Alzheimer's disease (AD) is the leading form of dementia worldwide. AD disrupts neuronal pathways and thus is commonly viewed as a network disorder. Many studies demonstrate the power of functional connectivity (FC) graph-based biomarkers for automated diagnosis of AD using electroencephalography (EEG). However, various FC measures are commonly utilised, as each aims to quantify a unique aspect of brain coupling. Graph neural networks (GNN) provide a powerful framework for learning on graphs. While a growing number of studies use GNN to classify EEG brain graphs, it is unclear which method should be utilised to estimate the brain graph. We use eight FC measures to estimate FC brain graphs from sensor-level EEG signals. GNN models are trained in order to compare the performance of the selected FC measures. Additionally, three baseline models based on literature are trained for comparison. We show that GNN models perform significantly better than the other baseline models. Moreover, using FC measures to estimate brain graphs improves the performance of GNN compared to models trained using a fixed graph based on the spatial distance between the EEG sensors. However, no FC measure performs consistently better than the other measures. The best GNN reaches 0.984 area under sensitivity-specificity curve (AUC) and 92% accuracy, whereas the best baseline model, a convolutional neural network, has 0.924 AUC and 84.7% accuracy.

This document is the author's post-print version, incorporating any revisions agreed during the peer-review process. Some differences between the published version and this version may remain and you are advised to consult the published version if you wish to cite from it.

Alzheimer's disease (AD) is a neurodegenerative disorder known to affect functional connectivity (FC) across many brain regions. Linear FC measures have been applied to study the differences in AD by splitting neurophysiological signals such as electroencephalography (EEG) recordings into discrete frequency bands and analysing them in isolation from each other. We address this limitation by quantifying cross-frequency FC in addition to the traditional within-band approach. Cross-bispectrum, a higher-order spectral analysis approach, is used to measure the nonlinear FC and is compared with the cross-spectrum, which only measures the linear FC within bands. This work reports the first use of cross-bispectrum to reconstruct a cross-frequency FC network where each frequency band is treated as a layer in a multilayer network with both inter- and intra-layer edges. An increase of within-band FC in AD is observed in low-frequency bands using both methods. Bispectrum also detects multiple cross-frequency differences, mainly increased FC in AD in delta-theta coupling. An increased importance of low-frequency coupling and decreased importance of high-frequency coupling is observed in AD. Integration properties of AD networks are more vulnerable than HC, while the segregation property is maintained in AD. Moreover, the segregation property of γ is less vulnerable in AD, suggesting the shift of importance from high-frequency activity towards low-frequency components. The results highlight the importance of studying nonlinearity and including cross-frequency FC in characterising AD. Moreover, the results demonstrate the advantages and limitations of using bispectrum to reconstruct FC networks.

No abstract