Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5...
Transcript of Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5...
![Page 1: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/1.jpg)
Dimitri Van De VilleMedical Image Processing Laboratory (MIP:lab)Institute of Bioengineering/Center for NeuroprostheticsEcole Polytechnique Fédérale de Lausanne
http://miplab.epfl.ch/
Juan-les-Pins / November 22, 2017
@dvdevill
Department of RadiologyUniversité de Genève
Functional connectivity models: from blobs to (dynamic) networks
![Page 2: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/2.jpg)
OverviewThe rise of (functional) MRIActivation mapping
Task-evoked activityBrain activation maps using confirmatory analysisGeneral Linear Model (GLM)
Connectivity mappingResting-state condition reveals intrinsic functional networksBlind source separation (independent component analysis)Statistical interdependencies and graph analysis
Dynamic connectivity mappingMoment to moment fluctuations of connectivityWindowed and event-based approaches
Open challenges2
![Page 3: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/3.jpg)
“the quest for an understanding of the functional organization of the [...] human brain,
using techniques to assess changes in brain circulation, [a search that has occupied]
mankind for more than a century”
Marcus Raichle, 1998
![Page 4: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/4.jpg)
4Siemens 3T Prisma MRI scanner @ Campus Biotech / picture (c) EPFL/Alban Kakulya
Widely deployed in hospitals and research centersEndogenous contrast mechanismNon-invasive imaging tool to study human brain anatomy and function
Magnetic resonance imaging (MRI)
Structural MRI Functional MRI
Single 3D volume1x1x1 mm3
takes couple of minutes
Series of 3D volumes3x3x3 mm3
20-30 slicesevery 2-4 secduring 5-10 minutes
![Page 5: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/5.jpg)
![Page 6: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/6.jpg)
[Buxton et al 1997; Friston et al. 1998, 2000; Iannetti & Wise, 2007]
FMRI blood-oxygenation-level-dependent (BOLD) signals are slow proxy for neuronal activity
Neurons in the visual cortex are active
(metabolic demand at synapses)
Neurovascular coupling: blood flow increases
locally to bring in glucose and oxygen
Oxygen is carried by hemoglobin in RBCs
Increase of ratio in oxygenated/deoxygenated
hemoglobin
Deoxygenated Hb is paramagnetic and drives
BOLD signal
![Page 7: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/7.jpg)
[Buxton et al 1997; Friston et al. 1998, 2000; Iannetti & Wise, 2007]
FMRI blood-oxygenation-level-dependent (BOLD) signals are slow proxy for neuronal activity
20 sec
hemodynamicresponse function
• Impulse response to short period of neural activation
• Notice the response timing: ~2 sec delay, 4-6 sec to peak, up to 20 sec back to baseline
![Page 8: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/8.jpg)
8
fMRI of evoked activity
visual stimuli
min 0 max
GLM / statistical testing
calcarine sulcus
movie accelerated 4 times
t-value
![Page 9: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/9.jpg)
fMRI of evoked and intrinsic activity
5% evoked activity
95% intrinsic activity
energy budget
![Page 10: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/10.jpg)
Condition of resting state: Conscious and unconscious brain activity happen without premeditation or external
stimulus
More profound change of viewpoint:From brain processing stimuli/performing tasks
to internal dynamics being modulated
![Page 11: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/11.jpg)
BOLD signal (PCC)
fMRI of spontaneous activity
resting-state scan (minimally preprocessed)
movie accelerated 4 times
min 0 maxchanges w.r.t. baseline
minimal compliance for patient studies!
