Nature BME: Modeling and prediction of dynamic response of large-scale brain networks during direct electrical stimulation

Direct electrical stimulation of the brain is a technique that regulates brain activity and helps treat various brain dysfunctions and promotes brain function.
, for example, deep brain stimulation (DBS) is effective for neurological disorders such as Parkinson's disease and epilepsy, as well as for neuropsychiasts such as chronic pain, incurable depression and obsessive compulsive disorder.
although the mechanism by which direct electrical stimulation alters brain activity is unclear, studies have shown that electrical stimulation alters activity in multiple brain regions (local and long-range) distributed over large brain networks.
the stimulation effects of this network level have been observed through various signal patterns, such as local field inductance (LFP), cortical electrogram (ECoG), functional magnetic resonance imaging (fMRI), and diffusion imaging (DTI).
observations highlight the need for simulation stimulation to affect large-scale multiregional brain network activity, which is largely impossible.
this modeling is especially important when the timing of stimuli needs to change in real time and when activity in multiple brain regions needs to be monitored.
, such as closed-loop DBS therapy for neurological and neuropsycological disorders.
to change stimulation patterns in real time based on feedback from changes in brain activity (e.g., the frequency and amplitude of the stimulus pulse sequence).
the ability to predict how continuous stimulation (input) drives the time evolution (i.e. dynamics) of large-scale multiregional brain network activity (output) remains a challenge.
In this paper, an enabling technique for precisely regulating brain function and dysfunction was developed, and the ability to predict the dynamic response of large-scale multi-region brain networks to duration changes was established in two sober rhesus monkeys.
using machine learning techniques to achieve this prediction by introducing and developing dynamic IO models of brain network activity.
use a customized semi-chronic micro-drive system to provide microstration of continuous time changes while recording through a large multi-region brain network.
use LFP power characteristics to measure the response of brain networks.
the model input as the excitation amplitude and frequency, changing the excitation amplitude and frequency in real time.
this paper develops a data-driven dynamic modeling and machine learning method to simulate the dynamic response of two macaques to brain network activity in short to microstrulating pulse sequences.
as an output, the LFP power feature time series and four bands -1-8 Hz (δ-Hz), 8-12 Hz (α), 12-30 Hz (β), 30-50 Hz (low gamma) are calculated from multiple brain regions after the stimulus artifacts are removed.
referred to each LFP power feature as a network node.
the amplitude and frequency of the stimulus pulse sequence as inputs, assuming that they are the key factors affecting the stimulus effect.
In each experiment, continuous bipolar microstration stimulation was performed at a given location, selected from the predation pre cortical cortical (OFC), front buckle-back cortical (ACC), amygdala (AMG), or top leaf upper leaf (SPL), while recording LFP activity cortical and sprite, pale ball, and AMG across multiple brain regions of the prea bash, motor cortical cortical, top-leaf.
the experiment, the monkeys were awake.
accurate IO modeling and machine learning require appropriate IO model structures and informative IO datasets to adapt to the model.
dynamic linear state spatial model (LSSM) structure (LSSM) is established.
model uses an incubation period to describe the effects of stimuli, and the time changes in the incubation period drive the dynamics of the brain network.
information-rich IO dataset needs to provide an input-stimulating waveform that fully stimulates brain network activity.
this can be achieved by designing a waveform of the white spectrum in the input space (amplitude and frequency).
the previous theoretical work, a multi-level noise modulation stimulus pulse sequence is designed and implemented to model IO.
MN modulation pulse sequence randomly switches amplitude and frequency between multiple discrete levels at time.
use IO datasets to fit and evaluate IO models through machine learning techniques and cross-validation.
a multi-experiment experiment designed to repeat the same random MN modulation pulse sequence in each experiment.
input is the same in each test, the dynamics of a single test input drive are the same in all tests, and the internal dynamics independent of the input change in different experiments.
IO model for forward prediction was evaluated using a four-way cross-validation.
to evaluate the statistical significantness of IO model prediction and the control of stimulating artifact suppression effect, the same model is applied to the artificially generated IO dataset.
use dynamic IO models to predict the overall brain network dynamics of a single experiment during stimulation.
conducted 16 MN stimulation experiments on two monkeys.
in each experiment, an MN modulation pulse sequence is generated, with durations from 60 to 270 seconds.
the resulting pulse sequence is used for multiple tests.
also recorded the LFP signals for the first 5 minutes of rest stimulation in the 208 (monkey A) and 165 (monkey M) channels, respectively.
duration of each experiment is 10 min to 120 min.
this paper proves that the response of large-scale multi-region brain networks to electrical stimulation of duration changes can be predicted by developing data-driven dynamic IO models.
reveals and signals the oscillation and damping dynamics of the brain network's response to stimulation, and achieves model-based simulation of internal state closed-loop control.
dynamic structure of the IO model, which quantifies complex brain network dynamics in response to stimuli with oscillation and damping characteristics, is the key to accurate prediction.
using random MN modulation pulse sequences can fully stimulate brain network activity.
closed-loop neuromodulation system can improve the effectiveness of DBS treatment, especially for open-loop stimuli with different effects of neuropsyurological disorders, such as depression.
, closed-loop neuromodulation can also promote brain function, such as memory.
dynamic modeling is particularly important in such closed-loop scenarios, where stimulus parameters such as frequency and amplitude need to change in real time, unlike open-loop scenes characterized by fixed stimulus patterns.
-stimulation closed-loop changes need to be guided by (1) real-time feedback on brain activity and the brain's internal state ( such as mood or pain levels), and (2) IO models to predict how the changes in stimulation will change brain activity, thereby changing the state of the brain's interior.
, the IO model needs to address two major challenges in these applications.
, it must predict the neural response during ongoing and time-varying stimuli, rather than the end of stimuli or fixed stimulation patterns.
, it must make this prediction in a large multi-region brain network because they are involved in many neuropsycological disorders, rather than in specific brain regions.
this paper proves that such a prediction provides a beneficial technique to facilitate the design of neuropsychiatural and neurological disorders of the future closed-loop neuromodulation system.
Yang, Y., Qiao, S., Sani, O.G. et al. Modelling and prediction of the dynamic responses of large-scale brain networks during direct electrical stimulation. Nat Biomed Eng (2021). MedSci Original Source: MedSci Original Copyright Notice: All text, images and audio and video materials on this website that indicate "Source: Mets Medicine" or "Source: MedSci Original" are owned by Mets Medicine and are not authorized to be reproduced by any media, website or individual, and are authorized to be reproduced with the words "Source: Mets Medicine".
all reprinted articles on this website are for the purpose of transmitting more information and clearly indicate the source and author, and media or individuals who do not wish to be reproduced may contact us and we will delete them immediately.
at the same time reproduced content does not represent the position of this site.
leave a message here