Modelling mouse auditory response dynamics along a continuum of consciousness using a deep recurrent neural network

Area of Science:

• Neuroscience
• Anesthesiology
• Consciousness Science

Background:

• Understanding neurophysiological shifts during anesthesia and consciousness transitions is crucial in anesthesiology and consciousness science.
• Auditory-evoked potentials (AEPs) are sensitive to changes in brain states, but their dynamic morphology during consciousness transitions requires further characterization.

Purpose of the Study:

• To characterize the dynamics of auditory-evoked potential morphology in mice across a continuum of consciousness.
• To develop and validate a computational model for predicting AEP changes during transitions between anaesthetized and conscious states.

Main Methods:

• Epidural field potentials were recorded from auditory cortices of urethane-anaesthetized and conscious mice.
• Auditory stimulation was delivered, and responses were ordered by sample entropy to represent a continuum of awareness.
• A recurrent neural network (RNN) was trained to model event-related potential (ERP) waveforms and their changes with varying state parameters.

Main Results:

• The RNN accurately synthesized ERP waveforms, closely matching experimental data (r² > 0.9).
• Model simulations demonstrated predictable changes in ERP morphology corresponding to transitions between anaesthetized and conscious states.
• Specific AEP peak amplitudes exhibited sigmoidal or linear relationships with the state parameter during consciousness transitions.

Conclusions:

• A recurrent neural network effectively models changes in auditory-evoked potential morphology associated with consciousness state transitions.
• This modeling approach provides a quantitative method to understand neurophysiological correlates of awareness.
• The methodology holds potential for clinical application in predicting consciousness transitions using event-related potentials in humans.