=Paper=
{{Paper
|id=Vol-3741/paper42
|storemode=property
|title=Integrating Brain Networks and Multi-Modal Data for Early Detection of Alzheimer’s Disease
|pdfUrl=https://ceur-ws.org/Vol-3741/paper42.pdf
|volume=Vol-3741
|authors=Carmela Comito,Clara Pizzuti,Marcello Sammarra,Annalisa Socievole
|dblpUrl=https://dblp.org/rec/conf/sebd/ComitoPSS24
}}
==Integrating Brain Networks and Multi-Modal Data for Early Detection of Alzheimer’s Disease==
https://ceur-ws.org/Vol-3741/paper42.pdf
Integrating Brain Networks and Multi-Modal Data for
Early Detection of Alzheimer’s Disease
Carmela Comito1 , Clara Pizzuti1 , Marcello Sammarra1 and Annalisa Socievole1
1
Institute for High Performance Computing and Networking, Via P. Bucci 8-9/C, Rende, 87036, Italy
Abstract
Early diagnosis of Alzheimer’s disease (AD) is crucial for providing timely treatment and care to patients.
However, current diagnostic methods rely on clinical symptoms and biomarkers, which are often
unreliable and invasive. Brain networks model the brain’s structure and function in AD and other brain
diseases. To fully capture their complexity, we need multi-modal models that combine different types of
data, such as structural and functional connectivity, clinical and genetic information. This gives us a
holistic view of the disease’s many aspects. In this paper, we argue that brain networks and multi-modal
data fusion can improve early diagnosis of AD by capturing the complex and heterogeneous nature of the
disease. Using brain network modeling and multi-modal data fusion, we envisage a novel framework for
detecting AD and its prodromal stages. The framework can simultaneously capture network properties
from multi-modal as well as longitudinal datasets, which provide complementary information.
Keywords
Alzheimer’s Disease, Brain Networks, Multi-modal data fusion, Artificial Intelligence, Prediction, Pro-
gression
1. Introduction
Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that affects millions of
people worldwide. It is characterized by cognitive impairment, memory loss, and behavioral
changes. AD is the most common cause of dementia and the sixth leading cause of death among
adults. Early diagnosis of AD is crucial for providing timely treatment and care to the patients,
as well as for reducing the social and economic burden of the disease.
The development of AD typically occurs in three primary phases. In the initial stage, known
as pre-clinical AD, changes in the brain may initiate without observable symptoms, making
detection of the disease challenging. Subsequently, in the second phase, referred to as mild
cognitive impairment (MCI), individuals and their families may start noticing symptoms related
to cognitive abilities, although these may not significantly affect daily functioning. Notably, not
all individuals diagnosed with MCI progress to AD.
Research on Alzheimer’s primarily focuses on identifying biomarkers capable of diagnosing
SEBD 2024: 32nd Symposium on Advanced Database System, June 23–26, 2024, Villasimius, Sardinia, Italy
∗
Corresponding author.
†
These authors contributed equally.
Envelope-Open carmela.comito@icar.cnr.it (C. Comito); clara.pizzuti@icar.cnr.it (C. Pizzuti); marcello.sammarra@icar.cnr.it
(M. Sammarra); annalisa.socievole@icar.cnr.it (A. Socievole)
Orcid 0000-0001-9116-4323 (C. Comito); 0000-0001-7297-7126 (C. Pizzuti); 0000-0002-7196-7994 (M. Sammarra);
0000-0001-5420-9959 (A. Socievole)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
the disease and predicting its progression. Various measurements have been developed and
assessed for detecting AD, typically encompassing physical health examinations, neuropsycho-
logical assessments, and brain imaging techniques. Biomarkers are usually categorized into
physiological, cognitive, behavioral, and psychological domains. Among the non-cognitive tests,
neuroimaging techniques are commonly utilized, while the Mini Mental State Examination
(MMSE) [1, 2] stands out as one of the most widely recognized cognitive assessments.
However, current diagnostic methods rely on clinical symptoms and biomarkers, which are
often unreliable and invasive. Clinical symptoms are subjective and vary across individuals
and stages of the disease. Biomarkers, such as cerebrospinal fluid (CSF) and amyloid-beta (Aβ),
require invasive procedures and expensive equipment. Moreover, both clinical symptoms and
biomarkers are not sensitive enough to detect the early and prodromal stages of AD, such as
mild cognitive impairment (MCI) and subjective cognitive decline (SCD).
To capture the various symptoms of AD, including subtle changes that occur throughout the
progression of the disease, there’s widespread agreement that a reliable method for detecting
early-stage Alzheimer’s disease cannot rely solely on measurements from one source. Instead,
it should employ a multi-modal approach by combining different types of biomarkers. Each
type of data reveals unique aspects of the condition, and integrating them all provides a more
comprehensive understanding, ultimately improving diagnostic accuracy.
