ppt

Report
Effective Connectivity Patterns Associated with P300 Unmask
Differences in the Level of Attention/Cognition between
Normal and Disabled Subjects
S.I. Dimitriadis, Yu Sun, N.A. Laskaris ,N.Thakor, A. Bezerianos
AIIA-Lab, Informatics dept., AUTH, Greece
NeuroInformatics.GRoup , AUTH, Greece
SINAPSE, National University of Singapore
1
Outline
Introduction
• P300 is an important event-related potential (ERP)
• α-ERD could be induced by various modalities and cognitive processing
• Visual P300 can be evoked through flashing images
• Effective Connectivity Patterns (ECPs)related to Visual Stimuli (6)
in both subjects (Normal & Disabled)
Methodology
• Brain is a complex network
• Effective Connectivity Graphs (ECGs) related to Visual P300 &
α-ERD were treated as tensors
• Tensor Subspace Analysis & Classification
• Network analysis
Results
Conclusions
2
Intro
Method
Results
Conclusions
P300 is an important event-related potential (ERP) component elicited by
infrequent and task-relevant stimulus, and it reflects the processes of attention,
stimulus classification, and memory updating (Zaslansky et al., 1996 ;
Comercheroa et al., 1999 ; Linden, 2005 ; Polich, 2007)
A significant α-band (8–13 Hz in frequency) ERD (α-ERD) could be induced
by both sensory stimulation (external event) (Pfurtscheller et al.,1994)
and cognitive processing (internal event) in various attention and memory
tasks (Basar et al., 2000,2001 ; Klimesch,1997 )
P300 was showed to be functionally associated with the cognitive
processing reflected by α-ERD (Yordanova et al., 1998 ; 2001)
3
Intro
Method
Results
Conclusions
Recent Findings related to the relationship of P300 & α-ERD
As revealed by time-varying effective connectivity, the cortical information was
consistently flowed from α-ERD sources to P300 sources in the target condition
for all four sensory modalities (Peng et al., 2012)
4
Intro
Method
Results
Conclusions
In order to study the effective connectivity patterns (ECPs) related with P300 in both
normal and disabled subjects, we adopted a well-known causality estimator called
Partial Directed Coherence (PDC).
Our analysis was based on α frequency range (8-13 Ηz) as an attempt to study high
cognitive activation and attention in terms of networks at the sensor level.
We first attempted to classify correctly the (ECPs) related to visual stimuli of
different images for each subject independently (Normal & Disabled)
Finally, we explored any topographic difference between normal and disabled
subjects in terms of Complex Networks
5
Intro
Method
Results
Conclusions
ΕΕG & CONNECTOMICS
FROM MULTICHANNEL RECORDINGS TO CONNECTIVITY GRAPHS
FUNCTIONAL/EFFECTIVE
MEASURES
6
Intro
Method
Results
Conclusions
EEG & CONNECTOMICS
COMPLEXITY IN THE HUMAN BRAIN
7
Intro
Method
Results
Conclusions
Motivation and problem statement
• Association of effective connectivity patterns (ECPs) with particular cognitive tasks
has been a hot topic in neuroscience
• A low number of P300 studies followed a connectivity analysis and especially for
decoding ECPs related to different stimuli of the same modality (visual)
• We used tensor subspace analysis (TSA) to reduce the initial high-dimensionality of
the pairwise coupling in the original effective connectivity network
which
1. would significantly decrease the computational cost
and
2. facilitate the differentiation of brain states related to each visual stimulus
• The classification scheme treated each ECPs in a trial fashion
• We analyzed ECPs by adopting complex network analysis (hubs, small-world index)
8
Intro
Method
Results
Conclusions
Experimental Protocol
(Visual stimulus) - The images were flashed in
random sequences, one image at a time. Each flash
of an image lasted for 100 ms and during the
following 300 ms none of the images was flashed,
i.e. the inter-stimulus-interval was 400 ms.
Electrodes = 32
Fs=2048 Hz
The disabled subject (Cerebral Palsy) was wheelchair-bound but had
varying communication and limb muscle control abilities. In addition,
disabled subject was able to perform simple, slow movements with
their arms and hands but were un-able to control other extremities.
