Graph theoretical analysis of Alzheimer's disease: Discrimination of AD patients from healthy subjects
Introduction
Human brain is a complex network comprised of a number of nodes (brain regions) connected through anatomical/functional links. Modelling the brain as a networked structure and studying its static and dynamic properties has received significant attention in recent years, which is mainly due to recent advances in network science tools and data processing techniques [8]. Data recorded from Electroencephalography (EEG), Magnetocephalography (MEG) and functional Magnetic Resonance Imaging (fMIR) can be used to extract functional brain networks [2], [16], [34], while anatomical brain networks can be extracted through Diffusion Tensor Imaging (DTI) technique [12]. In order to extract functional connectome of the brain, the first step is to obtain weighted connectivity matrices from EEG, MEG or fMIR data. To this end, linear techniques such as Pearson correlation and coherence analysis or nonlinear methods such as synchronization likelihood can be used. Often, the weighted connectivity matrices are binarized to obtain reliable network measures. To binarize a weighted network, one can use methods such as sparsity thresholding [1] or minimum spanning three [35].
Studying network properties in various brain disorders have revealed disease-specific abnormalities. Disorders such as schizophrenia [18], Alzheimer's disease [34] and epilepsy [6] show altered functional networks. AD symptoms are the most cause of dementia in the brain, often leading to the death. Previous studies showed that functional and anatomical networks in AD patients demonstrate altered network properties such as average path length, clustering coefficient, small-worldness and synchronizability [10], [33], [34], [35], [40]. EEG, as a non-invasive and cheap neuroimaging modality, has been used to study the AD mechanisms in the brain [3], [5], [24]. EEG-based networks have been studied in AD. For example, Stam et al. studied AD networks extracted from EEG and found that AD networks show longer shortest path length than healthy controls in beta band [34]. The clustering coefficient also altered in AD networks [40]. Tahaei et al. showed decreased synchronizability (as measured by the eigenratio of the Laplacian matrix of the connection graph) in AD subjects [35]. Afshari and Jalili studied directed networks in AD brains and showed abnormalities in global and local efficiency measures [4].
Previous studies on EEG-based brain functional networks were mainly in resting-state eyes-closed condition or when the subjects were performing a cognitive task. In this work, we study the resting-state with both eyes-open and eyes-closed conditions. We consider EEGs recorded from 25 healthy control subjects and 23 patients suffering from AD and analyse the network properties in different frequency bands. EEG signs have been previously used to classify AD patients from healthy control subjects [26,32,36]. We use EEG-based graph theory metrics to discriminate AD patients from healthy controls, which to the best of our knowledge, has not been studied in previous research works. Previous research works have used graph theory features of AD networks obtained from fMRI or MEG to classify AD from healthy controls. Zanin et al. studied MEG-based networks in Mild Cognitive Impairment (MCI), which is known to be early stages of AD development [39]. They showed that one can correctly classify MCI patients from healthy subjects with an accuracy of about 80% for networks with small to medium density values. By studying fMIR-based networks of MCI patients, Wang et al. showed significant decrease of global and functional connectivity in MCI [38]. Jie et al. also studied fMRI-based functional networks of MCI patients and obtained an accuracy of 89% using multi-kernel Support Vector Machines (SVM) classifier [21].
In this work, a number of graph theory features of functional networks are used to classify AD patients from healthy subjects. Also, a number of feature selection methods are used to choose a set of optimal features for the classification task. The feature selection algorithms are population-based optimization methods, which have been frequently applied to this problem [27], [30]. We find that the AD patients have statistically different local connectivity than healthy controls in the eyes-closed condition. Furthermore, SVM with optimal features can correctly predict up to 83% of the subjects.
Section snippets
Subjects and EEG recording
The EEGs of 23 newly diagnosed patients suffering from AD symptoms and 25 healthy controls are considered in this study. The subjects were recruited from the Memory Clinic of the Neurology Department (CHUV, Lausanne). The AD and control groups are not different in their age and educational level. AD patients show significantly less Mini Mental Test Examination (MMSE) scores than control subjects. Table 1 shows demographic information of the subjects used in this study. All the patients,
Results and discussion
Figs. 1–8 compare AD and healthy control subjects in terms of different graph theory metrics. The networks show stable properties across the frequency bands. AD brains show significantly (almost 10%) decreased local efficiency in the eyes-closed condition, but the decrease in the eyes-open condition is not significant (Fig. 1). The changes in the transitivity profile are not statistically significant in either of the conditions. Although the global connectivity measures (global efficiency,
Conclusion
The human brain can be modelled as a complex network where the nodes represent distinct brain regions and the links represent functional/anatomical connectivity between them. To construct brain networks, one can use brain neuroimaging modalities such as EEG, EMG, fMRI and DTI. In this work we considered EEGs recorded from 23 AD patients and 25 healthy control subjects in two conditions: resting-state eyes-open and eyes-closed. The network structures were extracted by applying Pearson
Acknowledgments
The author would like to thank Dr. Maria G. Knyazeva for EEG recording and preprocessing, and Homayoun Hamed Moghaddam Rafati for his help in implementing the feature selection and classification algorithms. This research was partially by Australian Research Council through grant number DE140100620.