Over the past few years, the progress in Artificial Intelligence-based methods for analyzing
multi-modal data has fueled research seeking new approaches for early disease detection. In
Comito et al. [3], an overview of the most recent approaches leveraging machine learning
and deep learning techniques for the prediction of Alzheimer’s disease by exploiting the huge
amount of multi-modal data now made available from the public repositories mentioned above
to researchers has been presented. However, current research does not consider an important
modality that recently has attracted the interest of many researchers: the brain connectome
introduced by Sporns et al. [4] in 2005. In particular, in the last ten years, the investigation of
AD progression and early detection has shown promise through the concurrent utilization of
advanced neuroimaging methods and complex network theory. By creating a brain network
from imaging data and representing it using network graphs allows to capture the complex and
dynamic interactions among brain regions, and reflect the changes in the brain structure and
function due to AD. Several studies have shown that brain network analysis can provide useful
insights into the pathophysiology and progression of AD, and can discriminate between AD and
normal aging. Analyzing the complex network of the human brain provides valuable insights
into its structural organization. This allows for the identification of abnormal interaction
patterns or irregularities in the modular structure of brains affected by AD.
The objective of this paper is to explore new avenues and methods to integrate brain networks
within multi-modal learning architecture to improve early diagnosis of Alzheimer’s disease.
To this purpose, the paper proposes a novel framework for detecting AD and its prodromal
stages using brain network modeling and multi-modal data fusion. Unlike existing methods,
our framework can simultaneously capture network properties from multi-modal as well as
longitudinal datasets, which provide complementary information. We use network models to
represent the structural and functional connectivity of the brain regions, and integrate multiple
types of data, such as images, text, audio, etc., to capture the complex and heterogeneous nature
of AD.
Figure 1: Data modalities used by the AI methods reviewed in the paper.
The paper is organized as follows. In Section 2, a summary of the current state-of-the-art
multi-modal approaches for AD prediction overviewed in [3] is reported. Section 3 discusses
brain networks, highlighting the most relevant trends of the network-based models. Section
4 presents the proposed framework integrating brain networks within multi-modal learning
architectures. Finally, Section 5 concludes the paper.
2. Multi-modal Approaches for AD Prediction
This section summarizes the main data modalities and AI methods utilized in Alzheimer’s
disease research reviewed in [3].
As far as data modalities are concerned (see Figure 1), neuroimaging (i.e. different tipology
of MRI and PET scans) stands out as the most prevalent data modality due to its non-invasive
nature, the availability of large datasets, the advancement of robust AI tools capable of extracting
significant features and classifying images, and the possibility of analyzing both structural and
functional brain anomalies and changes in AD patients.
Following neuroimaging, biological and genetic markers (APOE-e4, SNPs, CSF) are predomi-
nantly used to identify individuals likely to develop AD. Actually, certain genes and CSF markers
have been linked to an increased risk of the disease. Neuropsychological and cognitive assess-
ment tests are the third most common, primarily serving as screening tools to pinpoint those
requiring further evaluation. Lastly, demographic and clinical data, including blood markers,
are less frequently employed in the classification and progression tracking of AD.
The complexity and diversity of the data involved in AD research are, in turn, reflected in
the use of various AI techniques to classify the stages of the disease and predict its progression.
Traditional ML classifiers like SVM, DT, GB, RF, and LOR are widely used (see Figure 2),
particularly when clinical, demographic, and cognitive data are adopted. On the other hand,
DL methods, including NN, CNN, and RNN, show promise in medical image analysis, which is
crucial for AD diagnosis and monitoring. However, there is a remarkable number of AI methods
(referenced in Figure 2 as “OTHERS”) that are used only in one or two approaches. This
highlights that the application of AI in AD prediction and progression is indeed a multifaceted
field.
The mixed outcomes of the research in this field suggest that there is no one-size-fits-all
solution, and the choice of technique may depend on the specific dataset and task at hand.
Moreover, the need for careful parameter tuning, data selection, and experimental settings
cannot be overstated, especially when dealing with limited data availability, which is a common
challenge in AD research. The exploration of ensemble neural networks and the comparison of
various ML models underscore the ongoing efforts to refine predictive models for AD.
Overall, the continuous evolution of AI methods in AD prediction and progression demon-
strates the dynamic nature of the field and the potential for AI to contribute to our understanding
and management of this complex disease.
Figure 2: AI Methods used for Alzheimer’s Disease Prediction and Progression.
3. Brain networks
In the last decade, the study of the progression of Alzheimer’s disease and its early detection
has provided promising results from the joint use of advanced neuroimaging techniques and
complex network theory [5]. The construction of a brain network from an imaging model
through the formalism of network graphs has significantly improved the understanding of how
the brain of an AD patient behaves. The complex networks-based analysis of the human brain
provides better insights into the network structure, thus uncovering abnormal patterns of
interactions or randomness in the modular structure of an AD-infected brain.
