Towards model-based diagnostics of human brain pathophysiology

Kay H. Brodersen1,2 · Klaas E. Stephan1,3

1 Department of Economics, University of Zurich, Switzerland 2 Department of Computer Science, ETH Zurich, Switzerland 3 Wellcome Trust Centre for Neuroimaging, University College London, UK

2

Schizophrenia, depression, mania, etc.

diverse genetic basis, strong gene-environment interactions genetically based diagnoses impossible

multiple pathophysiological mechanisms even when symptoms are similar, causes can differ across patients

variability in treatment response and outcome

Consequences?

need to infer on pathophysiological mechanisms in individual patients!

Psychiatric spectrum diseases

3

Model-based inference on individual pathophysiology

model-based

diagnostic tests

application to brain activity data from individual patients

model of neuronal (patho)physiology

spectrum disease

type 2

type 3

type 1

diagnostic classification

treatment X

treatment Y

treatment Z

4

Model-based analyses

How do patterns of hidden quantities (e.g., connectivity among brain regions) differ between groups?

Towards model-based diagnostic tests

Structure-based analyses

Which anatomical structures allow us to separate patients and healthy controls?

Activation-based analyses

Which functional differences allow us to separate groups?

5

From models of pathophysiology to clinical applications

Developing models of (patho)physiological processes

• neuronal: synaptic plasticity, neuromodulation • computational: learning, decision making

Validation studies in animals & humans

• can models detect experimentally induced changes, e.g., specific changes in synaptic plasticity?

Clinical validation studies & translation

• clinical validation of classifications • predicting diagnosis, therapeutic response, outcome

x1 x2

x3

CuxDxBuAdt

dx n

j

j

j

m

i

i

i

1

)(

1

)(

u1

u2

-10

0

10

-0.5

0

0.5

-0.1

0

0.1

0.2

0.3

0.4

-0.4

-0.2

0 -0.5

0

0.5-0.4

-0.35

-0.3

-0.25

-0.2

-0.15

-10

0

10

-0.5

0

0.5

-0.1

0

0.1

0.2

0.3

0.4

-0.4

-0.2

0 -0.5

0

0.5-0.4

-0.35

-0.3

-0.25

-0.2

-0.15

gen

era

tive

em

bed

din

g

L.H

G

L.H

G

Vo

xel (

64,-

24,4

) m

m

L.MGB L.MGBVoxel (-42,-26,10) mm

Voxel (-56,-20,10) mm R.HG L.HG

controlspatients

Voxel-based feature space Generative score space

patients controls

6

From models of pathophysiology to clinical applications

Developing models of (patho)physiological processes

• neuronal: synaptic plasticity, neuromodulation • computational: learning, decision making

Validation studies in animals & humans

• can models detect experimentally induced changes, e.g., specific changes in synaptic plasticity?

Clinical validation studies & translation

• clinical validation of classifications • predicting diagnosis, therapeutic response, outcome

x1 x2

x3

CuxDxBuAdt

dx n

j

j

j

m

i

i

i

1

)(

1

)(

u1

u2

-10

0

10

-0.5

0

0.5

-0.1

0

0.1

0.2

0.3

0.4

-0.4

-0.2

0 -0.5

0

0.5-0.4

-0.35

-0.3

-0.25

-0.2

-0.15

-10

0

10

-0.5

0

0.5

-0.1

0

0.1

0.2

0.3

0.4

-0.4

-0.2

0 -0.5

0

0.5-0.4

-0.35

-0.3

-0.25

-0.2

-0.15

gen

era

tive

em

bed

din

g

L.H

G

L.H

G

Vo

xel (

64,-

24,4

) m

m

L.MGB L.MGBVoxel (-42,-26,10) mm

Voxel (-56,-20,10) mm R.HG L.HG

controlspatients

Voxel-based feature space Generative score space

patients controls

7

Model-based classification by generative embedding

step 2 — kernel construction

step 1 — model inversion

measurements from an individual subject

subject-specific inverted generative model

subject representation in the generative score space

A → B

A → C

B → B

B → C

A

C B

step 3 — analysis

separating hyperplane to discriminate between groups

A

C B

jointly discriminative connection strengths

step 4 — interpretation

-2 0 2 4 6 8-1

0

1

2

3

4

5

Voxel 1

Voxe

l 2

-0.4 -0.35 -0.3 -0.25 -0.2 -0.15-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

(1) L.MGB -> L.MGB

(14)

R.H

G -

> L

.HG

Brodersen et al. (2011) NeuroImage; Brodersen et al. (2011) PLoS Comput Biol

8

activity 𝑧1(𝑡)

Choosing a generative model: DCM for fMRI

intrinsic connectivity direct inputs

modulation of connectivity

neural state equation

haemodynamic forward model 𝑥 = 𝑔(𝑧, 𝜃ℎ)

BOLD signal

neuronal states

t

driving input 𝑢1(𝑡) modulatory input 𝑢2(𝑡)

t

activity 𝑧2(𝑡)

activity 𝑧3(𝑡)

signal 𝑥1(𝑡)

signal 𝑥2(𝑡)

signal 𝑥3(𝑡)

Friston, Harrison & Penny (2003) NeuroImage Stephan & Friston (2007) Handbook of Brain Connectivity

𝑧 = 𝐴 + ∑𝑢𝑗𝐵𝑗 𝑧 + 𝐶𝑢

9

Example: diagnosing stroke patients

L R

Brodersen et al. (2011) NeuroImage; Brodersen et al. (2011) PLoS Comput Biol

anatomical regions of interest

y = –26 mm

10

Example: diagnosing stroke patients

MGB

PT

HG (A1)

