Auditory steady-state response and gamma oscillations in an excitatory-inhibitory balanced neuronal network

Basic Information

• Paper ID: 2504.04329
• Title: Auditory steady-state response and gamma oscillations in an excitatory-inhibitory balanced neuronal network
• Authors: Duoyu Feng, Jiajia Li
• Classification: q-bio.NC (Quantitative Biology - Neurons and Cognition)
• Institutions: School of Information and Control Engineering, Xi'an University of Architecture and Technology; Department of Neurosurgery, Central Theater General Hospital
• Paper Link: https://arxiv.org/abs/2504.04329

Abstract

This study proposes a novel auditory neural network model that integrates speech signal input, cochlear processing, and cortical excitatory-inhibitory (E-I) balanced networks. The research reveals that increased noise intensity attenuates auditory steady-state responses in gamma oscillations, a mechanism verified through publicly available EEG data. Furthermore, enhancing the brain's E-I balance significantly improves auditory attention during speech recognition. This work not only elucidates the neural basis of selective attention in noisy environments but also provides promising therapeutic strategies for auditory attention disorders.

Research Background and Motivation

Core Research Questions

This study addresses the classic "cocktail party problem"—how the brain effectively perceives target speech signals in noisy environments. Specific research questions include:

1. How does the cerebral cortex perceive speech information amid environmental noise?
2. What is the relationship between gamma oscillations and auditory attention construction?
3. What are the mechanisms by which excitatory-inhibitory balance regulates attention?

Significance of the Problem

• Theoretical Significance: Deep understanding of the brain's speech recognition mechanisms holds important value for computational neuroscience
• Applied Value: Provides theoretical foundations for optimizing speech recognition systems in human-computer interaction (HCI)
• Clinical Significance: Offers potential therapeutic strategies for auditory attention disorders and hearing loss

Limitations of Existing Approaches

• Existing research predominantly employs "black-box" models lacking interpretability
• Insufficient in-depth analysis of neural oscillation dynamics mechanisms
• Inadequate understanding of how excitatory-inhibitory balance regulates attention

Core Contributions

1. Construction of a comprehensive auditory processing model: Integrates a complete auditory pathway from speech input through cochlea, thalamus, to cortex
2. Revelation of gamma oscillation encoding mechanisms: Discovers that gamma oscillation peak patterns encode speech signal features
3. Verification of noise masking effects: Validates through computational models and EEG data the inhibitory effects of noise on gamma oscillations
4. Proposal of attention regulation strategies: Finds that enhancing E-I balance improves auditory attention, providing new clinical insights
5. Establishment of state transition maps: Constructs cortical perceptual state transition diagrams under varying noise intensity and E-I balance parameters

Methodology Details

Task Definition

Investigation of auditory cortex neural network responses to speech signals under different noise conditions, with particular focus on:

• Input: Continuous speech signals, pure tones, noise of varying intensities
• Output: Local field potentials (LFP), neuronal firing patterns, gamma band oscillations (GBO)
• Objective: Understanding noise masking mechanisms and E-I balance regulation of attention

Model Architecture

1. Speech Input-Cochlear Coupling System

X = FFT_transform(Voice_Signal)                    (1)
x = envelope(X)                                    (2)
x' = (x - x_min)/(x_max - x_min)                  (3)
I_thalamus,i^E = A_i^E · x'                       (4)
I_thalamus,j^I = A_j^I · x'                       (5)

Where A_i^E : A_j^I = 5:2, simulating physiological parameter ratios in cortical networks.

2. Auditory Cortex Neural Network Model

Construction of an E-I balanced network containing 200 excitatory pyramidal neurons and 50 inhibitory interneurons:

Excitatory Neurons (Two-Compartment Model):

• Soma equation:

C_m,E dV_E,i/dt = f_E(V_E,i, m_i, n_i, h_i) + g_c/p(V_Ed,i - V_E,i)    (6)

• Dendrite equation:

C_m,E dV_Ed,i/dt = f_Ed(V_Ed,i, Ca^2+, s_n) + g_c/(1-p)(V_E,i - V_Ed,i) + I_syn,i^Ed + I_thalamus,i^Ed    (7)

Inhibitory Neurons (Fast-Spiking Interneuron Model):

C_m,I dV_I,j/dt = f_I(V_I,j, m_j, n_j, h_j) + I_syn,j^I + I_thalamus,j^I    (8)

3. Synaptic Current Model

Synaptic currents received by excitatory neurons:

