SlideShare a Scribd company logo
1 of 37
Download to read offline
Individual brain parcellations
in CNeuroMod 2020
Pierre Bellec
Département de Psychologie
pierre.bellec@criugm.qc.ca
Main objective
Build high quality individual
brain parcellations, which
generalize to a variety of
videos in CNeuroMod 2020.
“One parcellation to rule them
all, and in the embedding
space bind them.”
Background public domain photo by Erik Stein. Other
movie images are under copyright, and their inclusion
falls under “fair use” (hopefully).
Contributions
Pierre Bellec -
code, data analysis,
conceptual design
Amal Boukhdhir -
code, data analysis,
conceptual design
François Paugam -
code, data analysis,
conceptual design
Hanad Shamarke -
code, data analysis
Valentina Borghesani -
data analysis
Yu Zhang -
conceptual design
Max Mignotte -
conceptual design
Fondation Courtois
The CNeuroMod Team
The Subjects
& the Scanning Team
UNIQUE
Acknowledgments
Background
&
Objectives
Brain parcellation to compare ANNs with the brain
Works trying to quantitatively compare the activity of artificial neural networks (ANNs) with the brain often compare
specific ANN layers with specific brain parcels. Figure from Schrimpf et al. Biorxiv 2020 reused under CC-BY license.
Functional
connectivity
Slow spontaneous fluctuations and
seed-based connectivity map from the
posterior cingulate cortex identifies the
default-mode network. The method can
be extended using many different
seeds.
Method introduced by Biswal and colleagues
(1995). Application to the default-mode network
by Greicius and colleagues (2003). Figure
generated with nilearn.
Seed voxel in the
posterior cingulate (PCC)
Functional
parcellation
Data-driven cluster analysis automatically
detects brain parcels with homogeneous
connectivity patterns.
Biologically meaningful parcels can be generated
at various resolution (number of parcels).
Hard parcellations (top rows) are binary, non
overlapping. Soft parcellations (bottom) are
weighted and potentially overlapping, making
dynamic parcel reconfiguration possible.
Hard parcellations from Yeo, Krienen and colleagues (2011)
with figures generated using nilearn. Soft parcellations from
Dadi et al., Neuroimage (2020) under CC BY-NC-ND
license.
Difumo
Reproducibility
With very large amount of data, individual
parcellations can be generated with high
reproducibility. Here using ~7 hours of
resting-state fMRI per subject.
Figure from Xu et al. (2020) Journal of
Neurophysiology, re-used from a preprint on Biorxiv
under CC-BY-NC-ND license.
Reproducibility
Reproducibility estimated through split-half. Although connectivity maps converge within ~30 mns (left), binary
parcels are much slower to converge and reaches a lower asymptote (right). Resting-state data from the Midnight Brain
Scan (N=10). Figure 2 From Kraus et al. (2020) NeuroImage, re-used from a preprint on Biorxiv under CC-BY-NC-ND.
Homogeneity
Homogeneity of hard group brain parcels can closely be predicted from parcel size alone, and different algorithms
have only marginal impact.
Figure from Urchs et al. MNI open research (2019) under CC-BY license.
Homogeneity
Homogeneity can be generalized to soft parcellations by examining the R2 of compressing a brain image in the
parcellation space. Soft Difumo parcels substantially improve over hard parcels.
Figure from Dadi et al., Neuroimage (2020) under CC BY-NC-ND license.
Generalization
Salehi and colleagues noticed systematic
differences in parcellation reproducibility
across different fMRI task data (at the
individual level).
Figure from Salehi et al. (2020) Neuroimage, reused
from Biorxiv under CC-BY.
Midnight Brain Scan
Yale sample
1. Extend the individual soft parcellation method from Boukhdhir et al. (2020)
to the full brain.
We hypothesized that this method can scale to very large fMRI datasets.
2. Assess the reproducibility of individual parcels across cognitive contexts.
We hypothesized that dynamic parcels are largely context-independent.
3. Assess the homogeneity of individual parcels across cognitive contexts.
We hypothesized that dynamic parcels are homogeneous across contexts.
Specific objectives & Hypotheses
Methods
Data: cneuromod-2020 release
● Movie10 (12h)
● Bourne Supremacy, Wolf of Wall
Street, Hidden Figures (x2),
● Life (x2)
●
● Friends s1 & s2 (18h)
● HCP test-retest (9h)
○ 15 repetitions of HCP
8 domains: gambling,
motor, working memory,
social, language, relational,
emotion, rest.
��
��
Movie images are under copyright, and their inclusion falls under “fair use” (hopefully).
Data: preprocessing
1. fMRI data was preprocessed using the 2020 LTS release of the fmriprep pipeline
(Esteban et al., 2018). See the cneuromod docs for details.
2. fMRI data was resampled in the MNI 2009 asymmetric template (Fonov et al.,
2009) at 2 mm isotropic and smoothed at 8 mm isotropic (for parcellation
generation) and 5 mm isotropic (for assessing homogeneity of parcels).
3. fMRI time series were denoised using the Params36 strategy implemented in
load_confounds, including slow time drifts, second-order and derivatives
expansions of motion parameters, white matter and CSF averages, as well as
global signal (Ciric et al., 2017).
Dypac algorithm
Second level: a k-means clustering
procedure aggregates one-hot
encoders and generates a set of
state stability maps.
n_state=1024