![Page 12: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/12.jpg)
OverviewThe rise of (functional) MRIActivation mapping
Task-evoked activityBrain activation maps using confirmatory analysisGeneral Linear Model (GLM)
Connectivity mappingResting-state condition reveals intrinsic functional networksBlind source separation (independent component analysis)Statistical interdependencies and graph analysis
Dynamic connectivity mappingMoment to moment fluctuations of connectivityWindowed and event-based approaches
Open challenges12
![Page 13: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/13.jpg)
13[Biswal et al., 1995]
![Page 14: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/14.jpg)
14
Default-modenetwork
Auditory Visual
[Adapted from Raichle, TICS, 2010]
Somato-motor
Dorsalattention
Executivecontrol
Seed-based connectivity maps
![Page 15: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/15.jpg)
Cocktail problem: blind source separation
15[Gutierrez-Osuna, Introduction to Pattern Analysis]
![Page 16: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/16.jpg)
Mixing of spatial brain sources
16
instance offeature vector!spatial source
use of sourcesat time 1
use of sourcesat time 2
a11
a21
a22
a12
w11
w21
w22
w12
optimize unmixingbased on criterion of
statistical independence
network 1
network 2
mixture 1
mixture 2
... ...
fMRI: spatial ICA / EEG: temporal ICA
![Page 17: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/17.jpg)
Unmixed spatial brain sourcesICA reveals several large-scale brain networks (similar to task-evoked networks!)
17[Mantini et al, 2007]
Default-modenetwork
Visualnetwork
Somatosensorynetwork
Dorsal attentionnetwork
Auditorynetwork
Executive control/Saliency network
![Page 18: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/18.jpg)
Unmixing of brain regions from fMRI data
Compare against the classical GLM
estimated by OLS
estimated by ICA
Spatial ICA
18
Data (Y)
voxels
time
≈ Mixing matrix x Spatially independent
components (S)
Data (Y)
voxels
time
≈ Design matrix x Activation maps (beta)
![Page 19: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/19.jpg)
Graph analysisParcellation into brain regions based on atlasPairwise correlations between regionally-averaged timecourses of all brain regions
19[Richiardi et al, 2013]
brain regions
brain regions!
build connectivity matrix
![Page 20: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/20.jpg)
Graph analysisParcellation into brain regions based on atlasPairwise correlations between regionally-averaged timecourses of all brain regions
20[Richiardi et al, 2013]
brain regions
brain regions
apply graph/network analysis
![Page 21: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/21.jpg)
[Sole and Valverde, 2004]
![Page 22: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/22.jpg)
[Bullmore and Sporns, 2009; Bassett et al, 2010]
Small-worldhigh clustering ~functional segregationhigh efficiency ~functional integration
Cost-efficienthigh efficiency for low connection cost
Hubbyfat-tailed degree distributionshierarchical, but still resilient to attacks and errors
Modularmore dense connections to nodes in module than to nodes in other modules
![Page 23: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/23.jpg)
Show me your rest, and I tell your successLinking resting-state functional connectomes (280!) to life style/demographics/psychometric measures
23[Smith et al., Nature Neuroscience, 2015]
![Page 24: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/24.jpg)
Show me your rest, and I tell who you areFunctional connectome serves as fingerprint to identify individuals
126 individuals, across resting state & task sessions
24[Finn et al., Nature Neuroscience, 2015]
R1: rest 1, R2: rest 2, WM: working memory,Mt: motor task, Lg: language, Em: emotion
![Page 25: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/25.jpg)
OverviewThe rise of (functional) MRIActivation mapping
Task-evoked activityBrain activation maps using confirmatory analysisGeneral Linear Model (GLM)
Connectivity mappingResting-state condition reveals intrinsic functional networksBlind source separation (independent component analysis)Statistical interdependencies and graph analysis
Dynamic connectivity mappingMoment to moment fluctuations of connectivityWindowed and event-based approaches
Open challenges25
![Page 26: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/26.jpg)
Dynamic functional connectivityExtract network dynamics bysliding-window pairwise correlations
26[Chang and Glover, 2010]
regi
onal
tim
ecou
rses
dyna
mic
FC
![Page 27: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/27.jpg)
Dynamic functional connectivityProper preprocessing of timecourses is required
To avoid aliasing artefact: high-pass filtering of input timecourses with reciprocal of window length w
27
DynFC’s “no-free-lunch theorem”
frequency1/w
frequency1/w
frequency1/w
F {x}
F {y}
F {cxy
}
[Leonardi, VDV, NeuroImage, 2015]
![Page 28: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/28.jpg)
Sliding window correlation*Window length is 30 TRs, step of 2 TRs, TR=1.1sec
Three healthy subjects:
Dynamic functional connectivity
28
Frontal
Limbic
Occipit
al
Parieta
l
Subco
rtical
Tempo
ral
[Allen et al, 2014; Leonardi, VDV et al, 2013; Leonardi, VDV, 2014; Leonardi et al, 2014]
![Page 29: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/29.jpg)
FrontalLimbic
OccipitalParietal
TemporalSubcortical
0
positive
negative
Eigenconnectivities
time x subjects
PCA
- average dynamic FC issubtracted out(i.e., driven by fluctuationsof FC only!)
