained* MLP classifier
72. The HCP Multi-modal Parcellation:
Prospects
2.Potential to map to new populations e.g.
patients/developing neonates and fetuses
Must tune the HCP analysis pipelines to new data sets
Propagate HCP v1 parcellation to new data either through:
Surface registration
And/or retraining the MLP classifier
73. The HCP Multi-modal Parcellation:
Prospects
3. Predicting Cognition and Behaviour
• Prediction of age/gender/developmental outcome/disease
progression
• Using:
Classification
Regression
Unsupervised Learning
• HCP data comes with 280 behavioural and demographic
measures
74. The HCP Multi-modal Parcellation:
Prospects
2. Predicting Cognition and Behaviour
e.g. Canonical Correlation
Analysis (CCA)
Functional netmats vs
HCP demographics
Smith et al. Nature
NeuroScience 2015
Shared modes of variance
Complex relationships
between all 280
behaviours & all network
connections
75. The HCP Multi-modal Parcellation:
Prospects
2. Predicting Cognition and Behaviour
e.g. Canonical Correlation
Analysis (CCA)
Functional netmats vs
HCP demographics
Smith et al. Nature
NeuroScience 2015
Shared modes of variance
Complex relationships
between all 280
behaviours & all network
connections
76. The HCP Multi-modal Parcellation:
Prospects
2. Predicting Cognition and Behaviour
Same analysis applied to
HCP parcellation
Bijsterbosch, Janine Diane, et al. "The relationship between
spatial configuration and functional connectivity of brain
regions." bioRxiv (2017): 210195.
Regions most predictive of
the strongest CCA mode
77. HCP Fluid Intelligence Predictions
• 360*110 features
mean myelin/thickness/folding/
tfMRI/rfMRI per parcel
• R2 = 0.347
• Feature Importance mapped
back to the image space
Random Forest regression to predict fluid intelligence from
HCP features
78. HCP Fluid Intelligence Predictions
• 360*110 features
mean myelin/thickness/folding/
tfMRI/rfMRI per parcel
• R2 = 0.347
• Feature Importance mapped
back to the image space
L R
Random Forest regression to predict fluid intelligence from
HCP features
79. Future Challenges
What to do about topological variability?
• Does averaging (group ICA, during
annotation) obscure true variability
• Could different brains have different
numbers of parcels?
• How to compare across data sets if spatial
averaging no longer valid?
Q
80. Future Challenges
What to do about topological variability?
• Look for new ways to compare data that do
not rely on spatial averaging
• Revise methods for image registration that
break topological constraints
A…?
81. The Developing Human Connectome
Project (dHCP)
• Model dynamic, emerging
4D connectomes
~ 1500 fetuses, preterm and
term born neonates
• Multimodal imaging
HARDI
rfMRI
0.5mm3 (reconstructed) T1
& T2
• New surface extraction
pipeline
Brain extract
Bias correct T1T2
Align
White
PialMid-thickness Inflated Very Inflated Sphere
T1/T2w ratio
A. Pre-Processing
F
G
C
H
B
DE
I
I
Myelin MapSulcal DepthCurvatureThickness
Segmentation
Makropoulos and Robinson et al. The Developing Human Connectome Project: a Minimal Processing Pipeline for Neonatal Cortical Surface Reconstruction. bioRxiv, 125526.
82. Challenges of working
with developing data
• Developing data is affected by
Motion (severe cases account for < 2% )
Limited scan times
Relatively low resolution
spatio-temporal evolution
83. Challenges of working
with developing data
• Developing data is affected by
Motion (severe cases account for < 2% )
Limited scan times
Relatively low resolution
spatio-temporal evolution
84. Challenges of working
with developing data
• Developing data is affected by
Motion (severe cases account for < 2% )
Limited scan times
Relatively low resolution
spatio-temporal evolution
85. dHCP spatio-temporal atlases
• New volumetric and surface templates spanning 36-44
weeks gestation
Andreas Schuh et al.
Unbiased construction
of a temporally
consistent
morphological atlas of
neonatal brain
development
(in preparation)
86. dHCP spatio-temporal atlases
• New volumetric and surface templates spanning 36-44
weeks gestation
Jelena Bozek et al.
Construction of a Neonatal
Cortical Surface Atlas Using
Multimodal Surface
Matching in the Developing
Human Connectome Project
(under revision)
87. dHCP spatio-temporal atlases
• New volumetric and surface templates spanning 36-44
weeks gestation
Jelena Bozek et al.
Construction of a Neonatal
Cortical Surface Atlas Using
Multimodal Surface
Matching in the Developing
Human Connectome Project
(under revision)
88. dHCP spatio-temporal atlases
• New volumetric and surface templates spanning 36-44
weeks gestation
Jelena Bozek et al.
Construction of a Neonatal
Cortical Surface Atlas Using
Multimodal Surface
Matching in the Developing
Human Connectome Project
(under revision)
89. Finding trends in longitudinal
cortical development
• 10 preterm subjects
• Imaged twice at
• 32.7 ± 1.2 weeks
• 41.5 ± 1.6 weeks
• MSM generates
smooth maps of
deformation strain
over this time period
• Consistent and
statistically significant
across the population
Robinson, Emma C., et al. "Multimodal surface matching with
higher-order smoothness constraints." NeuroImage (2017).
SAS SSS
a)
e)
b)
x LABEL POINT
OPTIMAL LABEL
CONTROL POINT
c)
f)
d)
MSS
SSS + G
TSS
DAS TAS
g)
F
90. Finding trends in longitudinal
cortical development
• 10 preterm subjects
• Imaged twice at
• 32.7 ± 1.2 weeks
• 41.5 ± 1.6 weeks
• MSM generates
smooth maps of
deformation strain
over this time period
• Consistent and
statistically significant
across the population
Robinson, Emma C., et al. "Multimodal surface matching with
higher-order smoothness constraints." NeuroImage (2017).
91. New MSM: finding trends in
longitudinal cortical development
• 24 very preterm infants (born <30 weeks PMA, 15 male, 15 female)
• scanned 2-4 times before or at term-equivalent (36-40 weeks PMA)
Garcia, Kara E., et al. "Dynamic patterns of cortical expansion during folding of the preterm human brain." bioRxiv (2017): 185389.
92. Examples of dHCP rfMRI ICA
30 dimensional ICA estimated from 114 subjects imaged across 36-42
weeks PMA
93. Examples of dHCP rfMRI ICA
Predicting gestational age at scan
prediction train score 0.636331204848 test score 0.212042264421
94. Examples of dHCP dMRI
Visualisation of anatomical connections derived from multi-shell
high angular resolution diffusion data.
96. Conclusions
1. The HCP v 1.0 multi-modal parcellation is
Cytoarchitecturally accurate
Functionally consistent
Sensitive & Robust
2. Future iterations will
Map regions to diseased or developing populations
Fully account for topological variations in the data
3. HCP processing has inspired the dHCP
Leading to new insights wrt the functional and
morphological development of the neonatal cortex
97. • Prof. Daniel Rueckert
• Dr Bernhard Mainz
• Dr Ben Glocker
• Dr Antonis Makropoulos
• Dr Andreas Schuh
• Prof Jo Hajnal
• Prof David Edwards
• Prof Julia Schnabel
Acknowledgements
• Prof. David Van Essen
• Matthew Glasser
• Tim Coalson
• Dr Carl Hacker
• Kara Garcia
• Prof. Mark Jenkinson
• Prof. Steven Smith
• Prof. Saad Jbadi
• Dr Janine Bijsterbosch
• Dr Samuel Harrison