First level: run k-Means on sliding
windows of fMRI time series.
Parcels are represented with
one-hot encoders.
n_cluster=256
Number of windows / run n_replication=100
Dypac algorithm
… in numbers
First-level cluster analysis transforms a series
of ~ 50 brain volumes into a hard parcellation
of functionally connected brain regions,
which is represented as a sparse matrix.
Top image taken from the nilearn documentation
(under BSD license). https://nilearn.github.io
Dypac algorithm … in numbers (Ced)
Dypac algorithm … in numbers (Ced bis)
Dypac scalability
● Memory footprint is reasonable and does not depend on the
number of clusters, thanks to sparse boolean arrays.
● Running k-means on a 100k x 5M sparse array is feasible (a
few hours using 32 cores), thanks to scikit-learn, with support
for sparse arrays and multi-core processing.
● Implementation of full-brain “dynamic parcellation by
aggregation of clusters” (dypac) is available on github.
Results
Parcel reproducibility, friends s01 vs s02
Reproducibility is measured by maximizing spatial correlation of stability maps inside the grey matter between test and
retest between friends-s01 to friends-s02. Left: parcels from the same subject are matched. Right: parcels from different
subjects are matched.
excellent reproducibility
Parcel reproducibility, friends s01 vs s02
Parcels are matched from friends-s01 to friends-s02 by maximizing spatial correlation (sub-01).
high reproducibility low reproducibility
friends-s01 friends-s02 friends-s01 friends-s02
R2 friends-s02: subject- vs group- atlas
Individual dypac parcels are generated
from friends-s01, and R2 is estimated on
friends-s02.
The R2 of a number of group parcellation
with varying number of parcels is
presented, for reference.
R2 friends-s02: intra- vs inter-subject
Individual dypac parcels are generated
from friends-s01, and R2 is estimated on
friends-s02.
The R2 is compared when a subject is
embedded with its own parcellation, vs a
parcellation from another subject.
Parcel reproducibility, friends-s01 vs movie10
Left: parcels from the same subject are matched between friends-s01 and friends-s02. Right: parcels from the same
subject are matched between friends-s01 and movie10.
Moderate
reproducibility
Parcels matching, friends-s01 vs movie10
Parcels are matched from friends-s01 to movie10 by maximizing spatial correlation (sub-01).
high reproducibility low reproducibility
friends-s01 movie10 friends-s01 movie10
Parcel reproducibility, friends-s01 vs hcptrt
Left: parcels from the same subject are matched between friends-s01 and friends-s02. Right: parcels from the same
subject are matched between friends-s01 and hcptrt.
Good
reproducibility
Parcels matching, friends-s01 vs hcptrt
Parcels are matched from friends-s01 to hcptrt by maximizing spatial correlation (sub-01).
high reproducibility low reproducibility
friends-s01 hcptrt friends-s01 hcptrt
R2: friends vs movie10
Individual dypac parcels are
generated from friends-s01,
and R2 is estimated on
friends-s01, friends-s02, all
the movies from movie10,
and all the tasks in hcptrt.
R2 movie10 & hcptrt: subject- vs group- atlas
Individual dypac parcels are generated from friends-s01, and R2 is estimated on friends-s02 (left) and movie10
(middle) and hcptrt (right), along with R2 of a number of group parcellations.
friends-s02 movie10 hcptrt
Individual dypac parcels are generated from friends-s01, and R2 is estimated on friends-s02 (left), movie10 (middle)
and hcptrt (right). The R2 is compared when a subject is embedded with its own parcellation, vs a parcellation from
another subject.
R2 movie10 & hcptrt: intra- vs inter-subject
friends-s02 movie10 hcptrt
Conclusions
&
next steps
1. Extend the dynamic parcellation method to the full brain.
The proposed method (dypac) scales to very large individual fMRI datasets.
2. Assess the reproducibility of dynamic parcels across cognitive contexts.
○ Reproducibility is good to excellent with very long time series (~10h) and
similar types of stimuli (two different seasons of friends).
○ Some departures in parcellation were observed on movies from different
genres (moderate) or in the HCP tasks (small).
3. Assess the homogeneity of dynamic parcels across cognitive contexts.
○ Individual dypac parcels have markedly higher homogeneity (R2) than group
parcels or parcels generated on other subjects (except in hcptrt).
○ R2 was extremely stable across friends seasons, a slight decrease was
observed on movie10, with a marked decrease in some hcptrt tasks.
Conclusions
1. Investigate other cneuromod datasets
This is ongoing work for years to come.
2. Compare with other algorithms.
○ Dynamic parcels are soft, overlapping parcels. Convergence with ICA and
sparse matrix factorization should be investigated.
○ The API for model evaluation, dypac parcels and documentation will be
released for further assessment by the community.
3. Establish group vs individual parcellation best practices
○ Individual parcels embed data better than group atlases and generalize
adequately across cognitive contexts.
○ As a group atlas, Difumo performs very well and may be suitable for
situations where comparison of embeddings across subjects is required.
○ The importance of individual vs group parcellations for brain-augmented
learning remains an open question.
Next steps