- optimized for explainedvariance
- orthogonality constraint
[Leonardi, VDV, NeuroImage, 2013; Leonardi, VDV, HBM, 2014; image by M. Leonardi]
“Lego” of dynamic FC
![Page 30: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/30.jpg)
global fluctuations in FC
cingulate gyri, medial frontal gyri, precuneus (default-mode network)primary sensory in red
inferior and middle frontal gyri, inferior parietal(fronto-parietal)
posterior DMNtemporal and inferior frontal
subcortical
visual
FrontalLimbic
OccipitalParietal
TemporalSubcortical
0
positive
negative
Eigenconnectivities
![Page 31: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/31.jpg)
Altered dynamics in multiple sclerosis
31[Leonardi, ..., VDV, NeuroImage, 2013] [Damaraju et al. 2014, Ma et al. 2014, Yu et al. 2015, Rashid et al. 2014]
Temporal contributions of eigenconnectivities is altered in minimally-disabled relapse-remitting multiple sclerosis patients (F(10,19)=2.6, p=0.005)
Schizophrenia, autism, temporal lobe epilepsy, ...
![Page 32: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/32.jpg)
Origin and relevance of dynamic FC
32
Parts of dynFC survive rigorous statistical testing:non-parametric randomization
Betzel et al, 2016; Zalesky, 2014; Keilholz et al, 2013; Huang, et al, ArXiv
DynFC correlates with electrical activityThompson et al, 2013; Tagliazucchi et al, 2013; Chang et al, 2013
DynFC varies along demographic variables: age, genderHutchison, Morton, 2015; Yaesoubi et al, 2015
DynFC is modulated by changes in arousal:anesthesia, caffeine
Rack-Gomer and Liu, 2012; Barttfeld et al, 2015; Tagliazucchi et al, 2014
DynFC correlates with cognition: daydreaming, cognitive flexibility
Kucyi, Davis, 2013; Yang et al, 2014; Cheng et al, 2016
![Page 33: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/33.jpg)
Spatially resolved dynamic FCVoxel-wise (dynamic) connectome
105 timecoursesConnectivity matrix is huge!Sliding-window approach... %@U$)!
But matrix is low-rank!Rank is at most #timepoints << 105
33[Preti, VDV, Scientific Reports, 2017]
?
![Page 34: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/34.jpg)
Representative dominant patterns
34
...
k-means(10x CV)
[Preti, VDV, Scientific Reports, 2017]
leading eigenvector
Interplay between major networksRDP5: full DMN vs sensorimotorRDP4,6: segregation of DMN in dorsal/ventral parts
s1 s2 s3
voxels
...
concatenation over windows/subjectsA) B)
C)
representative dominant patterns
voxel BOLD timecourses
voxelwise dominant FC patterns
D)
ui
+ -+ + - -
patterns of voxels belonging to clusters
r1 r2 r3 r4 r5 r6ROI1
+ - + + - ROI2
+ - + - - ROI3
dFC-driven cortical parcellation
{voxels
...