Graphs are a mathematical model widely used for studying complex systems. In the literature
on brain networks and neurodegenerative diseases in general, there are different ways to
formalize and model the brain through a graph. More formally, the entities of a brain and their
relationships can be represented with a brain network 𝐵𝑁 modeled as a graph 𝐺 = (𝑉 , 𝐸, 𝑊 )
where 𝑉 is a set of 𝑛 objects, called nodes or vertices, 𝐸 ⊆ 𝑉 × 𝑉 is a set of links, called edges,
that connect two elements of 𝑉, and 𝑊 ∶ 𝑉 × 𝑉 → 𝑅 is a function which assigns a weight to a
couple (𝑖, 𝑗) of nodes 𝑖 and 𝑗, if there exists an edge connecting 𝑖 and 𝑗, and 0 if an edge between 𝑖
and 𝑗 does not exist. In almost all the AD-related studies, the nodes of the graph are usually
brain regions, while the edges may capture different relationships between regions [6, 7]. We
therefore classify brain networks and study their connectivity accordingly, as follows.
• Anatomical/structural brain networks: constructed from structural magnetic reso-
nance imaging (MRI), the edges represent physical connections between regions (e.g., the
estimated white matter connection strength in terms of number of fibers between any
pair of brain regions [7], interregional similarity [8] [9], etc.).
• Functional brain networks: the edges capture the functional interaction (magnetic,
electrical, or hemodynamic/metabolic) between brain regions that are not necessarily
adjacent or physically connected. These networks are usually constructed from imaging
models like functional magnetic resonance imaging (fMRI), electroencephalography
(EEG), and magnetoencephalography (MEG). In recent studies, resting-state functional
magnetic resonance imaging (rs-fMRI) has also been used in AD progression studies. This
imaging technique evaluates the BOLD (Blood oxygenation level-dependent) signal in
various regions of the brain. Its fluctuations, together with other functional connectivity
alterations, are used as AD biomarkers.
• Cortical thickness networks: these are hybrid networks based on structural data with
functional-like edges representing correlations between regions.
• Directed progression networks (DPNets): closely related to cortical thickness networks,
they attempt to capture the temporal progression of the disease, more than the correlations
between regions, similar to the epidemic network models where an edge represents the
spreading of the disease. In this network formalism, the edges are directed and capture
the degree to which one brain region thinning precedes the second region thinning.
To characterize network topology, thresholding sparsification techniques are usually applied to
structural or functional connectivity matrices to remove edges with noisy weights [10].
Even if these types of networks are clearly related, the comparison between them and their
joint analysis with the goal of having a systemic and multi-layer view of the disease progression
is not straightforward [7].
An important aspect that the study of brain networks has highlighted in recent years is the
detection of building blocks, i.e. the presence of overabundant small subgraphs sharing patterns
of interconnections, called network motifs, occurring with a frequency higher than that in a
random network [11]. Network motifs have been recognized as fundamental building blocks
of networks [11] giving insights into the functional mechanisms of the analyzed system, and
revealing different organization models of the same network.
A 𝑚𝑜𝑡𝑖𝑓 of a graph 𝐺 is defined as an unordered subset 𝑀 = {𝑣1 , … , 𝑣𝑟 } of nodes of 𝐺 presenting
a particular pattern of interconnections. Fig. 3 shows five types of motifs among three and four
nodes (Fig. 3(a)-Fig. 3(e)). Their labeling follows the same convention adopted in [12]. The
bi-fan motif, for example, is over-expressed in neuronal networks.
Motifs have been largely analyzed in brain networks, especially in structural and functional
networks. In [13], Meier et al. exploit motifs for clustering the functional brain network built
on MEG data. This type of network, called effective connectivity network, is composed by ROIs
linked by their effective connectivity, a measure describing the causal effect of one brain region
ve
nda-
haveproven
proven funda-
funda-
ional
tion regulation
iptional
nscriptional regulation
regulation
. are
)fs1A,
Fig. M11A,
1A,
(Fig. M–M 7)1are
1–M
M 7) are
–M 7) are R ES E A RC H | R E PO R TS
en
open
onal
4); bidirectional
openbidirectional
bidirectional
structural
o structural
hubs
ey hubs
to structural hubs hubs
p1A,
paths
paths
-hop (Fig.
(Fig.
paths 1A,1A,1A,
(Fig.
rstanding
anding
traf-
derstandingairair traf-
traf-
air traf-
work
have
network
rk motifsmotifs
motifs
havehave have (a) (b) (c) (d) (e)
net-
mental
ntal
al units
units
units of of of
net-
net- net-
Figure 3: (a) 𝑀4 , (b) 𝑀5 (feed-forward loop), (c) 𝑀6 , (d) 𝑀8 , (e) 𝑀𝑏𝑖𝑓 𝑎𝑛 motifs.