More Related Content

What's hot

pantech Image processing projects 2017 18
pantech Image processing projects 2017 18pantech Image processing projects 2017 18
pantech Image processing projects 2017 18Senthil Kumar
 
Deep Learning And Business Models (VNITC 2015-09-13)
Deep Learning And Business Models (VNITC 2015-09-13)Deep Learning And Business Models (VNITC 2015-09-13)
Deep Learning And Business Models (VNITC 2015-09-13)Ha Phuong
 
Promises of Deep Learning
Promises of Deep LearningPromises of Deep Learning
Promises of Deep LearningDavid Khosid
 
行動認識手法の論文・ツール紹介
行動認識手法の論文・ツール紹介行動認識手法の論文・ツール紹介
行動認識手法の論文・ツール紹介Kensho Hara
 
Deep Learningによる超解像の進歩
Deep Learningによる超解像の進歩Deep Learningによる超解像の進歩
Deep Learningによる超解像の進歩Hiroto Honda
 
【CVPR 2020 メタサーベイ】Video Analysis and Understanding
【CVPR 2020 メタサーベイ】Video Analysis and Understanding【CVPR 2020 メタサーベイ】Video Analysis and Understanding
【CVPR 2020 メタサーベイ】Video Analysis and Understandingcvpaper. challenge
 
Poster Toward a realistic retinal simulator
Poster Toward a realistic retinal simulatorPoster Toward a realistic retinal simulator
Poster Toward a realistic retinal simulatorHassan Nasser
 
Deep Learning for Computer Vision (3/4): Video Analytics @ laSalle 2016
Deep Learning for Computer Vision (3/4): Video Analytics @ laSalle 2016Deep Learning for Computer Vision (3/4): Video Analytics @ laSalle 2016
Deep Learning for Computer Vision (3/4): Video Analytics @ laSalle 2016Universitat Politècnica de Catalunya
 
Modeling perceptual similarity and shift invariance in deep networks
Modeling perceptual similarity and shift invariance in deep networksModeling perceptual similarity and shift invariance in deep networks
Modeling perceptual similarity and shift invariance in deep networksNAVER Engineering
 
Deep Convnets for Video Processing (Master in Computer Vision Barcelona, 2016)
Deep Convnets for Video Processing (Master in Computer Vision Barcelona, 2016)Deep Convnets for Video Processing (Master in Computer Vision Barcelona, 2016)
Deep Convnets for Video Processing (Master in Computer Vision Barcelona, 2016)Universitat Politècnica de Catalunya
 
Color-plus-Depth Level-of-Detail in 3D Tele-immersive Video: A Psychophysical...
Color-plus-Depth Level-of-Detail in 3D Tele-immersive Video: A Psychophysical...Color-plus-Depth Level-of-Detail in 3D Tele-immersive Video: A Psychophysical...
Color-plus-Depth Level-of-Detail in 3D Tele-immersive Video: A Psychophysical...Wanmin Wu
 
Depth estimation do we need to throw old things away
Depth estimation do we need to throw old things awayDepth estimation do we need to throw old things away
Depth estimation do we need to throw old things awayNAVER Engineering
 
11338 13634 July 07 (I)
11338 13634 July 07 (I)11338 13634 July 07 (I)
11338 13634 July 07 (I)kunal2279
 
Learning Representations for Sign Language Videos - Xavier Giro - NIST TRECVI...
Learning Representations for Sign Language Videos - Xavier Giro - NIST TRECVI...Learning Representations for Sign Language Videos - Xavier Giro - NIST TRECVI...
Learning Representations for Sign Language Videos - Xavier Giro - NIST TRECVI...Universitat Politècnica de Catalunya
 
Self Attested Images for Secured Transactions using Superior SOM
Self Attested Images for Secured Transactions using Superior SOMSelf Attested Images for Secured Transactions using Superior SOM
Self Attested Images for Secured Transactions using Superior SOMIDES Editor
 
Cassandra audio-video sensor fusion for aggression detection
Cassandra  audio-video sensor fusion for aggression detectionCassandra  audio-video sensor fusion for aggression detection
Cassandra audio-video sensor fusion for aggression detectionJoão Gabriel Lima
 
Deep Learning Representations for All - Xavier Giro-i-Nieto - IRI Barcelona 2020
Deep Learning Representations for All - Xavier Giro-i-Nieto - IRI Barcelona 2020Deep Learning Representations for All - Xavier Giro-i-Nieto - IRI Barcelona 2020
Deep Learning Representations for All - Xavier Giro-i-Nieto - IRI Barcelona 2020Universitat Politècnica de Catalunya
 

What's hot (20)

pantech Image processing projects 2017 18
pantech Image processing projects 2017 18pantech Image processing projects 2017 18
pantech Image processing projects 2017 18
 
Deep Learning And Business Models (VNITC 2015-09-13)
Deep Learning And Business Models (VNITC 2015-09-13)Deep Learning And Business Models (VNITC 2015-09-13)
Deep Learning And Business Models (VNITC 2015-09-13)
 
Promises of Deep Learning
Promises of Deep LearningPromises of Deep Learning
Promises of Deep Learning
 
行動認識手法の論文・ツール紹介
行動認識手法の論文・ツール紹介行動認識手法の論文・ツール紹介
行動認識手法の論文・ツール紹介
 
Deep Learningによる超解像の進歩
Deep Learningによる超解像の進歩Deep Learningによる超解像の進歩
Deep Learningによる超解像の進歩
 
【CVPR 2020 メタサーベイ】Video Analysis and Understanding
【CVPR 2020 メタサーベイ】Video Analysis and Understanding【CVPR 2020 メタサーベイ】Video Analysis and Understanding
【CVPR 2020 メタサーベイ】Video Analysis and Understanding
 
