1 of 33

Graph theory based method to identify pathological networks involved in �Alzheimer’s disease

Presented by: Sahar ALLOUCH, Faculty of Engineering, Lebanese University

Supervised by: Dr. Aya KABBARA, Université de Rennes 1, LTSI

Dr. Mahmoud HASSAN, Université de Rennes 1, LTSI

Dr. Mohamad KHALIL, AZM Center for Research in Biotechnology and its Applications

Lebanese university

Faculty of engineering – Branch I

Electrical and Electronics department

Rennes University

LTSI laboratory

2 of 33

OUTLINE

Introduction

Data Acquisition

Data Pre-Processing

Classic Analysis: Voltage topography

Networks Analysis

Conclusion

2

7/30/2019

01

02

03

04

05

06

3 of 33

Alzheimer’s disease

3

7/30/2019

Memory loss

Troubles with familiar tasks

Confusion with time or places

Changes in mood and personality

Speaking or writing problems

Difficulties in concentration

Alzheimer’s disease (AD) is the most common cause of dementia.

WHO, 2019

Alzheimer’s disease symptoms

4 of 33

General objective

4

7/30/2019

Sociodemographic variables

Mood disorders

Not objective: depend on

Clinical symptoms
Neuropsychological tests

Diagnosis

Develop direct, objective and easy-to-use neuro-markers able to differentiate between AD patients and healthy controls.

Problematic

Objective

Use Electroencephalography (EEG) to derive a measure that correlates with the level of cognitive impairment

5 of 33

Advantages of EEG

5

7/30/2019

fMRI

PET

SPECT

MEG

EEG

Temporal resolution (s)

Spatial resolution (mm)

WHY EEG?

Non-invasive

Easy-to-use

Low cost

Excellent temporal

Resolution

Poor spatial resolution

Laureys et al., 2002

6 of 33

OUTLINE

Introduction

Data Acquisition

Data Pre-Processing

Classic Analysis: Voltage topography

Networks Analysis

Conclusion

6

7/30/2019

01

02

03

04

05

06

7 of 33

Data Acquisition

7

7/30/2019

Addenbrooke’s Cognitive Examination - Revised

4. Button Press task

2. Picture Naming task

3 sec

13 Healthy controls

11 AD patients

2 Subjects

at risk

1. Resting-state

3. One-Back task

Neurologists: Assef NASSER

Rashid AL MOHAMMAD

Discussion with Prof. BENQUET: Neurologist

8 of 33

Data Acquisition

8

7/30/2019

The clinical scores of AD patients and healthy controls. Wilcoxon test was performed.

ACE-R: p=0.001 Attention: p=0.004

Memory: p=0.001

MMSE: p=0.0007 Language: p=0.004 Visuospatial: p=0.008 Fluency: p=0.001

ACE-R: Addenbrooke’s Cognitive Examination – Revised

MMSE: Mini-Mental State Examination

Cognitive scores:

9 of 33

OUTLINE

Introduction

Data Acquisition

Data Pre-Processing

Classic Analysis: Voltage topography

Networks Analysis

Conclusion

9

7/30/2019

01

02

03

04

05

06

10 of 33

Data Pre-processing

10

7/30/2019

Eye blinks
Muscle artifacts
Cardiac artifacts

2. Non-physiological artifacts

1. Physiological artifacts

EEG Artifacts

50Hz power line
Cellphones
Chargers/Electrical equipment
Electrode/Lead movement

11 of 33

Data Pre-processing

11

7/30/2019

Bandpass filtering: [3 – 45] Hz

1

Independent component analysis -

Eye blink removal

2

Epochs extraction: [-200, 800] ms

3

Bad Channels Interpolation

4

Pre-processing steps:

12 of 33

OUTLINE

Introduction

Data Acquisition

Data Pre-Processing

Classic Analysis: Voltage topography

Networks Analysis

Conclusion

12

7/30/2019

01

02

03

04

05

06

13 of 33

Classic Analysis: Voltage Topography

13

7/30/2019

Objective:

Investigating differences in brain activity between AD patients and Healthy controls using a static and a dynamic approach

Static approach

Dynamic approach

14 of 33

Classic Analysis: Voltage Topography

14

7/30/2019

EEG signals

Dynamic analysis

Static analysis

Mann-Whitney test

Shapiro-Wilk test

Test the normality of the data

Investigate differences in voltage topographies

15 of 33

Topological Analysis

15

7/30/2019

2D topography showing significant t-values resulting from the Student t-test (Bonferroni corrected, p<0.05).

Mean Control > Mean AD

Mean Control < Mean AD

Results – Button press

16 of 33

Topological Analysis

16

7/30/2019

Results – Picture naming

2D topography showing significant t-values resulting from the Student t-test (Bonferroni corrected, p<0.05).