orksof of
anization
zation
ion of networks
networks
networks
fs
s an
gely largely
argely remains
remains
remains anonan
an
another region. To calculate this pairwise value between brain regions, the measure of
Phase Transfer Entropy (PTE) is applied. Here, some 3 and 4 nodes motifs are over-expressed,
r network
ures moreover, motif-based clustering reveals a strong symmetry between the two hemispheres,
structures
etwork
work structures
structures supporting the idea of a higher-order organization of the effective connectivity brain network.
organization
om-
ganization of of
com-com- et al. [7] analyze the presence of motifs spanning over a two-layer brain network
nization of com-Battiston
ramework
den-
mework
work that
that thatof iden-
healthy subjects composed of a structural and a functional layer. The multiplex two-layer
iden-
iden-network is built by forming a first structural network layer with DW-RMI data (Diffusion
tifs.
work
orFor For
eacheacheach network
network
network Magnetic RMI) and a second functional layer with rs-fMRI data. The motifs analyzed are
gher-order
ring clustering
er-order
order clustering
multiplex, which is statistically overabundant small subgraphs spanning over the two layers
clustering
and jointly considered. Results show that motifs in whichFig.links between brain
C. elegansregions at[Seethe
hich
hdif- means
means that
means that dif- that
dif-dif- 2. Higher-order cluster in the neuronal network.
(A) The four-node bi-fan motif, which is overexpressed in neuronal networks
structural layer also have a functional activity are frequent, but there are also frequent multiplex exposes their organization.The clus
(1). Intuitively, this motif describes a cooperative propagation of
(29).]
information
have been proposed as pioneers
neurons are known regulators of n
rns
de-
are are exposed,
exposed,
re exposed, de- motifsde-
de- having distant regions with strong functional dynamics.
from the nodes on the left to the nodes on the right. (B) The best higher-order
cluster in the
The cluster
Finally,
contains three
the
C. elegans frontal
ring
analysis
neuronal
motor
shows
network based
neurons (RMEL, -V,
on the motif
and -R;
the
cyan)
in (A).
with
intermediary of information proces
three RME neurons, outgoing conn
the largest total number of conne
existence of a reinforcement mechanism between the structural many outgoingand the functional layers for
tif. many incoming
connections,
connections,
which
serving
serve as the source
as the destination of
of information;
inner labial sensory neurons (IL2DL, -VR, -R, -DR, -VL, and -L; orange) with
which the probability of having a link is related to the intensity of the connection in the other Most information; and
six
four
lustration of the higher-order cluste
locations are the true two-dimens
information flows from left to
twork
our motif M,
ourour
layer.
URA motor neurons (purple) acting as intermediaries. These RME neurons RIH serve as sources of informatio
kork motif
motif M,M, our • Step 2: Compute the spectral ordering s of in the network, in practice, the algorithm is much tion) (1
uster
with
er of nodes S with
Recent studies have highlighted that network motif analysis the nodes canfrom provide
the normalized new markers
motif Laplacian faster. for the16 real-world networks where higher-
By analyzing
of of nodes
nodes S with
S with diagnosis and progression of Alzheimer’s disease. Friedman
matrix constructed via W (15).
et al.
• Step 3: Find [14] were
the prefix set ofable
s with to
the number of edges m ranges from 159,000 to
M
the distinguish
2 billion, we found the computational com-
knowle
the sup
nshould
S should
pate participate
participate
AD from HC patients with directed motif analysis of their
smallest motif conductance (the argument of the
DPNets. formally,InS : =addition, the
plexity to scale as Q(m ). Moreover, the algorithm
entropy of
1.2
when a
ould participate minimum),
S = {s , …, s }.
arg min f (S ), where can easily be parallelized,
niques can
r
be
M r
used to
and sampling tech-
further improve perform-
is not k
also ser
ond, the set S should
r 1 r
ould
d, the set S
the
should
motif distribution provides a new imaging marker for MCI.
For triangular motifs, the algorithm scales to ance (16). for the
he set S shouldIn the next section, a review of the most recent proposals
networks with billions of edges and, typically, only The framework can be applied to directed, un- (13). Su
takes
studying brain networks
several hours to process graphs of such size.
for the
directed, and weighted networks, as well as motifs higher-
M,hen
which occurs when
Fig. 1.diagnosis
Higher-order Fig. 1. Higher-order
network structures network and structures
the higher-order
On smaller networks with hundreds of thousands
and the networkhigher-orde
(13). Moreover, it can also be applied to networks
clust ber of
hich occurs when of Fig.
ch occurs 1. Higher-order
Alzheimer’s disease is reported. network
whenin theFig. 1. Higher-order network structures and the higher-order n structures
of edges, the algorithm can process motifs up to
size 9 (13). Although the worst-case computational andwith the higher-order
positive and negative signs on the edges, which
are common in social networks (friend versus foe
motifs
mathem
nomthea motif
(A)are (A) Higher-order
Higher-order structures are capturedstructures are captured
by network motifs.byFor network
example, motifs.
all 13For exampl
connected
complexity of the algorithm for triangular mo- or trust versus distrust edges) and metabolic net- spectra
a motif
motif areare
in in
thethe
3.1. (A) (A) Higher-order
Higher-order structures
structures are are captured
captured by by network
network motifs.
motifs. For For example
tifs is Q(m1.5), where m is the number of edges
example, a
works (edges signifying activation versus inhibi- that fin
ely,
f M,givenmotifs areNetwork-based
a motif M, here.
shown motifs (B)approaches
are shown here.