Poster Toward a realistic retinal simulator
Poster Toward a realistic retinal simulatorPoster Toward a realistic retinal simulator
Poster Toward a realistic retinal simulator
 
Deep Learning for Computer Vision (3/4): Video Analytics @ laSalle 2016
Deep Learning for Computer Vision (3/4): Video Analytics @ laSalle 2016Deep Learning for Computer Vision (3/4): Video Analytics @ laSalle 2016
Deep Learning for Computer Vision (3/4): Video Analytics @ laSalle 2016
 
Modeling perceptual similarity and shift invariance in deep networks
Modeling perceptual similarity and shift invariance in deep networksModeling perceptual similarity and shift invariance in deep networks
Modeling perceptual similarity and shift invariance in deep networks
 
Skipping and Repeating Samples in Recurrent Neural Networks
Skipping and Repeating Samples in Recurrent Neural NetworksSkipping and Repeating Samples in Recurrent Neural Networks
Skipping and Repeating Samples in Recurrent Neural Networks
 
Deep Convnets for Video Processing (Master in Computer Vision Barcelona, 2016)
Deep Convnets for Video Processing (Master in Computer Vision Barcelona, 2016)Deep Convnets for Video Processing (Master in Computer Vision Barcelona, 2016)
Deep Convnets for Video Processing (Master in Computer Vision Barcelona, 2016)
 
Neural Architectures for Video Encoding
Neural Architectures for Video EncodingNeural Architectures for Video Encoding
Neural Architectures for Video Encoding
 
Color-plus-Depth Level-of-Detail in 3D Tele-immersive Video: A Psychophysical...
Color-plus-Depth Level-of-Detail in 3D Tele-immersive Video: A Psychophysical...Color-plus-Depth Level-of-Detail in 3D Tele-immersive Video: A Psychophysical...
Color-plus-Depth Level-of-Detail in 3D Tele-immersive Video: A Psychophysical...
 
Depth estimation do we need to throw old things away
Depth estimation do we need to throw old things awayDepth estimation do we need to throw old things away
Depth estimation do we need to throw old things away
 
11338 13634 July 07 (I)
11338 13634 July 07 (I)11338 13634 July 07 (I)
11338 13634 July 07 (I)
 
Learning Representations for Sign Language Videos - Xavier Giro - NIST TRECVI...
Learning Representations for Sign Language Videos - Xavier Giro - NIST TRECVI...Learning Representations for Sign Language Videos - Xavier Giro - NIST TRECVI...
Learning Representations for Sign Language Videos - Xavier Giro - NIST TRECVI...
 
Self Attested Images for Secured Transactions using Superior SOM
Self Attested Images for Secured Transactions using Superior SOMSelf Attested Images for Secured Transactions using Superior SOM
Self Attested Images for Secured Transactions using Superior SOM
 
Cassandra audio-video sensor fusion for aggression detection
Cassandra  audio-video sensor fusion for aggression detectionCassandra  audio-video sensor fusion for aggression detection
Cassandra audio-video sensor fusion for aggression detection
 
Deep Learning Representations for All - Xavier Giro-i-Nieto - IRI Barcelona 2020
Deep Learning Representations for All - Xavier Giro-i-Nieto - IRI Barcelona 2020Deep Learning Representations for All - Xavier Giro-i-Nieto - IRI Barcelona 2020
Deep Learning Representations for All - Xavier Giro-i-Nieto - IRI Barcelona 2020
 
Speaker ID II (D4L1 Deep Learning for Speech and Language UPC 2017)
Speaker ID II (D4L1 Deep Learning for Speech and Language UPC 2017)Speaker ID II (D4L1 Deep Learning for Speech and Language UPC 2017)
Speaker ID II (D4L1 Deep Learning for Speech and Language UPC 2017)
 

Similar to Cneuromod20210329 bellec parcellation

Image Translation with GAN
Image Translation with GANImage Translation with GAN
Image Translation with GANJunho Cho
 
TensorFlow London: Cutting edge generative models
TensorFlow London: Cutting edge generative modelsTensorFlow London: Cutting edge generative models
TensorFlow London: Cutting edge generative modelsSeldon
 
Improved Brain Segmentation using Pixel Separation and Additional Segmentatio...
Improved Brain Segmentation using Pixel Separation and Additional Segmentatio...Improved Brain Segmentation using Pixel Separation and Additional Segmentatio...
Improved Brain Segmentation using Pixel Separation and Additional Segmentatio...AFEng1
 
System for Detecting Deepfake in Videos – A Survey
System for Detecting Deepfake in Videos – A SurveySystem for Detecting Deepfake in Videos – A Survey
System for Detecting Deepfake in Videos – A SurveyIRJET Journal
 
最近の研究情勢についていくために - Deep Learningを中心に -
最近の研究情勢についていくために - Deep Learningを中心に - 最近の研究情勢についていくために - Deep Learningを中心に -
最近の研究情勢についていくために - Deep Learningを中心に - Hiroshi Fukui
 
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATUREMINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATUREijcsit
 
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATUREMINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATUREAIRCC Publishing Corporation
 