MGB

PT

HG (A1)

stimulus input

L R

Brodersen et al. (2011) NeuroImage; Brodersen et al. (2011) PLoS Comput Biol

11

Multivariate analysis: connectional fingerprints

patients controls

12

16218917734729133230781 24338936050

60

70

80

90

100

bala

nced

accu

racy

Classification performance

Activation-based analyses a anatomical feature selection c mass-univariate contrast feature selection s locally univariate searchlight feature selection p PCA-based dimensionality reduction

Correlation-based analyses m correlations of regional means e correlations of regional eigenvariates z Fisher-transformed eigenvariates correlations

Model-based analyses o gen.embed., original full model f gen.embed., less plausible feedforward model l gen.embed., left hemisphere only r gen.embed., right hemisphere only

activation- based

correlation- based

model- based

a c s p m e z o f l r

bal

ance

d a

ccu

racy

100%

50%

90%

80%

70%

60%

n.s. n.s.

*

Brodersen et al. (2011) PLoS Comput Biol

13

Full Bayesian approach to performance evaluation

Brodersen, Mathys, Chumbley, Daunizeau, Stephan (in preparation)

Full Bayesian

mixed-effects inference

(beta-binomial model)

𝑘𝑗+ 𝑘𝑗

−

𝜋𝑗+ 𝜋𝑗

−

𝛼+,𝛽+ 𝛼−,𝛽−

𝑘𝑗+ 𝑘𝑗

−

𝜌𝑗

𝜇,Σ

𝑗 = 1…𝑚

Bin 𝑘𝑗− 𝜎 𝜌𝑗,2 ,𝑛𝑗

− Bin 𝑘𝑗

+ 𝜎 𝜌𝑗,1 ,𝑛𝑗+

𝒩2 𝜌𝑗 𝜇, Σ

𝒩 𝜇 𝜇0, Σ/𝜅0

Bin 𝑘𝑗+ 𝜋𝑗

+, 𝑛𝑗+

Bin 𝑘𝑗− 𝜋𝑗

−, 𝑛𝑗−

Beta 𝜇𝑗−|𝛼−,𝛽− Beta 𝜇𝑗

+|𝛼+,𝛽+

𝑝 𝛼−, 𝛽− 𝑝 𝛼+, 𝛽+

𝑗 = 1…𝑚

Inv-Wish𝑣0 Σ|Λ0−1

Full Bayesian

mixed-effects inference

(normal-binomial model)

14

The generative projection

-10

0

10

-0.5

0

0.5

-0.1

0

0.1

0.2

0.3

0.4

-0.4

-0.2

0 -0.5

0

0.5-0.4

-0.35

-0.3

-0.25

-0.2

-0.15

-10

0

10

-0.5

0

0.5

-0.1

0

0.1

0.2

0.3

0.4

-0.4

-0.2

0 -0.5

0

0.5-0.4

-0.35

-0.3

-0.25

-0.2

-0.15generative

emb

edd

ing

L.H

G →

L.H

G

Vo

xel (

64

,-2

4,4

) m

m

L.MGB → L.MGB Voxel (-42,-26,10) mm Voxel (-56,-20,10) mm R.HG → L.HG

controls

patients

Voxel-based activity space Model-based parameter space

Brodersen et al. (2011) PLoS Comput Biol

15

Discriminative features in model space

MGB

PT

HG (A1)

MGB

PT

HG (A1)

stimulus input

L R

16

Discriminative features in model space

MGB

PT

HG (A1)

MGB

PT

HG (A1)

stimulus input

L R

highly discriminative somewhat discriminative not discriminative

17

Summary

-10

0

10

-0.5

0

0.5

-0.1

0

0.1

0.2

0.3

0.4

-0.4

-0.2

0 -0.5

0

0.5-0.4

-0.35

-0.3

-0.25

-0.2

-0.15

-10

0

10

-0.5

0

0.5

-0.1

0

0.1

0.2

0.3

0.4

-0.4

-0.2

0 -0.5

0

0.5-0.4

-0.35

-0.3

-0.25

-0.2

-0.15

gen

era

tive

em

bed

din

g

L.H

G

L.H

G

Vo

xel (6

4,-

24,4

) m

m

L.MGB L.MGBVoxel (-42,-26,10) mm

Voxel (-56,-20,10) mm R.HG L.HG

controlspatients

Voxel-based feature space Generative score space

patients controls

x1 x2

x3

Non-invasive, model-based assays of human brain pathophysiology

Quantitative diagnostics

Patient A Patient B

Individualized treatment

18

Translational Neuromodelling Unit (TNU) @ IBT

Prof. Dr. Dr. med. Klaas E. Stephan

Laboratories for multimodal studies

of brain activity

Ultra high field MRI

Clinical unit for patient recruitment

Validation & Translation of Computational Models of

Brain Pathophysiology

Development of computational

models & software

19

Colleagues & collaborators*

Klaas E. Stephan’s group in Zurich

K. Brodersen J. Daunizeau A. Diaconescu S. Iglesias L. Kasper F. Lieder E. Lomakina C. Matthys M. Piccirelli

London (FIL/ION)

R. Dolan K. Friston M. Garrido A. Leff R. Moran W. Penny

MPI Cologne Germany

H. Endepols R. Graf M. Tittgemeyer

Elsewhere

M. Breakspear H. den Ouden L. Harrison

Zurich (ETH/UZH)

J. Buhmann F. Helmchen K. Prüssmann F. Vollenweider B. Weber

Supported by • Branco Weiss Foundation • European Union FP7 • NCCR "Neural Plasticity" • SystemsX – The Swiss Systems Biology

Initiative • URPP "Foundations of Human Social

Behaviour„ • Zurich Neuroscience Centre (ZNZ)

* listed alphabetically