+
-!me
voxels(1
05)
Xi
r1 r2 r3
r4 r5 r6
k-means clustering
r1 r2 r3 r4 r5 r6
r1 r2 r3 r4 r5 r6
(s)
(s)
1 2 3
4 5 6
![Page 35: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/35.jpg)
Dynamic FC voxel-wise parcellation
35
z=-24 z=-14 z=-4 z=6
z=16 z=36 z=46 z=56
37 distributed patterns
![Page 36: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/36.jpg)
Dynamic FC voxel-wise parcellationSplit into contiguousregions
Leads to 449 parcels
Fluctuations ofdynamic FC aremeaningful in terms of long-range andshort-rangeinteractionsExtensions
Higher rankClustering...
36
12 3
4
1 23
4 5 6
AAL proposed method
Margulies 2009 proposed method
[Preti, VDV, Scientific Reports, 2017]
cerebellumprecuneus
available online at https://neurovault.org/collections/3091/
![Page 37: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/37.jpg)
Better capturing of dynamics
37
timec
ours
es
statistical dependencies characterize
“average”behavior during
resting state
[Chang and Glover, 2010; Leonardi, VDV, 2013, 2014]
windowed correlation is temporally restricted, but suffers from low SNR and potential aliasing
can we extract ‘key points’ that capture the important events?
![Page 38: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/38.jpg)
Selection of timepoints with extreme values of seed region (point process model)
Averaging is proxy of seed connectivity:
Co-activation patterns (CAPs)
38
timec
ours
es
[Tagliazucchi et al., 2012; Liu and Duyn, PNAS, 2013]
average over all selected frames: seed-based correlation:
![Page 39: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/39.jpg)
Co-activation patterns (CAPs)Temporal clustering of selected frames
Averaging of spatial activity patterns for each temporal cluster leads to “co-activation patterns” (CAPs)
39
timec
ours
es
[Tagliazucchi et al., 2012; Liu and Duyn, PNAS, 2013]
1
2
3
temporalclusters
![Page 40: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/40.jpg)
Co-activation patterns (CAPs)Averaging of spatial activity patterns for each temporal cluster leads to “co-activation patterns” (CAPs)
Temporal clustering allows CAPs to have spatial overlap, but it does not disentangle temporal overlap:
#CAPs suffers from combinatorial explosion:N “fundamental” networks that can temporally overlap K at a time would lead to ~ NK CAPs
CAPs can be contaminated by non-seed related activity
40[Tagliazucchi et al., 2012; Liu and Duyn, PNAS, 2013]
![Page 41: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/41.jpg)
[Khalidov, VDV et al, IEEE Transactions on SP, 2011; Karahanoglu, VDV et al, NeuroImage, 2013]
signal flow
fin
volume
dHb
BOLDy
q
s
v
�s/�s
�fin � 1�f
s
fin
�0
finE(fin, E0)�0E0
�fout(v,�) q
v
⇥0
�fout(v,�)⇥0H
activity-inducingsignal
Total activation regularized deconvolution of fMRI BOLD timeseries
H�1 =
Q4i=1(D � ↵iI)
D � �I
First-order Volterra kernelH-1
measured BOLD signal
innovation signal
D
recoveredactivity-inducing
signal
Total activation regularization:
x = argminx
1
2||y � x||22
| {z }data fitness
+�����DH�1 {x}
����1| {z }
regularization
![Page 42: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/42.jpg)
Innovation-driven co-activation patternsAfter total-activationdeconvolution, innovationsignals are obtainedTransients rather thanBOLD enter into theclustering
42[Karahanoglu, VDV, Nature Communications, 2015; Current Opinion in Biomedical Engineering, 2017]
![Page 43: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/43.jpg)
posterior cingular cortex (L)
angular gyrus (L)
superior frontal medial (R)
Innovation never comes alone
and 10’000 voxels more...