Review : Inter-slice Context Residual Learning for 3D Medical Image Segmentation
Review : Inter-slice Context Residual Learning for 3D Medical Image SegmentationReview : Inter-slice Context Residual Learning for 3D Medical Image Segmentation
Review : Inter-slice Context Residual Learning for 3D Medical Image SegmentationDongmin Choi
 
Scene Description From Images To Sentences
Scene Description From Images To SentencesScene Description From Images To Sentences
Scene Description From Images To SentencesIRJET Journal
 
CARLsim 3: Concepts, Tools, and Applications
CARLsim 3: Concepts, Tools, and ApplicationsCARLsim 3: Concepts, Tools, and Applications
CARLsim 3: Concepts, Tools, and ApplicationsMichael Beyeler
 
Quantitative Propagation of Chaos for SGD in Wide Neural Networks
Quantitative Propagation of Chaos for SGD in Wide Neural NetworksQuantitative Propagation of Chaos for SGD in Wide Neural Networks
Quantitative Propagation of Chaos for SGD in Wide Neural NetworksValentin De Bortoli
 
IRJET- Image Captioning using Multimodal Embedding
IRJET-  	  Image Captioning using Multimodal EmbeddingIRJET-  	  Image Captioning using Multimodal Embedding
IRJET- Image Captioning using Multimodal EmbeddingIRJET Journal
 
IRJET-Lossless Image compression and decompression using Huffman coding
IRJET-Lossless Image compression and decompression using Huffman codingIRJET-Lossless Image compression and decompression using Huffman coding
IRJET-Lossless Image compression and decompression using Huffman codingIRJET Journal
 
Unsupervised Cross-Domain Image Generation
Unsupervised Cross-Domain Image GenerationUnsupervised Cross-Domain Image Generation
Unsupervised Cross-Domain Image GenerationJunho Cho
 
Beginner's Guide to Diffusion Models..pptx
Beginner's Guide to Diffusion Models..pptxBeginner's Guide to Diffusion Models..pptx
Beginner's Guide to Diffusion Models..pptxIshaq Khan
 

Similar to Cneuromod20210329 bellec parcellation (20)

Introduction to Deep learning
Introduction to Deep learningIntroduction to Deep learning
Introduction to Deep learning
 
Image Translation with GAN
Image Translation with GANImage Translation with GAN
Image Translation with GAN
 
TensorFlow London: Cutting edge generative models
TensorFlow London: Cutting edge generative modelsTensorFlow London: Cutting edge generative models
TensorFlow London: Cutting edge generative models
 
Improved Brain Segmentation using Pixel Separation and Additional Segmentatio...
Improved Brain Segmentation using Pixel Separation and Additional Segmentatio...Improved Brain Segmentation using Pixel Separation and Additional Segmentatio...
Improved Brain Segmentation using Pixel Separation and Additional Segmentatio...
 
System for Detecting Deepfake in Videos – A Survey
System for Detecting Deepfake in Videos – A SurveySystem for Detecting Deepfake in Videos – A Survey
System for Detecting Deepfake in Videos – A Survey
 
Europy17_dibernardo
Europy17_dibernardoEuropy17_dibernardo
Europy17_dibernardo
 
最近の研究情勢についていくために - Deep Learningを中心に -
最近の研究情勢についていくために - Deep Learningを中心に - 最近の研究情勢についていくために - Deep Learningを中心に -
最近の研究情勢についていくために - Deep Learningを中心に -
 
Literature Review
Literature ReviewLiterature Review
Literature Review
 
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATUREMINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
 
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATUREMINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
MINIMIZING DISTORTION IN STEGANOG-RAPHY BASED ON IMAGE FEATURE
 
Review : Inter-slice Context Residual Learning for 3D Medical Image Segmentation
Review : Inter-slice Context Residual Learning for 3D Medical Image SegmentationReview : Inter-slice Context Residual Learning for 3D Medical Image Segmentation
Review : Inter-slice Context Residual Learning for 3D Medical Image Segmentation
 
Scene Description From Images To Sentences
Scene Description From Images To SentencesScene Description From Images To Sentences
Scene Description From Images To Sentences
 
CARLsim 3: Concepts, Tools, and Applications
CARLsim 3: Concepts, Tools, and ApplicationsCARLsim 3: Concepts, Tools, and Applications
CARLsim 3: Concepts, Tools, and Applications
 
Lausanne 2019 #2
Lausanne 2019 #2Lausanne 2019 #2
Lausanne 2019 #2
 
Quantitative Propagation of Chaos for SGD in Wide Neural Networks
Quantitative Propagation of Chaos for SGD in Wide Neural NetworksQuantitative Propagation of Chaos for SGD in Wide Neural Networks
Quantitative Propagation of Chaos for SGD in Wide Neural Networks
 
IRJET- Image Captioning using Multimodal Embedding
IRJET-  	  Image Captioning using Multimodal EmbeddingIRJET-  	  Image Captioning using Multimodal Embedding
IRJET- Image Captioning using Multimodal Embedding
 
IRJET-Lossless Image compression and decompression using Huffman coding
IRJET-Lossless Image compression and decompression using Huffman codingIRJET-Lossless Image compression and decompression using Huffman coding
IRJET-Lossless Image compression and decompression using Huffman coding
 
Unsupervised Cross-Domain Image Generation
Unsupervised Cross-Domain Image GenerationUnsupervised Cross-Domain Image Generation
Unsupervised Cross-Domain Image Generation
 