Spoken communication with disabled subject was possible, although he
was suffered from mild dysarthria.
9
Intro
Method
Results
Conclusions
Experimental Schedule
• Each subject completed four recording sessions.
• The first two sessions were performed on one day and the last two
sessions on another day.
• For all subjects the time between the first and the last session was less than
two weeks
• Each of the sessions consisted of six runs, one run for each of the six images.
• For further details about the protocol followed on this experiment see the
original paper related to this dataset (Hoffmann et al.,2012)
10
Intro
Method
Results
Conclusions
Preprocessing
Signals were filtered within frequency range of 8 to 13 Hz. Biological
artifacts were diminished by means of ICA employing function runica
from EEGLAB.
Effective Connectivity measure
• ARFIT package estimates both the time-invariant parameters of
the MVAR model and its optimum order (Tapio and Arnold, 2001)
• The order estimation uses Schwarz’s Bayesian Criterion (SBC) (Schwarz,
1978)
• PDCs was computed based on the MVAR model fitted to the signal
using an AR algorithm (ARFIT)
• Partial Directed Coherence (PDC) is frequency domain characterizations
of causality
PDC [0, 1] where high values in a certain frequency band reflect a
directionally linear influence from channel j to channel i in that band
and is normalized by the sum of the influenced processes.
11
Intro
Method
Results
Conclusions
Significant Links
• The maximum number of possible directed connections in a network with k nodes is
N =k*(k-1) and for k = 32 , Nmax = 992 .
Thus, the EFG is extremely dense and should be filtered out.
• We performed a Statistical filtering analysis:
- A surrogate data method with 200 realizations
- Surrogates were obtained by randomize signal to remove all causal relationships between
them (Hesse et al., 2003).
- To correct for multiple testing, the false discovery rate (FDR) method was adopted (Benjamini
and Hochberg, 1995)
Small-World Index
• We estimated Small-world index based on directed global (dGE) and local efficiency (dLE).
• The calculation of GErand and LErand was based on the procedure described here
which preserves the out-strength but not the in-strength distribution. We repeated this
procedure 250 times.
• We then averaged across all random networks to obtain GErand and LErand. The small-world
indices γ=LE/LErand and λ= GE/GErand were then calculated for the FCG under study,
and the ratio S=γ/λ was defined.
• This was greater than 1 for small-world networks
12
Intro
Method
Outline of the Tensorial Approach
Effective Connectivity
Results
Conclusions
ECG matrix
TENSOR
VECTOR
13
Intro
Method
Results
Conclusions
The linear Dimensionality Reduction Problem in Tensor Space
Tensor Subspace Analysis (TSA) is fundamentally based on Locality Preserving Projection
• X can be thought as a 2nd order tensor (or 2-tensor) in the tensor space
• The generic problem of linear dimensionality reduction in the second order space is
the following:
1. Given a set of tensors (i.e. matrices) X1,..., X m  n1  n2
find two transformation matrices U of size n1  l1 and V of n2  l2 that maps these
tensors to a set of tensors Y1,..., Ym   l1   l 2 (l1  n1, l2  n2 ) such that Yi “represents” ,
where Yi  U T X iV
Xi
14
Intro
Method
Results
Conclusions
Optimal Linear Embedding
In the case of supervised learning (classification labels are available), the label
information can be easily incorporated into the graph as follows:
  Xi  X j

Sij  e

0
2
t
If X j and Xi share the same label;
(2)
Otherwise
Let and be the transformation matrices. A reasonable transformation respecting the graph
structure can be obtained by solving the following objective functions:
min  U T X iV  U T X jV Sij
2
U ,V
(3)
ij
The objective function incurs a heavy penalty if neighboring points are mapped far apart
D  S
With D be a diagonal matrix
the optimization problem was restricted to the following equations
ii
(D U - SU )V   D U V
j
ij
(4)
Once is obtained, can be updated by solving the following generalized eigenvector
problem
(DV - SV )V  λDV V
(5)
Thus, the optimal and can be obtained by iteratively computing the generalized
eigenvectors of (4) and (5)
15
Intro
Method
Results
Conclusions
Machine learning validation
• The TSA algorithm, followed by a k-nearest-neighbor classifier (with
k=6), was tested on trial-based connectivity data from all bands.