Clustering of a(B) Clustering
network basedofon a network
motif M 7based on motif
. For a given 164motif M 7M,. For
ou
given a motif M, ofmotifs motifsareare shown here. (B) Clustering of a network based on motifMM 7. For ga
8 JULY 2016 • VOL 353 ISSUE 6295
ven a motif
sframework
M, shown here.
nodes(B) Clustering of methods
a network based onthe
motif . For aw
to find aaims
setpartofto
As find
the Network
nodes aMedicine
S that setminimizes
of approaches, that
S the
motif minimizes
network-based
conductance, motif conductance,
fMare able
(S), to reveal
which fM(S),
we define 7which
as the ra
amework
ework aimsnetwork
aims to disruptions
find aorset of nodes S that minimizes motif conductance, offM(S), which wewe
motifsto find amotifs
set ofcutnodes S that minimizes motif conductance, M(S),
fof which ind
the structural changes initiated by the disease through the application
set of nodes
that S)
cutthat
(filled triangles cut) (filled
to thetriangles
minimum cut) to
number the minimum
of nodes number
in instances
methodologies proper of Network Science and Graph Theory, offering practical techniques able nodes
of the in instan
motif
tf of nodesS) S) that motifs
motifs cut cut (filled
(filled triangles
triangles cut) to the minimum number of nodes in in instanc
nodes thisthat
case, there
to tackle theis this
one case,
motif
limitations there
cut.
of the (C) onecut)
is The
consolidated motif tocut.
the
higher-order
biomarkers minimum
(C)
[5]. Themethods
network
These number
higher-order
clustering ofnetwork
nodes
framework.
are increasingly instance
clustering
Given a f
thiscase,
case, this case,
due to there is one motif cut. (C) The higher-order network clustering fra
Mthere isthis
onecase,
motif cut. (C) The higher-order network clustering fram
attracting interest their ability to simplify the understanding of the modeling of the
athematical
ng, interest (in this
Engineering,
disease at
interest
multiple
(inlevels
7), the
biological
framework
[15]. In
Mforms
the
), the aframework
7following, motif
we review
forms
adjacency
the most
amatrix
motif
relevant
adjacency
(W
works
matrix
M) by counting
matical
tical
ment Engineering,
2
4305,Engineering,
USA. Department interest
interest two (in (in
this
nodes
this
case,case,M7M
co-occur
), the
),of7the
in an
framework
framework
instance of
forms
formsthe
a motif
amotif.
motif adjacency
adjacency
An
matrix
matrix
eigenvector of (W
a
(
2 two nodes co-occur
in this area in
by reporting an instance
in Table 1 the typeof the
brain motif.
network An eigenvector
modeling. of a Laplacian transfor
,5,
INUSA. Department
2
USA.
rsity, Department
West Lafayette, IN two two nodes
nodes
adjacency co-occur
co-occur
matrix inis in
an
thenan instance
instance
computed.Theof of
thethe motif.
motif.
ordering AnsAn eigenvector
eigenvector
of the nodes of of
aa Lap
provided L
y, West
epartment,
adjacency
Lafayette,
Stanford IN matrix is then computed.The ordering s of the nodes provided by the componen
West Lafayette, IN • Dragomir and Vrahatis [16] review the current state-of-the-art in network-based biomark-
adjacency matrix is then computed.The ordering s of the nodes provided
tment, (15)
Stanford produces ersadjacency
nested
for preclinicalmatrix
(15) sets
produces is nested
r = {s
ADSdiagnosis then
1, by
computed.The
ssets
…,subdividing = {s
Sr the
r} of increasing
ordering
1, …,size
current sr} r.
trends We s of
ofinto the
increasing
prove
two nodes
size
that
research r.provided
the We
setprove
Sr witbyt
A. Stanford
ent,
(15)
lines: ADproduces
biomarkers
based nested
focusing
conductance, (15) produces nested sets S = {s , …, s } of increasing size r. We prove th
onsets fMS(Sr=
the modular {s
is1,a…, sr} atof
r substructures r a increasing
(1) molecular size
level We prove
(i.e.r. cluster tha
r),higher-order
1near-optimal higher-order (13).
based conductance,
cs.stanford.edu fM (S r),
blood omics-biomarkers
is
in
a near-optimal
molecular networks) and at a (2)
cluster
brain level
(13).
(i.e. volumetric
anford.edu
ord.edu based
based conductance,
conductance, Mr(S
fMf(S ), aisnear-optimal
), ris a near-optimal higher-order
higher-order cluster
cluster (13).
(13).
