Francisco del Pozo
Centro de Tecnología Biomédica CTBUniversidad Politécnica de Madrid UPM
Parque Científico y Tecnológico de Montegancedo
www.ctb.up.es
EA-DIAGNOSTIC
Early detection of Alzheimer
Disease
20
14
The Cognitive Continuum
MCI
Mild cognitive
Impairment
Normal
Alzheimer's disease
SMCSUBJECTIVE
MEMORY
COMPLAINTS
Memory complaints
with normal scores
on
neuropsychological
test: no medical
evidence of disease
Experimental
target
4,2 - 6 %
The problem – the market
Alzheimer Disease (AD)
affects 4.2 - 6% of the
population over 65
Spain: 600.000 people
with dementia (400.000
AD)
AD vs Healthy ageing
Are there early functional synchronization differences?
The Objectives
1. RESEARCH ON BIOMARKERS FOR THE
EARLY DETECTION OF ALZHEIMER DISEASE
2. THE CLINICAL TRANSLACION. CLINICAL
TRIALS
3. INDUSTRIAL TRANSFER TO MAKE THAT
POSSIBLE
Toward the early detection of
Alzheimer Disease (AD)
Multicenter-
Clinical trial
Amiloid
Genetics
TAU
RMI
Neuro-
psicology
DTI/MEGAD Biomarker
Clinical individual
AD Biomarkers
Biomarkers
exploitation
MCI/SMC
Analysis
(resting/memory
)
MEG-based brain
conectivity analysis
Graph models
Multidimentional
Classification
Synchronization
MEG analysis
Describing brain connectivity
• Cross-correlation
• Wavelet coherence
• Sync. likelihood
• Generalized Sync.
• Phase Sync.
- EEG
- MEG
- fMRI
- Histological Analysis
- DTI (MRI)
Anatomical Networks Functional Networks
From Bullmore & Sporns, Nature Rev. 10, 186
(2009)
MEG continous recordings
Channel #1
Channel #340TIME
MAGNETIC
FIELD
Visualización conectividad
306 sensors: Planar
radiometers and
magnetometers
Powerful connectivity tools are required
Statistics between groups/conditions
Functional and Effective Connectivity
Signals: Time domain
Signals: Frecuency domain
a)C
PU
/GP
U
Pa
ralle
liza
tio
n
CU
DA
/Op
en
MP
HERMES Integrated toolbox to characterize
functional and effective brain connectivity
1. CLASSICAL MEASURES◦ Cross-Correlation
◦ Correlation
◦ Coherence
◦ [C PARAMETERS]
2. PHASE SYNCRONIZATION
MEASURES◦ Phase Locking Value (PLV)
◦ Phase-Lag Index (PLI)
◦ Weighted Phase-Lag Index (WPLI)
◦ RHO
◦ [PS PARAMETERS]
3. GENERALIZED SYNCRONIZATION
MEASURES◦ S Index
◦ H Index
◦ N Index
◦ M Index
◦ L Index
◦ Synchronization Likelihood
◦ [GS PARAMETERS]
4. GRANGER CAUSALITY◦ Clasical Linear Granger Causality
◦ Direct Transfer Function (DTF)
◦ Partial Directed Coherence (PDC)
◦ [GC PARAMETERS]
There are conectivity differences
Synchronization profiles
Frontoparietal Synchronization
Memory tasks
MCI vs Controls
Synchronization profiles
Lateral Synchronization
Memory tasks
MCI vs Controls
There are conectivity differences
(López, Bruña et al, 2014)
MCI AD Conversion (memory tasks)
Subjetive Memory Complaints (SMC)
Alzheimer'
s disease
SMC Memory
complaints with
normal scores on
neuropsychological
test: no medical
evidence of disease
SMCSUBJECTIVE
MEMORY
COMPLAINT
S
MCI
Normal
Default Mode Network
How to describe synchronization changes.
Modeling Brain Newtworks
Control
MCI
Degree, clustering, outreach
and knn distributions:
MCI networks have nodes with
higher connectivity.
The clustering increases with the
degree (in both Control and MCI).
For the same degree, outreach is
higher at the MCI group.
Networks are assortative.
How to describe synchronization changes.
Modeling Brain Newtworks
Differences between the MCI and
Control groups:
Randomness
Ne
two
rk s
tre
ng
th
Control
M.C.I.
High
Synchronization
Low clustering
Higher outreach
Low modularity
Higher Randomness
Alzheimer
Low
Synchronization
Low clustering
Higher Randomness
Synchronization results vs. other AD-
related potential markers
Multicenter-
Clinical trial
Amiloid
Genetics
TAU
RMI
Neuro-psicology
DTI/MEG
Biomarkers
Clinical individual
AD Biomarkers
Biomarkers
exploitation
MCI/SMC
Resting/memory
analysis
MEG-based brain
conectivity analysis
Graph models
Multidimentional
Classification
Synchronization
MEG analysis
Looking for biomarkers from image
multimodalities
Neuroimaging
Neuroimaging Biomarkers in Aging and Dementia. Application of volumetric
techniques (Voxel-Based Morphometry, VBM) at different stages of
neurodegenerative diseases.