Mahdi Jalili received his B.S. degree in electrical engineering from Tehran Polytechnique in 2001, his M.S. degree in electrical engineering from the University of Tehran in 2004, and his PhD from Swiss Federal Institute of Technology Lausanne (EPFL) in 2008. He then joined Sharif University of Technology as assistant professor. He is now with the School of Electrical and Computer Engineering, RMIT University, Melbourne, Australia and holds Australian Research Council DECRA (Discovery Early
References (40)
- et al.
Perception-related EEG is more sensitive to Alzheimer's disease effects than resting EEG
Neurobiol. Aging
(2016) - et al.
Network analysis for a network disorder: the emerging role of graph theory in the study of epilepsy
Epilepsy Behav.
(2015) - et al.
Membership-margin based feature selection for mixed type and high-dimensional data: theory and applications
Inf. Sci.
(2015) - et al.
EEG brain functional networks in schizophrenia
Comput. Biol. Med.
(2011) - et al.
Towards scalable fuzzy-rough feature selection
Inf. Sci.
(2015) - et al.
Topography of EEG multivariate phase synchronization in early Alzheimer's disease
Neurobiol. Aging
(2010) - et al.
Feature selection with partition differentiation entropy for large-scale data sets
Inf. Sci.
(2016) Binary social impact theory based optimization and its applications in pattern recognition
Neurocomputing
(2014)- et al.
A hybrid particle swarm optimization for feature subset selection by integrating a novel local search strategy
Appl. Soft Comput.
(2016) - et al.
A discrete particle swarm optimization method for feature selection in binary classification problems
Eur. J. Oper. Res.
(2010)
Disrupted functional brain connectome in individuals at risk for Alzheimer's disease
Biol. Psychiatry
Efficiency and cost of economical brain functional networks
PLoS Comput. Biol.
A resilient, low-frequency, small-world human brain functional network with highly connected association cortical hubs
J. Neurosci.
EEG coherence in Alzheimer's dementia
J. Neural Transm.
Directed functional networks in Alzheimer's disease: disruption of global and local connectivity measures
IEEE J. Biomed. Health Inform.
Pattern Recognition and Machine Learning
The economy of brain network organization
Nat. Rev. Neurosci.
Structural covariance of superficial white matter in mild Alzheimer's disease compared to normal aging
Brain Behav.
Functional neural network analysis in frontotemporal dementia and Alzheimer's disease using EEG and graph theory
BMC Neurosci.
Swarmed Feature Selection
Cited by (50)
Structural connectivity modifications in the brain of selected patients with tumour after its removal by surgery (a case study)
2023, Physica A: Statistical Mechanics and its ApplicationsIntraoperative enhancement of effective connectivity in the default mode network predicts postoperative delirium following cardiovascular surgery
2023, Future Generation Computer SystemsLongitudinal study of irradiation-induced brain functional network alterations in patients with nasopharyngeal carcinoma
2022, Radiotherapy and OncologyFeatures of the superficial white matter as biomarkers for the detection of Alzheimer's disease and mild cognitive impairment: A diffusion tensor imaging study
2022, HeliyonCitation Excerpt :Recently, the study of network properties in Alzheimer's disease has attracted the attention of several researchers. Among these researchers are Daianu et al. [25], Seo et al. [26], Jalili et al. [27], Sheng et al. [28] and Sulaimany et al. [29]. They believe that brain network connectivity analysis provides a significant understanding of how neural pathways break down in Alzheimer's disease.
Exploring Alzheimer's disease: a comprehensive brain connectome-based survey
2024, Psychoradiology
Mahdi Jalili received his B.S. degree in electrical engineering from Tehran Polytechnique in 2001, his M.S. degree in electrical engineering from the University of Tehran in 2004, and his PhD from Swiss Federal Institute of Technology Lausanne (EPFL) in 2008. He then joined Sharif University of Technology as assistant professor. He is now with the School of Electrical and Computer Engineering, RMIT University, Melbourne, Australia and holds Australian Research Council DECRA (Discovery Early Carrier Research Award) Fellowship and RMIT's Vice-Chancellor Research Fellowship. His research interests are in network science, dynamical systems, social networks analysis and mining, and human brain functional connectivity analysis. Dr. Jalili is a senior member of IEEE and associate editor of IEEE Canadian Journal of Electrical and Computer Engineering and an editorial board member of Complex Adaptive Systems Modelling.