Mean Control > Mean AD

Mean Control < Mean AD

17 of 33

Topological Analysis

17

7/30/2019

Discussion

AD patients

Problems in naming objects (Appell et al., 1982; Huff et al., 1986)

Motor dysfunction (Albers et al., 2015; Vidoni et al., 2012)

Static analysis

Dynamic analysis

Significant activities occurring at fast timescales are hidden

Significant activities occurring at fast timescales are detected.

No significant differences between the two groups

Significant differences between the two groups

18 of 33

OUTLINE

Introduction

Data Acquisition

Data Pre-Processing

Classic Analysis: Voltage topography

Networks Analysis

Conclusion

18

7/30/2019

01

02

03

04

05

06

19 of 33

7/30/2019

19

7/30/2019

Network Analysis

Network = Nodes + Edges

Example: Social network

Objective: Derive a network-based index that correlates with the level of cognitive impairment

Network?

20 of 33

Network Analysis

20

7/30/2019

Nodes

Edges

Brain networks

abnormal

Interaction disruption

Brain diseases are network diseases

21 of 33

Data Analysis

21

7/30/2019

32 Scalp EEG signals

Inverse Problem (wMNE)

15000 Source EEG signals

Head model

Dipole orientation and location

2. Functional connectivity

1. Source reconstruction

EEG source connectivity

15000 Source EEG signals

Clustering into 68 ROIs

68 Source EEG signals

Desikan-Killiany atlas

1. Source reconstruction

wMNE : Weighted Minimum Norm Estimate

In order to construct the brain networks, we are using a relatively new technique called “EEG source connectivity”.

It consists of assessing functional connectivity at the cortical level.

It consists of two main steps: Source reconstruction and functional connectivity.

Source reconstruction aims at constructing the temporal dynamics of the cortical sources

In order to pass from scalp signals to cortical signal, we must solve an inverse problem. Since the inverse problem is ill posed, assumptions must be taken:

We suppose therefore that the cortical sources are homogeneously distributed over the cortex and normal to the cortical surface.

We used wMNE as an inverse solution.

Since the number of the reconstructed sources is too great, they are clustered based on a defined set of regions of interest.

Using the desikan killiany atlas, we obtain 68 source signals.

22 of 33

Network Analysis

22

7/30/2019

2. Functional connectivity

Sliding window

Phase-locking value (PLV)

Time

68 ROIs

Dynamic Functional connectivity

23 of 33

Network Analysis

23

7/30/2019

0.42

0.35

0.47

0.53

0.22

0.15

0.61

0.31

0.67

0.57

0.78

0.62

0.54

0.84

Node’s strength

Sum of all edge's weights connected to a node

Network measure: strength variability

How much a node changes its strength during time?

How much a network is dynamically flexible?

Node’s strength variability

Kabbara et al., 2019

24 of 33

Network Analysis

24

7/30/2019

Results

p=0.006

Mean (±std) strength variability in AD patients and healthy controls, in the two tasks-related paradigms.

AD patients have a higher strength variability than healthy controls in both tasks.

25 of 33

Network Analysis

25

7/30/2019

Results – Button press

Correlation between cognitive scores and strength variability relative to the button press paradigm.

AD patients

Healthy controls

26 of 33

Network Analysis

26

7/30/2019

Results – Picture naming

Correlation between cognitive scores and strength variability relative to the picture naming paradigm.

AD patients

Healthy controls

27 of 33

Network Analysis

27

7/30/2019

Discussion

Strength variability in AD group

Strength variability in control group

Excessive networks reconfiguration

Loss of Focus

Higher network flexibility

28 of 33

OUTLINE

Introduction

Data Acquisition

Data Pre-Processing

Classic Analysis: Voltage topography

Networks Analysis

Conclusion

28

7/30/2019

01

02

03

04

05

06

29 of 33

Conclusion

29

7/30/2019

We found that AD brain networks have a higher dynamic flexibility compared to healthy controls.

We found significant negative correlations between the network-dynamicity and the level of cognitive impairment.

We highlighted the advantage of dynamic analysis over static analysis in exploring brain activity.

Main Contributions

We used the strength variability measure for the first time in AD context in order to differentiate between AD patients and controls.

30 of 33

Limitations and Future work

30

7/30/2019

Main Limitations

Future work

Analyze data obtained from the resting state-paradigm, and one back-task.

Test other network measure to quantify dynamic networks.

Increase the number of participants

Include the “at risk” group in the analysis

Use machine learning algorithms for groups classification

Low number of patients

Low number of channels

31 of 33

Valorization

31

7/30/2019

Allouch, S., Tabbal, J., Nasser, A., Hassan, M., Khalil, M., & Kabbara, A. (2019). Altered motor performance in Alzheimer ’ s disease : a dynamic analysis using EEG. Fifth International Conference on Advances in BioMedical Engineering IEEE, Lebanon. Accepted.

Conference paper

First prize at the 11^th Annual Conference for Engineering and Architecture Graduates in North Lebanon (ACEAG) – July 27, 2019

Honors and Awards

32 of 33

THANK YOU

33 of 33

Network Analysis

33

7/30/2019

Results

Button press

Picture naming