Default Mode Network
SC FC
Pr: Precuneus
IP: Inferior parietal
PC: Posterior cingulate
AC: Anterior cingulate
Green lines
significant
decrease
in connectivity of
MCI vs Controls
Correlation between structural and
functional connectivity
Relationship between
neuropsychology and network scores
MEG-Functional connectivity vs p-TAU in the CSF
TAU concentration
has been associated
with neuronal damage
and cognitive
impairment
CSF
Frontal Inferior OrbLeft
Frontal Superior OrbRight
(R2=-0.52; p<0.05)
Frontal SuperiorLeft
Frontal Middle OrbLeft
(R2=-0.38; p<0.05)
Functional connectivity and
genetic profiles
lLIOC <PLV>
<PLV
>
ApoE Genotype Effect p = 0.049 (F = 4.03)
lLIOC Disrupted Links
lLIOC rTP
mPFClPFC
Delta Band APOE effect
carriers versus non
carriers of apoe 3/4
Multicenter clinical trial
Multicenter-
Clinical trial
Amiloid
Genetics
TAU
RMI
Neuro-psicology
DTI/MEG
Biomarkers
Clinical individual
AD Biomarkers
Biomarkers
exploitation
MCI/SMC
Resting/memory
analysis
MEG-based brain
conectivity analysis
Graph models
Multidimentional
Classification
Synchronization
MEG analysis
Clinical trial: MAGIC-AD
It has been examined differences in functional connectivity between
MCI and healthy controls with MEG at the group level.
- In order for the MEG-based biomarkers to be useful, it must be able
to detect abnormal function at the level of the individual patient.
-There were two goals to the present study:
- To develop a model, using data mining techniques, that reliably
distinguishes between MCI patients and healthy controls.
- Test this model using an unseen dataset of MCI and control
subjects acquired by the MAGIC-AD consortium.
MAGIC-AD: MEG International Consortium
for the study of Alzheimer’s Disease
Elekta-Neuromag supported the annual meeting of the
consortium
Obu – Japan
Pittsburg – USA
Salt Lake – USA
Houston – USA
Cambridge – UK
Helsinki – Finland
Brussels - Belgium
Madrid - Spain
DMN
MEG Preprocessing
102
Magnetometer
Sensor Data
Synchronization
Matrix
102
Magnetometer
Sensor Data
Synchronization
Matrix
102
Magnetometer
Sensor Data
Synchronization
Matrix
Severa
lepochs
(2 s
ec)
per
subje
ct
Aggregated
Synchronization
Matrix
One matrix per subject
Several statistics for each pair of
channels
Mean
Standard Deviation
Median
Median Absolute Deviation (MAD)
Variance coefficient
Mutual Information / Broadband
Resting state
data
.03-330 Hz
Analysis of MAGIC-AD DATA
Center for Biomedical
Technology
(Madrid, Spain)
Recordings of MEG
data
University of Pittsburgh
(USA)
University of Houston (USA)
University of Cambridge
(UK)
University of Helsinki (FN)
National Center for
Gerontology and Geriatrics
(Obu-Japan)
Blind
Data Mining
analysis
University of Utah (SLC,
USA)
Meeting to assess
results
Classification Phase• Random Forest [RF] (Breiman, 2001)
• Naïve Bayes Network [NB] (Buntine, 1991)
• C4.5 induction tree [C4.5] (Quinlan, 1993)
• K-Nearest Neighbour [KNN] (Cover and Hart,
1967)
• Logistic Regression [Logistic](Ng and
Jordan, 2002)
• Support vector machine [SVM] (Platt, 1999),
Information
gain filter
Features: 5150x5
statistics for a pairs
of channels
Subjects: 83 MCI
and 54 Control
1 subject
for
validation
146
subjects for
trainingMulti
Boruta
Classifier
Validation: Internal
validation (leave-one-out
validation scheme)
Classification
algorithms: State-of-the-
art data mining
techniques.
Validation
Bootstrapx147 repetitions
Results of the data mining blind
classification. MAGIC-AD DATA
External validation
Real class
Predicted class MCI Control
MCI 12 4 75,00% PPV
Control 1 11 91,67% NPV
92,31% 73,33% 82,14% Accuracy
Sensitivity Specificity
(Jack et al, 2010; Sperling et al, 2011)
MEG-based
Functional
ConnectivityYears