• The following results have been obtained through a cross-validation
scheme that shuffle the trials and get 90% for training and 10% for
testing.
• The average classification rates derived after applying the above crossvalidation scheme 100 times was 100% for both subjects (Dimitriadis et
al., 2012)
16
Intro
Method
Results
Conclusions
Effective Connectivity Patterns (ECPs)
DISABLED
Effective Connectivity Graphs related to P300
for each one of the six target images
-The color of the arrow (both body and head) is
related with the strength of PDC.
Red color (hubs) (outgoing degree = k-1=31).
Comparison of ECPs
NORMAL
• The spatial distribution of hubs for the
disabled subject is located mainly in
bilateral frontal sites, left fronto-central, in
P7 and in CP6.
• For the normal subject, two more sensor
areas were detected as hubs: PO3 and
O1.
• Visual inspection of EFGs uncovered
greater connectivity strength for disabled
17
subject compared to normal
Intro
Method
Results
Conclusions
Small-World Index
Small-World index for each flashing image of Fig.1 in both subjects (* denotes
significant difference p <0.001 of SW between the two subjects across trials
using Wilcoxon rank test; Bonferroni corrected p’ < p/6).
EFGs of normal subject are closer to a small-world network compared to
the EFGs of the disabled subject for each of the flashing image
18
Intro
Method
Results
Conclusions
• Based on our experiments, the tensorial treatment of ECGs succeeded to identical
discriminate the six flashing images in normal and disabled subject
• Our network analysis revealed important hubs areas consistent in both subjects
located over frontal areas (bilaterally), left fronto-central and at sensors P7 and CP6
• Additionally, in normal subject two more hubs were revealed: PO3 and O1
• By a visual inspection of the topographies based on the connection strength, it is obvious
that hubs sensors in disabled subject entrusting the demands of the task related to high
cognitive activation and attention
• Even though our analysis based on sensor level, the lack of hubs located over left
parieto-occipital sites in disabled subject can be related to visional issues related to
cerebral palsy that balanced with more effort over frontal and fronto-central areas.
• Finally, small-world index uncovered a more well organized network in normal
compared to disabled subject
19
References
[1] Dimitriadis, S.I.,Yu, Sun., K, Kwok., et.al.,” A Tensorial Approach to Access Cognitive
Workload related to Mental Arithmetic from EEG Functional Connectivity Estimates,”
35th Annual of the IEEE EMBC, Osaka (Japan) 3 – 7 July 2013
[2] Peng, W., Hu, L., Zhang, Z., et al., “Causality in the Association between P300 and
Alpha Event-Related Desynchronization,” PLoS ONE, vol.7: e34163.
doi:10.1371/journal.pone.0034163, 2012
[3] Hoffmann, U., Vesin, J.M., Ebrahimi, T., et. al., “An efficient P300 - based braincomputer interface for disabled subjects,” J Neurosci. Methods, vol.167, pp.115-25, Jan
15, 2008.
[4] Sameshima, K., and Baccala, L.A.,”Using partial directed coherence to describe
neuronal ensemble interactions,” J. Neurosci. Methods, vol.94,pp.93–103,1999
[5] Ioannides AA, Dimitriadis SI, Saridis G, et. al., “Source Space Analysis of EventRelated Dynamic Reorganization of Brain Networks. Computational and Mathematical
Methods in Medicine Special Issue", Graph Theoretical Approaches in Brain Networks",
vol. 2012, Article ID 452503, doi:10.1155/2012/452503.
20
NeuroInformatics.GRoup , AUTH, Greece
http://neuroinformatics.gr/
Cognitive Engineering Lab – SINAPSE Laboratory –
SINGAPORE
http://www.sinapseinstitute.org/projects/cognitiveengr/
21

similar documents