Optics group research overview
Optics group research overviewOptics group research overview
Optics group research overview
 
Beginner's Guide to Diffusion Models..pptx
Beginner's Guide to Diffusion Models..pptxBeginner's Guide to Diffusion Models..pptx
Beginner's Guide to Diffusion Models..pptx
 

Recently uploaded

Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43bNightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43bSérgio Sacani
 
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptxUnlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptxanandsmhk
 
Pests of cotton_Sucking_Pests_Dr.UPR.pdf
Pests of cotton_Sucking_Pests_Dr.UPR.pdfPests of cotton_Sucking_Pests_Dr.UPR.pdf
Pests of cotton_Sucking_Pests_Dr.UPR.pdfPirithiRaju
 
Zoology 4th semester series (krishna).pdf
Zoology 4th semester series (krishna).pdfZoology 4th semester series (krishna).pdf
Zoology 4th semester series (krishna).pdfSumit Kumar yadav
 
Biopesticide (2).pptx .This slides helps to know the different types of biop...
Biopesticide (2).pptx  .This slides helps to know the different types of biop...Biopesticide (2).pptx  .This slides helps to know the different types of biop...
Biopesticide (2).pptx .This slides helps to know the different types of biop...RohitNehra6
 
Hubble Asteroid Hunter III. Physical properties of newly found asteroids
Hubble Asteroid Hunter III. Physical properties of newly found asteroidsHubble Asteroid Hunter III. Physical properties of newly found asteroids
Hubble Asteroid Hunter III. Physical properties of newly found asteroidsSérgio Sacani
 
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...Sérgio Sacani
 
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...jana861314
 
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Lokesh Kothari
 
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCRStunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCRDelhi Call girls
 
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdfPests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdfPirithiRaju
 
Natural Polymer Based Nanomaterials
Natural Polymer Based NanomaterialsNatural Polymer Based Nanomaterials
Natural Polymer Based NanomaterialsAArockiyaNisha
 
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral AnalysisRaman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral AnalysisDiwakar Mishra
 
Disentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOSTDisentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOSTSérgio Sacani
 
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...Sérgio Sacani
 
Nanoparticles synthesis and characterization​ ​
Nanoparticles synthesis and characterization​  ​Nanoparticles synthesis and characterization​  ​
Nanoparticles synthesis and characterization​ ​kaibalyasahoo82800
 
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptxSOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptxkessiyaTpeter
 
Spermiogenesis or Spermateleosis or metamorphosis of spermatid
Spermiogenesis or Spermateleosis or metamorphosis of spermatidSpermiogenesis or Spermateleosis or metamorphosis of spermatid
Spermiogenesis or Spermateleosis or metamorphosis of spermatidSarthak Sekhar Mondal
 
GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)Areesha Ahmad
 
Botany krishna series 2nd semester Only Mcq type questions
Botany krishna series 2nd semester Only Mcq type questionsBotany krishna series 2nd semester Only Mcq type questions
Botany krishna series 2nd semester Only Mcq type questionsSumit Kumar yadav
 

Recently uploaded (20)

Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43bNightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
 
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptxUnlocking  the Potential: Deep dive into ocean of Ceramic Magnets.pptx
Unlocking the Potential: Deep dive into ocean of Ceramic Magnets.pptx
 
Pests of cotton_Sucking_Pests_Dr.UPR.pdf
Pests of cotton_Sucking_Pests_Dr.UPR.pdfPests of cotton_Sucking_Pests_Dr.UPR.pdf
Pests of cotton_Sucking_Pests_Dr.UPR.pdf
 
Zoology 4th semester series (krishna).pdf
Zoology 4th semester series (krishna).pdfZoology 4th semester series (krishna).pdf
Zoology 4th semester series (krishna).pdf
 
Biopesticide (2).pptx .This slides helps to know the different types of biop...
Biopesticide (2).pptx  .This slides helps to know the different types of biop...Biopesticide (2).pptx  .This slides helps to know the different types of biop...
Biopesticide (2).pptx .This slides helps to know the different types of biop...
 
Hubble Asteroid Hunter III. Physical properties of newly found asteroids
Hubble Asteroid Hunter III. Physical properties of newly found asteroidsHubble Asteroid Hunter III. Physical properties of newly found asteroids
Hubble Asteroid Hunter III. Physical properties of newly found asteroids
 
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
All-domain Anomaly Resolution Office U.S. Department of Defense (U) Case: “Eg...
 
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
Traditional Agroforestry System in India- Shifting Cultivation, Taungya, Home...
 
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
 
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCRStunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
 
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdfPests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
 
Natural Polymer Based Nanomaterials
Natural Polymer Based NanomaterialsNatural Polymer Based Nanomaterials
Natural Polymer Based Nanomaterials
 
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral AnalysisRaman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
Raman spectroscopy.pptx M Pharm, M Sc, Advanced Spectral Analysis
 
Disentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOSTDisentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOST
 
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
 
Nanoparticles synthesis and characterization​ ​
Nanoparticles synthesis and characterization​  ​Nanoparticles synthesis and characterization​  ​
Nanoparticles synthesis and characterization​ ​
 
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptxSOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
SOLUBLE PATTERN RECOGNITION RECEPTORS.pptx
 