I_syn,i^Ed = Σ[g_I w_k^I→E y_GABA,k(V_Ed,i - V_GABA)] + Σ[g_E y_AMPA,k(V_Ed,i - V_AMPA)/N_E]    (9)

Synaptic currents received by inhibitory neurons:

I_syn,j^I = Σ[g_E w_k^E→I y_AMPA,k(V_I,j - V_AMPA)] + g_GABA,j^autapse y_GABA(V_I,j - V_GABA) + Σ[g_I y_GABA,k(V_I,j - V_GABA)/N_I]    (13)

Technical Innovations

1. Multi-scale Integration Model: First integration of cochlear frequency separation, thalamic feature analysis, and cortical E-I networks within a unified framework
2. Dynamical Analysis Methods: Employs bifurcation analysis to reveal mechanisms of noise effects on neuronal firing patterns
3. Gamma Oscillation Encoding Theory: Proposes a novel mechanism whereby gamma oscillation peak patterns encode speech features
4. State Transition Control: Discovers methods for achieving controllable transitions in perceptual states through parameter adjustment

Experimental Setup

Datasets

1. Simulation Data:
   • Continuous speech signals (with/without noise conditions)
   • Pure tone signals (200-1000 Hz)
   • White noise (20-80 dB)
2. Validation Data:
   • Public EEG dataset41: 13 subjects
   • Stimuli: 1000 Hz and 500 Hz pure tones, 76 dB white noise
   • Stimulus duration: 60 ms per trial, total experiment duration: 13 minutes

Evaluation Metrics

1. Gamma Band Oscillations (GBO): Power in the 30-100 Hz frequency band
2. Peak Amplitude: Maximum value of the GBO curve
3. Peak Entropy: Information content of GBO peaks based on Shannon entropy
4. E-I Ratio: Ratio of excitatory to inhibitory postsynaptic currents

Analysis Methods

1. IIR Digital Filtering: Extraction of 30-100 Hz gamma frequency band
2. Power Spectral Analysis: Computation of squared power of filtered signals
3. Bifurcation Analysis: Investigation of system stability and firing pattern transitions
4. Time-Frequency Transforms: Analysis of speech signal frequency domain characteristics

Experimental Results

Main Findings

1. Verification of Noise Masking Effects

• No-noise condition: GBO peak amplitude in the 40-60 range, firing frequency >50 Hz
• Noise condition: GBO peak amplitude reduced to 0-20 range, firing frequency <35 Hz
• Critical threshold: 40 dB as the critical point for significant noise effects, consistent with results from Hahad et al.45

2. Frequency-Dependent Response

• As pure tone frequency increases from 200 Hz to 1000 Hz, GBO peak shows increasing trend
• GBO peak under white noise stimulation consistently below 20, significantly lower than pure tone stimulation
• EEG validation data shows similar frequency-dependent patterns

3. E-I Balance Regulation Effects

• As excitatory synaptic conductance g_E increases from 0.1 to 0.6:
  • E-I ratio increases significantly
  • Maximum GBO amplitude improves from approximately 20 to 60
  • Peak encoding entropy improves substantially

Ablation Studies

Bifurcation Dynamics Analysis

• Excitatory neurons: Enter firing state between Hopf bifurcation points HBPE,L and HBPE,R
• Inhibitory neurons: Similar bifurcation characteristics, but smaller IPSC variations
• Key finding: Noise primarily regulates neuronal firing patterns through effects on EPSC dynamics

State Transition Analysis

Construction of two-dimensional parameter space with noise intensity (20-80 dB) and g_E (0.1-1.0):

1. State ① Perception: Low noise, good speech encoding capability
2. State ② Masking: High noise, loss of speech perception ability
3. State ③ Recovery: Recovery of perception through enhanced g_E
4. State ④ Sharp Wave Ripples: Over-excitation state (100-200 Hz)

Experimental Discoveries

1. Encoding Mechanism: Spatiotemporal patterns of gamma oscillation peaks encode speech signal features
2. Masking Mechanism: Noise primarily reduces neuronal excitability by decreasing EPSC
3. Recovery Strategy: Enhancing E-I balance can restore attention in noisy environments
4. Critical Phenomena: Existence of clear noise intensity threshold (~40 dB) and regulation parameter ranges

Related Work

Auditory Attention Mechanism Research

• Kerlin et al.4: Attention gain control in cocktail party environments
• Petkov et al.20: Attention regulation in human auditory cortex
• Jensen et al.47: Relationship between gamma oscillations and attention memory