LR
1 2 133 1 4 1 4 2 3 25 151 121 52 15 3 51534 1 4 1 2 23 3 4 4 13511 2 1 552 1 2 52125 2[Karahanoglu and VDV, Nature Communications, 2015]
![Page 44: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/44.jpg)
iCAPs ordered in terms of occurrence
sensory components
Brain’s repertoire of functional networks
FPN
MOTVISP DMN pDMNEXEC
SUBSAL ACC
pVIS PREsVISAUD
![Page 45: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/45.jpg)
45Siemens 3T Prisma MRI scanner @ Campus Biotech / picture (c) EPFL/Alban Kakulya
![Page 46: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/46.jpg)
iCAPs ordered in terms of occurrence
iCAP-8: “full” DMN, longest duration
Brain’s repertoire of functional networks
FPN
MOTVISP DMN pDMNEXEC
SUBSAL ACC
pVIS PREsVISAUD
![Page 47: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/47.jpg)
Access to subsystems of the default-mode network
Brain’s repertoire of functional networks
FPN
MOTVISP DMN pDMNEXEC
SUBSAL ACC
pVIS PREsVISAUD
![Page 48: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/48.jpg)
PCC seed correlation
z-score
.5-.5-2 2
z-score
1.5-1.5-4 4
Default mode network de-CAP-sulated13
5
10DMN
8
pDMN
ACC
PRE
11
14
2
SAL
FPN
iCAPs dynamically assemble DMN(and other known RSNs)
![Page 49: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/49.jpg)
0
7
14
21
28
35
0 1 2 3 4 5
% o
f tot
al re
stin
g st
ate
number of overlapping iCAPs
Temporal overlap of iCAPs
1 AUD2 FPN3 PVIS4 SVIS5 PRE6 VISP7 MOT8 DMN9 EXEC10 pDMN11 SAL12 SUB13 ACC
20 most frequent combinations
+ positive- negative
on average 3.6 iCAPs active
2/3/42/3/4
4/7/87/8/132/4/8
![Page 50: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/50.jpg)
Relationship between iCAPs and behavior
12-SUB
1-AU
D
7-MOT
3-pVIS
4-sVIS
6-VISP
5-PR
E10-pDM
N8-DM
N13-ACC
9-EXEC
2-ATT
11-ASAL
2-2
Intro
cep
tion
Emot
ion
Actio
nCo
gniti
onPe
rcep
tion
Execution (Other)Execution (Speech)ImaginationInhibitionMotor LearningObservationPreparationRestAttentionLanguage (Orthography)Language (Other)Language (Phonology)Language (Semantics)Language (Speech)Language (Syntax)Memory (Explicit)Memory (Other)Memory (Working)MusicOtherReasoningSocial CognitionSomaSpaceTimeAngerAnxietyDisgustFearHappiness (Humor)Happiness (Other)OtherSadnessAir-HungerBladderHungerOtherSexualitySleepThermoregulationThirstAuditionGustationOlfactionSomesthesis (Other)Somesthesis (Pain)Vision (Color)Vision (Motion)Vision (Other)Vision (Shape)
sensorydefault attention
1-1behavioral correlation
behavioral scoresBrainMap:[Lancaster et al, Frontiers Neuroinformatics, 2012]
12-SUB
1-AU
D
7-MOT
3-pVIS
4-sVIS
6-VISP
5-PR
E10-pDM
N8-DM
N13-ACC
9-EXEC
2-ATT
11-ASAL
2-2
Intro
cep
tion
Emot
ion
Actio
nCo
gniti
onPe
rcep
tion
Execution (Other)Execution (Speech)ImaginationInhibitionMotor LearningObservationPreparationRestAttentionLanguage (Orthography)Language (Other)Language (Phonology)Language (Semantics)Language (Speech)Language (Syntax)Memory (Explicit)Memory (Other)Memory (Working)MusicOtherReasoningSocial CognitionSomaSpaceTimeAngerAnxietyDisgustFearHappiness (Humor)Happiness (Other)OtherSadnessAir-HungerBladderHungerOtherSexualitySleepThermoregulationThirstAuditionGustationOlfactionSomesthesis (Other)Somesthesis (Pain)Vision (Color)Vision (Motion)Vision (Other)Vision (Shape)
sensorydefault attention
1-1behavioral correlation
behavioral scores
2098 iCAPs combinations
hierarchicalclustering
• Highest level of hierarchy: sensory / DMN / attention
