Premium
Neuronal network hyperactivity in computational models of AD
Author(s) -
de Haan Willem,
van Nifterick Anne M.,
Gouw Alida A.,
Stam Cornelis J.,
Scheltens Philip
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.040407
Subject(s) - neurophysiology , neuroscience , biological neural network , computational model , electroencephalography , network model , computer science , network dynamics , artificial neural network , excitatory postsynaptic potential , inhibitory postsynaptic potential , psychology , artificial intelligence , mathematics , discrete mathematics
Background Neuronal excitation/inhibition imbalance is a core feature of early Alzheimer’s disease (AD). Although non‐invasive clinical diagnostic methods have not yet reported reliable detection of this phenomenon, MEG studies do report transient increases in oscillatory power and functional network connectivity. However, the relation between these phenomena at different scales is unclear. Computational modeling can be employed to simulate multiscale brain dynamics, relating abnormal neuronal behaviour to brain network impairment. This way, a macroscopic EEG ‘signature’ of underlying neuronal hyperactivity can be predicted. Method Our dynamic brain network model consists of 78 neural mass models (representing AAL‐atlas brain regions) coupled according to human DTI‐based structural network topology. By altering neural mass parameters, the level of excitability of inhibitory and/or excitatory neurons can be tuned, in order to assess their influence on overall oscillatory network behavior (absolute and relative power, peak frequency and spike density). The model produces EEG‐like neurophysiological data that can be analyzed the same way as patient data. We compared model outcomes with clinical data of amyloid‐positive (CSF) MCI patients. Result Recent computational brain network modeling experiments show a strong effect of inhibition on oscillatory activity as compared to excitation, in line with existing literature. The scenario most compatible with MCI patient data is ‘global decreased inhibition’, producing a decrease in peak frequency and lower relative alpha power, and higher total and relative theta power. This scenario also resulted in the highest spike density levels. Conclusion This preliminary study illustrates how computational brain network modeling can aid to construct a coherent framework for the neuropathophysiological findings in AD at different scales of investigation. Although further systematic exploration is needed, computational brain network modeling in AD can become a powerful tool for both research and clinical purposes.