8 JULY 2016 • VOL835JU
8 JU
8 JULY
approaches and connectivity from MRI data on connectivity networks). In this context,
the benefits of monitoring the early stages of the disease by jointly considering genotype
and brain phenotype information as noninvasive biomarkers are discussed. Particular
emphasis is placed on methods that analyze the network substructure through methods
of complex network analysis since disruptions usually occur locally, in different brain
regions and molecular pathways, and at different rates. Finally, a framework integrat-
ing knowledge from the two information levels, molecular omics-based data collected
from blood samples and brain connectivity obtained from neuroimaging techniques, is
proposed, showing that this multi-level solution can further improve diagnosis.
• In [8], Tijms et al. analyze the structural brain network of gray matter extracted from
MRI images, and the clinical progression in nondemented subjects who have abnormal
amyloid markers in the cerebrospinal fluid (CSF), that is a marker of predementia AD.
The study investigates if network structural measures like size, connectivity density,
degree, clustering coefficient, path length, betweenness centrality, and the small-world
property are somehow associated with the rate of progression to MCI or dementia, using
Cox proportional hazard models to assess associations between the structural measures
and time to clinical progression. Results indicate that when these measures have low
values there is an increased risk of fast progression to MCI. In particular, lower clustering
values, indicative of a more randomly organized network, in specific anatomical areas are
associated with clinical outcomes and fast clinical progression.
• In recent work, [9] Ding et al. investigate the relationship between topological features
of gray matter morphological networks and the clinical cognitive performance of healthy
control subjects (HCs) and patients with SCD or MCI. Analyzing local graph measures, the
networks of SCD and MCI show a significant decrease of degree centrality in the caudate
neucleus and of nodal efficiency in the caudate neucleus, right insula, lenticular nucleus
and putamen. In terms of global topological measures, SCD and MCI patients show
lower values of path length, normalized path length, and global efficiency in their brain
networks. The study concludes that the topological features of the structural gray matter
network can be considered biomarkers that can improve AD prognosis and interventions
in its early stage.
• Friedman et al. [6] define and analyze the DPNets in AD, particularly directed brain
networks where an edge between two regions represents nor the physical connectivity
nor a functional connectivity but the temporal spreading of the pathology. The DPNets
are constructed by evaluating the change in cortical thickness measurements: when a
node A is thinning over time, it is considered infected with a certain probability and
may spread its infection to other nodes. A directed edge connects node A to a node B
if in a late period B shows a higher thinning rate (i.e. B has been ”infected” by A), with
a weight representing an infectious similarity (ISIM). By using several local and global
measures (degree, indegree, outdegree, size of the giant component, path length, global
efficiency, clustering coefficient, modularity and small world properties), the results show
that DPNets are able to classify AD patients looking at clustering (low) and small-world
property (low) values.
• Lama and Kwon [17] design an AD diagnosis approach able to classify subjects into AD,
MCI, or HC, modeling the brains as functional graphs and exploiting graph theory-based
Table 1
Type of brain networks graphs in the reviewed works of AD brain network analysis (GM= gray matter).
Study Brain network type Data # of nodes Node type Edge type
[8] structural MRI 62 SCD, 160 MCI GM 3x3x3 voxels interregional similarity
[9] structural MRI 39 SCD, 39 MCI. 26 HC GM 2x2x2 anatomical regions interregional similarity
[6] DPNet MRI 39 AD, 97 NC GM ROI with 0.94 x 0.94 x 1.2 voxels infectious similarity
[17] functional rs-fMRI 31 AD, 31 MCI, 31 HC GM ROI with 3.3 thickness voxels ROI correlation
[14] DPNet MRI 39 AD, 65 MCI, 54 CONV, 97 HC GM ROI with 0.94 x 0.94 x 1.2 voxels infectious similarity
features. The functional brain network is built by setting on each edge the Pearson’s
correlation functional connectivity between ROIs. Then, graph embedding (node2vec)
is used to transform graphs into a vector and applying machine learning techniques.
To classify the subjects into classes, different classification techniques including linear
support vector machines (LVSM) and regularized extreme machine learning (RELM) are
explored. The highest accuracy is obtained by combining LASSO with LSVM.
4. Fusing Brain Networks and Multi-modal data: Proposed
Learning Model
Brain networks are complex and intricate systems that reflect the brain’s structure and function
in various brain diseases, such as AD. Multi-modal data is essential to capture their complexity.
Multi-modal data integrates different types of data, such as structural and functional, con-
nectivity, clinical, and genetic information, to form a holistic understanding of the disease’s
multifaceted nature. Previous research has tried to fuse various modalities.
However, studies investigating the combination and the relationship of the brain connectome
with biomarkers and genetics are very few. Only recently, Yu et al. [10] pointed out the link
between the changes in structural and functional network organization in Alzheimer’s disease
brain and the accumulation of amyloid-𝛽 and tau in particular parts of the brain. Moreover,
Badhwar et al. [18] proposed a roadmap to fusing multiomics measurement for the diagnosis of
Alzheimer’s disease.
