Brain simulation augments machine‐learning–based classification of dementia

Abstract

Introduction Computational brain network modeling using The Virtual Brain (TVB) simulation platform acts synergistically with machine learning (ML) and multi‐modal neuroimaging to reveal mechanisms and improve diagnostics in Alzheimer's disease (AD). Methods We enhance large‐scale whole‐brain simulation in TVB with a cause‐and‐effect model linking local amyloid beta (Aβ) positron emission tomography (PET) with altered excitability. We use PET and magnetic resonance imaging (MRI) data from 33 participants of the Alzheimer's Disease Neuroimaging Initiative (ADNI3) combined with frequency compositions of TVB‐simulated local field potentials (LFP) for ML classification. Results The combination of empirical neuroimaging features and simulated LFPs significantly outperformed the classification accuracy of empirical data alone by about 10% (weighted F1‐score empirical 64.34% vs. combined 74.28%). Informative features showed high biological plausibility regarding the AD‐typical spatial distribution. Discussion The cause‐and‐effect implementation of local hyperexcitation caused by Aβ can improve the ML–driven classification of AD and demonstrates TVB's ability to decode information in empirical data using connectivity‐based brain simulation.

Personalized virtual brains of Alzheimer’s Disease link dynamical biomarkers of fMRI with increased local excitability

“…These studies have demonstrated the value of this approach. Especially important results indicated in Triebkorn et al, 2022, showed the added values of the model for classification based on the structural and functional data features.…”

“…These models can also contain regionally variant parameters, when this is supported by clinical (Jirsa et al, 2017; Courtiol et al, 2020) or data informed (Kringelbach et al, 2020) hypotheses. In the context of AD, studies shifted their focus to linking local and global dynamics to the main driver of the disease to pathophysiology driven neuronal hyperactivity via The Virtual Brain platform (Sanz Leon et al, 2013; Zimmermann et al, 2018; Stefanovski et al, 2021; Triebkorn et al, 2022; Arbabyazd et al, 2021; Patow et al, 2022). These studies have demonstrated the value of this approach.…”

Whole-brain modeling of the differential influences of Amyloid-Beta and Tau in Alzheimer’s Disease

“…Following this approach, recent work by Stefanovski and co-authors [21] focused on the connection of A β with neural function in The Virtual Brain (TVB) platform [33], using a network of interconnected (through the corresponding structural connectivity matrix) Jansen-Rit models [34], addressing the phenomenon of hyperexcitability in AD, examining how A β burden modulates regional Excitation-Inhibition balance, leading to local hyperexcitation with high A β loads in the model, reproducing what has been previously observed in experimental studies. The resulting simulated local field potentials improved previous diagnostic classifications between AD and controls [22]. However, all these works study the effect of a single burden, namely A β , on the brain neuronal dynamics, while the work we present here focuses mostly on the interplay of both burdens, i.e., A β and tau, assessing their relative impacts on these dynamics.…”

“…They were able to reproduce what has been previously observed in experimental studies. The resulting simulated local field potentials improved previous diagnostic classifications between AD and controls [16]. However, all these works study the effect of a single burden, namely A β , on the brain neuronal dynamics, while the work we present here focuses mostly on the interplay of both burdens, i.e., A β and tau, assessing their relative impacts on brain dynamics.…”

“…While interesting results regarding E/I imbalance were derived ex-vivo (in humans), studies in-vivo regarding E/I imbalance in AD are lacking, as the activity of excitatory vs. inhibitory neuronal populations cannot be directly measured using neuroimaging. Most works focusing on whole-brain dynamics studied different measures of brain activation patterns, e.g., from its connectivity, but were not informative regarding the role of excitatory vs. inhibitory neuronal populations [18, 19, 20, 21, 22]. To disentangle mechanistic contributions of separate neuronal populations, whole-brain dynamic models can contribute to analyze collective properties of the brain [23, 24, 25], such as the fMRI signal [26, 27, 28, 29].…”