Neural Network Modeling

• Wang & Buzsáki33: Gamma oscillations in hippocampal interneuron networks
• Economo & White48: Control of gamma oscillations by excitatory-inhibitory balance
• Advantages over existing work: Integration of complete auditory pathway with interpretable dynamical mechanisms

E-I Balance Theory

• Existing research primarily focuses on single-scale E-I balance
• This work first connects E-I balance with auditory attention and speech recognition
• Provides quantitative regulation strategies and parameter ranges

Conclusions and Discussion

Main Conclusions

1. Gamma oscillations serve as neural markers of attention: Gamma oscillation amplitude directly reflects attention level
2. Noise affects attention through EPSC pathway: Noise primarily weakens attention by reducing excitatory synaptic currents
3. E-I balance can regulate attention states: Enhancing excitatory-inhibitory balance improves speech perception in noisy environments
4. Controllable state transition mechanisms exist: Reversible transitions in perceptual states can be achieved through parameter adjustment

Limitations

1. Model Simplification: Cochlear-thalamic system employs simplified signal processing model
2. Fixed Parameters: Certain physiological parameters based on literature values may exhibit individual variations
3. Limited Verification Scope: Primarily addresses pure tones and simple speech; verification in complex speech environments is limited
4. Clinical Translation: Further validation needed for translation from computational models to practical therapeutic applications

Future Directions

1. Multimodal Integration: Incorporation of information processing from other sensory channels such as vision
2. Personalized Modeling: Parameter optimization considering individual differences
3. Clinical Applications: Development of treatment schemes based on E-I balance regulation
4. Neural Modulation: Experimental verification combining optogenetic and other techniques

In-Depth Evaluation

Strengths

1. Theoretical Innovation:
   • First proposal of gamma oscillation peak encoding mechanism for speech
   • Establishment of quantitative relationships between E-I balance and auditory attention
   • Provision of interpretable neural dynamical models
2. Methodological Completeness:
   • Integration of complete auditory pathway from cochlea to cortex
   • Combination of computational modeling with experimental data validation
   • Employment of multiple analysis methods (bifurcation analysis, time-frequency analysis, etc.)
3. Practical Value:
   • Provision of potential therapeutic strategies for auditory attention disorders
   • Biologically-inspired insights for artificial intelligence speech recognition
   • Establishment of actionable parameter adjustment frameworks

Limitations

1. Model Complexity:
   • Contains numerous parameters with high tuning complexity
   • Certain biological details may be oversimplified
   • High computational costs
2. Verification Limitations:
   • Relatively small sample size for EEG validation (13 subjects)
   • Lack of verification in more complex speech environments
   • Clinical efficacy requires further validation
3. Generalizability Issues:
   • Primarily addresses normal hearing populations
   • Model applicability in pathological states unknown
   • Cross-cultural and cross-linguistic applicability requires verification

Impact and Significance

1. Academic Contributions:
   • Provides new modeling frameworks for computational neuroscience
   • Advances understanding of auditory attention mechanisms
   • Bridges theory and experiment
2. Application Prospects:
   • Algorithm optimization for hearing aids and cochlear implants
   • Enhancement of noise robustness in speech recognition systems
   • Novel treatment methods for attention deficit disorders
3. Reproducibility:
   • Provides detailed mathematical models and parameters
   • Uses publicly available EEG datasets for validation
   • Relatively complete method descriptions

Applicable Scenarios

1. Basic Research: Auditory neuroscience and cognitive neuroscience research
2. Clinical Applications: Diagnosis and treatment of auditory attention disorders and hearing loss
3. Engineering Applications: Algorithm optimization for intelligent speech systems and hearing devices
4. Educational Applications: Teaching cases for neuroengineering and computational neuroscience

References

This paper cites 65 relevant references, primarily including:

Core Theoretical References:

• Wang, X. J., & Buzsáki, G. (1996). Gamma oscillation by synaptic inhibition in a hippocampal interneuronal network model
• Jensen, O., Kaiser, J., & Lachaux, J. P. (2007). Human gamma-frequency oscillations associated with attention and memory

Validation Data:

• Delorme, A. (2022). EEG data from an auditory oddball task. OpenNeuro

Methodological References:

• Economo, M. N., & White, J. A. (2012). Membrane properties and the balance between excitation and inhibition control gamma-frequency oscillations

This paper makes significant contributions to computational neuroscience and auditory processing, providing not only new theoretical frameworks but also opening new directions for clinical applications. Its integrative modeling approach and systematic validation establish a solid foundation for subsequent research in this field.