To bridge this gap we propose a novel multi-modal architecture that integrates brain network
with biomarkers and genetics data. Integrating brain networks and multi-modal data within a
learning framework for Alzheimer’s prediction is a promising approach that leverages diverse
sources of information to improve the accuracy and reliability of predictive models. Brain
networks provide a comprehensive representation of the brain’s structural and functional
connectivity patterns, offering valuable insights into the underlying neurobiology of Alzheimer’s
disease. Multi-modal data, on the other hand, encompasses various types of information such
as neuroimaging scans (e.g., MRI, PET), clinical assessments, genetic markers, and cognitive
scores, each providing unique perspectives on the disease.
Figure 4 shows the proposed learning framework. The model is crafted for a multi-modal
multitask objective, aiming to grasp Alzheimer’s disease progression and cognitive scores based
on a variety of data. As illustrated in Figure 4, the model initially processes data from five
modalities: neuroimaging, biomarkers, genomics, clinical, and demographics. Specifically,
Figure 4: Multi-modal Learning Framework
multimodal data including MRI scans, demographics, medical history, functional assessments,
and neuropsychological test results, are used to develop learning models on various classification
tasks. Local and temporal feature learning in the model is facilitated by exploiting machine and
deep learning cutting-edge methods. The features from various modalities (cognitive scores,
neuropsychological battery, MRI, PET, and assessment modalities) undergo preprocessing to
enhance data quality. AI techniques are utilized for feature reduction, extracting principal
components from high-dimensional MRI and PET data. For example, neuroimaging features
from MRI and PET modalities can be extracted using FreeSurfer. Subsequently, deep features
are independently learned from each modality using both ML and DL approaches like stacked
CNN-BiLSTM models. Abstract deep features obtained from the previous step are then fused to
extract common features from all modalities using a series of dense layers within a deep neural
network. As the final stage of the learning framework, a classification is produced.
By combining brain networks with multi-modal data in a learning framework, we can
capitalize on the complementary nature of these sources to enhance prediction accuracy and
understand the complex progression of AD. Here are some key points to consider:
• Feature Fusion. Integrating information from brain networks and multi-modal data
involves fusing features extracted from different sources into a unified representation. This
can be achieved through techniques such as feature concatenation, attention mechanisms,
or graph-based fusion methods.
• Learning Architecture. The learning framework should incorporate neural network archi-
tectures capable of processing multi-modal inputs and capturing complex relationships
within the data. This may involve the use of deep learning models such as convolutional
neural networks (CNNs), recurrent neural networks (RNNs), or graph neural networks
(GNNs) tailored to handle multi-modal data.
• Regularization and Adaptation. Given the high-dimensional and heterogeneous nature of
multi-modal data, regularization techniques such as dropout, and batch normalization,
can help prevent overfitting and improve generalization performance. Additionally, model
adaptation strategies may be employed to adapt the learning process to different data
modalities and patient cohorts.
• Evaluation and Validation. Robust evaluation metrics and validation procedures are
essential for assessing the performance of the learning framework. Cross-validation,
hold-out validation, and external validation on independent datasets can help validate
the generalizability of the predictive models.
• Clinical Interpretability. Interpretable models are crucial for translating predictive insights
into actionable clinical decisions. Techniques such as attention mechanisms, feature
importance analysis, and visualization methods can provide insights into the contribution
of different modalities to the prediction task and aid in clinical interpretation.
• The rapid growth of new AI approaches in the last decade has largely overlooked the
importance of computational efficiency in algorithm design and data generation. This has
led to the widespread adoption of complex AI techniques with high computational costs
and energy consumption. This is true also in medical applications, such as AD detection,
where improvements in accuracy come at the cost of increased data availability. The
development of novel algorithms able to deal with limited resources while maximizing
the quality of the results obtained is a main objective of the emerging so-called Green AI
methods. Considering the large volumes of multimodal data nowadays available for AD,
the design and development of new energy-aware AI techniques with low computational
needs and reduced data while reaching high predictive accuracy is a desirable objective
in the near future of AD research.
Overall, integrating brain networks and multi-modal data within a learning framework holds
great potential for advancing Alzheimer’s prediction research, leading to more accurate and
reliable predictive models that can aid in early diagnosis and personalized treatment planning
for patients.
5. Conclusion
In conclusion, early detection of Alzheimer’s disease is paramount for effective intervention
and patient management. However, existing diagnostic methods often fall short, relying heavily
on clinical symptoms and biomarkers that may not provide accurate or timely results. Brain
networks offer a promising avenue for understanding the complex structural and functional
changes associated with AD. To fully grasp this complexity, a multi-modal approach is neces-
sary, incorporating various data types such as structural and functional connectivity, clinical
assessments, and genetic information.