• Behavioral profiles can be determined (BrainMap)
• ... and form consistent groupings as driven by iCAPs’ combinations
12-SUB
1-AU
D
7-MOT
3-pVIS
4-sVIS
6-VISP
5-PR
E10-pDM
N8-DM
N13-ACC
9-EXEC
2-ATT
11-ASAL
2-2
Intro
cep
tion
Emot
ion
Actio
nCo
gniti
onPe
rcep
tion
Execution (Other)Execution (Speech)ImaginationInhibitionMotor LearningObservationPreparationRestAttentionLanguage (Orthography)Language (Other)Language (Phonology)Language (Semantics)Language (Speech)Language (Syntax)Memory (Explicit)Memory (Other)Memory (Working)MusicOtherReasoningSocial CognitionSomaSpaceTimeAngerAnxietyDisgustFearHappiness (Humor)Happiness (Other)OtherSadnessAir-HungerBladderHungerOtherSexualitySleepThermoregulationThirstAuditionGustationOlfactionSomesthesis (Other)Somesthesis (Pain)Vision (Color)Vision (Motion)Vision (Other)Vision (Shape)
sensorydefault attention
1-1behavioral correlation
behavioral scores
![Page 51: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/51.jpg)
Default mode network
Salience network
Time to rethink our models?
Fronto-parietal / executive networks
[Menon, Uddin, Brain Struct Func, 2010; Menon, TICS, 2011; Nekovarova et al, Frontiers, 2014]
anticorrelation
![Page 52: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/52.jpg)
Default mode network
Salience network
Time to rethink our models?
Fronto-parietal / executive networks
anticorrelation?
![Page 53: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/53.jpg)
Default mode network
Salience network
Time to rethink our models? opposite signssame signs
mixed signs
Fronto-parietal / executive networks
![Page 54: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/54.jpg)
Unraveling crosstalk of ongoing spontaneous activity
activenetworks
activebehavior
[Karahanoglu and VDV, Nature Communications, 2015]
![Page 55: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/55.jpg)
Summaryactivation mapping
objec
tive
assu
mption
s
outco
me
experimental paradigm
baseline isconstant
activation maps
functional connectivity
mapping
networks
method
sGLM BSS (ICA,...),
seed connectivity, graph analysis
dynamic functional
connectivity mapping
co-fluctuations are unstable
(“non-stationary”)
dynamicalnetworks
baseline isfluctuating
co-fluctuations are stable (“stationary”)
sliding-window FC, point process analysis,
CAPs, iCAPs,...
![Page 56: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/56.jpg)
Take home message
56
Seeing the brain in action, at any moment, and systems level Multidisciplinary endeavor where computational approaches are essentialUnderstand how it all fits together
PerspectivesTracking of brain states
Naturalistic stimuli and tasksNeurofeedbackGraph signal processing: connect function (signal) with structure (graph)
Towards new models and markers of brain (dys)functionBenefit from “big MRI data” in health and disease
![Page 57: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/57.jpg)
Two recent review papers
57
![Page 58: Functional connectivity models: from blobs to … r2 r3 r4 r5 r6 r1 r2 r3 r4 r5 r6 (s) (s) 1 2 3 4 5 6 Dynamic FC voxel-wise parcellation 35 z=-24 z=-14 z=-4 z=6 z=16 z=36 z=46 z=56](https://reader036.fdocuments.us/reader036/viewer/2022062317/5b9df94109d3f2443d8d29ec/html5/thumbnails/58.jpg)
MIP:lab @ Campus Biotechhttp://miplab.epfl.ch
DimitriYury
KirstenDjalel
LucaThomas
ZaferNaghmeh
Valeria
GwladysGiulia
IsikDaniela
Fondation Leenaards
LorenaAnjali