Spermiogenesis or Spermateleosis or metamorphosis of spermatid
Spermiogenesis or Spermateleosis or metamorphosis of spermatidSpermiogenesis or Spermateleosis or metamorphosis of spermatid
Spermiogenesis or Spermateleosis or metamorphosis of spermatid
 
GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)
 
Botany krishna series 2nd semester Only Mcq type questions
Botany krishna series 2nd semester Only Mcq type questionsBotany krishna series 2nd semester Only Mcq type questions
Botany krishna series 2nd semester Only Mcq type questions
 

Cneuromod20210329 bellec parcellation

  • 1. Individual brain parcellations in CNeuroMod 2020 Pierre Bellec Département de Psychologie pierre.bellec@criugm.qc.ca
  • 2. Main objective Build high quality individual brain parcellations, which generalize to a variety of videos in CNeuroMod 2020. “One parcellation to rule them all, and in the embedding space bind them.” Background public domain photo by Erik Stein. Other movie images are under copyright, and their inclusion falls under “fair use” (hopefully).
  • 3. Contributions Pierre Bellec - code, data analysis, conceptual design Amal Boukhdhir - code, data analysis, conceptual design François Paugam - code, data analysis, conceptual design Hanad Shamarke - code, data analysis Valentina Borghesani - data analysis Yu Zhang - conceptual design Max Mignotte - conceptual design
  • 4. Fondation Courtois The CNeuroMod Team The Subjects & the Scanning Team UNIQUE Acknowledgments
  • 6. Brain parcellation to compare ANNs with the brain Works trying to quantitatively compare the activity of artificial neural networks (ANNs) with the brain often compare specific ANN layers with specific brain parcels. Figure from Schrimpf et al. Biorxiv 2020 reused under CC-BY license.
  • 7. Functional connectivity Slow spontaneous fluctuations and seed-based connectivity map from the posterior cingulate cortex identifies the default-mode network. The method can be extended using many different seeds. Method introduced by Biswal and colleagues (1995). Application to the default-mode network by Greicius and colleagues (2003). Figure generated with nilearn. Seed voxel in the posterior cingulate (PCC)
  • 8. Functional parcellation Data-driven cluster analysis automatically detects brain parcels with homogeneous connectivity patterns. Biologically meaningful parcels can be generated at various resolution (number of parcels). Hard parcellations (top rows) are binary, non overlapping. Soft parcellations (bottom) are weighted and potentially overlapping, making dynamic parcel reconfiguration possible. Hard parcellations from Yeo, Krienen and colleagues (2011) with figures generated using nilearn. Soft parcellations from Dadi et al., Neuroimage (2020) under CC BY-NC-ND license. Difumo
  • 9. Reproducibility With very large amount of data, individual parcellations can be generated with high reproducibility. Here using ~7 hours of resting-state fMRI per subject. Figure from Xu et al. (2020) Journal of Neurophysiology, re-used from a preprint on Biorxiv under CC-BY-NC-ND license.
  • 10. Reproducibility Reproducibility estimated through split-half. Although connectivity maps converge within ~30 mns (left), binary parcels are much slower to converge and reaches a lower asymptote (right). Resting-state data from the Midnight Brain Scan (N=10). Figure 2 From Kraus et al. (2020) NeuroImage, re-used from a preprint on Biorxiv under CC-BY-NC-ND.
  • 11. Homogeneity Homogeneity of hard group brain parcels can closely be predicted from parcel size alone, and different algorithms have only marginal impact. Figure from Urchs et al. MNI open research (2019) under CC-BY license.
  • 12. Homogeneity Homogeneity can be generalized to soft parcellations by examining the R2 of compressing a brain image in the parcellation space. Soft Difumo parcels substantially improve over hard parcels. Figure from Dadi et al., Neuroimage (2020) under CC BY-NC-ND license.
  • 13. Generalization Salehi and colleagues noticed systematic differences in parcellation reproducibility across different fMRI task data (at the individual level). Figure from Salehi et al. (2020) Neuroimage, reused from Biorxiv under CC-BY. Midnight Brain Scan Yale sample
  • 14. 1. Extend the individual soft parcellation method from Boukhdhir et al. (2020) to the full brain. We hypothesized that this method can scale to very large fMRI datasets. 2. Assess the reproducibility of individual parcels across cognitive contexts. We hypothesized that dynamic parcels are largely context-independent. 3. Assess the homogeneity of individual parcels across cognitive contexts. We hypothesized that dynamic parcels are homogeneous across contexts. Specific objectives & Hypotheses
  • 16. Data: cneuromod-2020 release ● Movie10 (12h) ● Bourne Supremacy, Wolf of Wall Street, Hidden Figures (x2), ● Life (x2) ● ● Friends s1 & s2 (18h) ● HCP test-retest (9h) ○ 15 repetitions of HCP 8 domains: gambling, motor, working memory, social, language, relational, emotion, rest. �� �� Movie images are under copyright, and their inclusion falls under “fair use” (hopefully).
  • 17. Data: preprocessing 1. fMRI data was preprocessed using the 2020 LTS release of the fmriprep pipeline (Esteban et al., 2018). See the cneuromod docs for details. 2. fMRI data was resampled in the MNI 2009 asymmetric template (Fonov et al., 2009) at 2 mm isotropic and smoothed at 8 mm isotropic (for parcellation generation) and 5 mm isotropic (for assessing homogeneity of parcels). 3. fMRI time series were denoised using the Params36 strategy implemented in load_confounds, including slow time drifts, second-order and derivatives expansions of motion parameters, white matter and CSF averages, as well as global signal (Ciric et al., 2017).