In this paper, we advocate for the integration of brain networks and multi-modal data fusion
to advance early AD diagnosis. By combining these approaches, we can better capture the
diverse and nuanced characteristics of the disease. Our proposed framework seeks to leverage
brain network modeling and multi-modal data fusion to develop a comprehensive understanding
of AD and identify prodromal stages. This innovative approach aims to extract network features
from diverse datasets, including longitudinal data, to enhance diagnostic accuracy and inform
timely interventions for individuals at risk of AD.
Acknowledgments
This work has been partially supported by the project ALCMAEON (F/260014/01-03/X51)
funded by MISE. We acknowledge the support of the PNRR project FAIR - Future AI Research
(PE00000013), Spoke 9 - Green-aware AI, under the NRRP MUR program funded by the NextGen-
erationEU.
References
[1] F. MF, F. SE, M. PR., Mini-mental state. a practical method for grading the cognitive state
of patients for the clinician, Journal of Psychiatric Research 12 (1975) 189–198.
[2] A. J. Mitchell, The Mini-Mental State Examination (MMSE): Update on Its Diagnostic
Accuracy and Clinical Utility for Cognitive Disorders, Springer International Publishing,
Cham, 2017, pp. 37–48.
[3] M. S. Carmela Comito, Clara Pizzuti, Overview of artificial intelligence methods for
alzheimer’s disease prediction and progression, in: SITIS Conference, 2023.
[4] O. Sporns, G. Tononi, R. Kötter, The human connectome: A structural description of the
human brain, PLoS Comput. Biol. 1 (2005).
[5] L. Zhang, J. Qu, H. Ma, T. Chen, T. Liu, D. Zhu, Exploring alzheimer’s disease: a compre-
hensive brain connectome-based survey, Psychoradiology (2024) kkad033.
[6] E. J. Friedman, K. Young, D. Asif, I. Jutla, M. Liang, S. Wilson, A. S. Landsberg, N. Schuff,
A. D. N. Initiative, Directed progression brain networks in alzheimer’s disease: properties
and classification, Brain connectivity 4 (2014) 384–393.
[7] F. Battiston, V. Nicosia, M. Chavez, V. Latora, Multilayer motif analysis of brain networks,
Chaos: An Interdisciplinary Journal of Nonlinear Science 27 (2017).
[8] B. M. Tijms, M. Ten Kate, A. A. Gouw, A. Borta, S. Verfaillie, C. E. Teunissen, P. Scheltens,
F. Barkhof, W. M. van der Flier, Gray matter networks and clinical progression in subjects
with predementia alzheimer’s disease, Neurobiology of aging 61 (2018) 75–81.
[9] H. Ding, Z. Wang, Y. Tang, T. Wang, M. Qi, W. Dou, L. Qian, Y. Gao, Q. Zhong, X. Yang, et al.,
Topological properties of individual gray matter morphological networks in identifying
the preclinical stages of alzheimer’s disease: a preliminary study, Quantitative Imaging in
Medicine and Surgery 13 (2023) 5258.
[10] M. Yu, O. Sporns, A. J. Saykin, The human connectome in alzheimer disease—relationship
to biomarkers and genetics, Nature Reviews Neurology 17 (2021) 545–563.
[11] R. Milo, S. Shen-Orr, S. Itzkovitz, N. Kashtan, D. Chklovskii, U. Alon, Network motifs:
Simple building blocks of complex networks, Science 353 (2002) 824–827.
[12] A. R. Benson, D. F. Gleich, J. Leskovec, Higher-order organization of complex networks,
Science 353 (2016) 163–166.
[13] J. Meier, M. Märtens, A. Hillebrand, P. Tewarie, P. Van Mieghem, Motif-based analysis
of effective connectivity in brain networks, in: Complex Networks & Their Applica-
tions V: Proceedings of the 5th International Workshop on Complex Networks and their
Applications (COMPLEX NETWORKS 2016), Springer, 2017, pp. 685–696.
[14] E. J. Friedman, K. Young, G. Tremper, J. Liang, A. S. Landsberg, N. Schuff, A. D. N. Initiative,
Directed network motifs in alzheimer’s disease and mild cognitive impairment, PLoS One
10 (2015) e0124453.
[15] E. Nagele, M. Han, C. DeMarshall, B. Belinka, R. Nagele, Diagnosis of alzheimer’s disease
based on disease-specific autoantibody profiles in human sera, PloS one 6 (2011) e23112.
[16] A. Dragomir, A. G. Vrahatis, A. Bezerianos, A network-based perspective in alzheimer’s
disease: Current state and an integrative framework, IEEE journal of biomedical and
health informatics 23 (2018) 14–25.
[17] R. K. Lama, G.-R. Kwon, Diagnosis of alzheimer’s disease using brain network, Frontiers
in Neuroscience 15 (2021) 605115.
[18] B. A., M. GP, S. S., B. SE, C. H., D. S., M. M., L. L., D. RA., B. P., A multiomics approach
to heterogeneity in alzheimer’s disease: focused review and roadmap, BRAIN 143 (2020)
1315–1331.