  • 18. Dypac algorithm Second level: a k-means clustering procedure aggregates one-hot encoders and generates a set of state stability maps. n_state=1024 First level: run k-Means on sliding windows of fMRI time series. Parcels are represented with one-hot encoders. n_cluster=256 Number of windows / run n_replication=100
  • 19. Dypac algorithm … in numbers First-level cluster analysis transforms a series of ~ 50 brain volumes into a hard parcellation of functionally connected brain regions, which is represented as a sparse matrix. Top image taken from the nilearn documentation (under BSD license). https://nilearn.github.io
  • 20. Dypac algorithm … in numbers (Ced)
  • 21. Dypac algorithm … in numbers (Ced bis)
  • 22. Dypac scalability ● Memory footprint is reasonable and does not depend on the number of clusters, thanks to sparse boolean arrays. ● Running k-means on a 100k x 5M sparse array is feasible (a few hours using 32 cores), thanks to scikit-learn, with support for sparse arrays and multi-core processing. ● Implementation of full-brain “dynamic parcellation by aggregation of clusters” (dypac) is available on github.
  • 24. Parcel reproducibility, friends s01 vs s02 Reproducibility is measured by maximizing spatial correlation of stability maps inside the grey matter between test and retest between friends-s01 to friends-s02. Left: parcels from the same subject are matched. Right: parcels from different subjects are matched. excellent reproducibility
  • 25. Parcel reproducibility, friends s01 vs s02 Parcels are matched from friends-s01 to friends-s02 by maximizing spatial correlation (sub-01). high reproducibility low reproducibility friends-s01 friends-s02 friends-s01 friends-s02
  • 26. R2 friends-s02: subject- vs group- atlas Individual dypac parcels are generated from friends-s01, and R2 is estimated on friends-s02. The R2 of a number of group parcellation with varying number of parcels is presented, for reference.
  • 27. R2 friends-s02: intra- vs inter-subject Individual dypac parcels are generated from friends-s01, and R2 is estimated on friends-s02. The R2 is compared when a subject is embedded with its own parcellation, vs a parcellation from another subject.
  • 28. Parcel reproducibility, friends-s01 vs movie10 Left: parcels from the same subject are matched between friends-s01 and friends-s02. Right: parcels from the same subject are matched between friends-s01 and movie10. Moderate reproducibility
  • 29. Parcels matching, friends-s01 vs movie10 Parcels are matched from friends-s01 to movie10 by maximizing spatial correlation (sub-01). high reproducibility low reproducibility friends-s01 movie10 friends-s01 movie10
  • 30. Parcel reproducibility, friends-s01 vs hcptrt Left: parcels from the same subject are matched between friends-s01 and friends-s02. Right: parcels from the same subject are matched between friends-s01 and hcptrt. Good reproducibility
  • 31. Parcels matching, friends-s01 vs hcptrt Parcels are matched from friends-s01 to hcptrt by maximizing spatial correlation (sub-01). high reproducibility low reproducibility friends-s01 hcptrt friends-s01 hcptrt
  • 32. R2: friends vs movie10 Individual dypac parcels are generated from friends-s01, and R2 is estimated on friends-s01, friends-s02, all the movies from movie10, and all the tasks in hcptrt.
  • 33. R2 movie10 & hcptrt: subject- vs group- atlas Individual dypac parcels are generated from friends-s01, and R2 is estimated on friends-s02 (left) and movie10 (middle) and hcptrt (right), along with R2 of a number of group parcellations. friends-s02 movie10 hcptrt
  • 34. Individual dypac parcels are generated from friends-s01, and R2 is estimated on friends-s02 (left), movie10 (middle) and hcptrt (right). The R2 is compared when a subject is embedded with its own parcellation, vs a parcellation from another subject. R2 movie10 & hcptrt: intra- vs inter-subject friends-s02 movie10 hcptrt
  • 36. 1. Extend the dynamic parcellation method to the full brain. The proposed method (dypac) scales to very large individual fMRI datasets. 2. Assess the reproducibility of dynamic parcels across cognitive contexts. ○ Reproducibility is good to excellent with very long time series (~10h) and similar types of stimuli (two different seasons of friends). ○ Some departures in parcellation were observed on movies from different genres (moderate) or in the HCP tasks (small). 3. Assess the homogeneity of dynamic parcels across cognitive contexts. ○ Individual dypac parcels have markedly higher homogeneity (R2) than group parcels or parcels generated on other subjects (except in hcptrt). ○ R2 was extremely stable across friends seasons, a slight decrease was observed on movie10, with a marked decrease in some hcptrt tasks. Conclusions
  • 37. 1. Investigate other cneuromod datasets This is ongoing work for years to come. 2. Compare with other algorithms. ○ Dynamic parcels are soft, overlapping parcels. Convergence with ICA and sparse matrix factorization should be investigated. ○ The API for model evaluation, dypac parcels and documentation will be released for further assessment by the community. 3. Establish group vs individual parcellation best practices ○ Individual parcels embed data better than group atlases and generalize adequately across cognitive contexts. ○ As a group atlas, Difumo performs very well and may be suitable for situations where comparison of embeddings across subjects is required. ○ The importance of individual vs group parcellations for brain-augmented learning remains an open question. Next steps