Heartbeat-related spectral perturbation of electroencephalogram reflects dynamic interoceptive attention states in...

Area of Science:

• Neuroscience
• Physiology
• Machine Learning

Background:

• Interoceptive awareness, the sense of the body's internal state, is linked to distinct electrophysiological responses synchronized with heartbeats, such as the heartbeat-evoked potential (HEP).
• Existing methods for measuring interoceptive states often rely on specific evoked potentials, necessitating further exploration of more comprehensive electrophysiological markers.
• The need for robust and dynamic measures of interoceptive attention states is critical for understanding their neural underpinnings and clinical relevance.

Purpose of the Study:

• To introduce and characterize heartbeat-related spectral perturbation (HRSP), a novel time-frequency analysis of R-peak locked electroencephalogram (EEG).
• To evaluate the efficacy of HRSP in classifying different interoceptive attention states using machine learning.
• To assess the dynamic properties of HRSP, including optimal frequency bands and minimal averaging epochs for accurate state classification.

Main Methods:

• Heartbeat-related spectral perturbation (HRSP) was computed as a time-frequency map of R-peak locked EEG data, with independent component analysis (ICA) applied to isolate brain components.
• A convolutional neural network (CNN) was developed and applied for both offline and online classification of interoceptive states (attending to heartbeats, white noise, time passage, toe).
• Classification accuracy was compared between HRSP derived from actual R-peaks versus sham (randomly positioned) R-peaks, and the utility of group ICA for online classification was assessed.

Main Results:

• CNNs utilizing HRSP with actual R-peaks achieved significantly higher classification accuracy for interoceptive states compared to those using sham R-peaks.
• Gradient-weighted class activation mapping identified theta and alpha frequency bands (200-600 ms post-R-peak) as crucial features for HRSP-based classification.
• Online classification using group ICA and a pre-trained model demonstrated reliable accuracy without requiring individual EEG precollection, highlighting efficiency.

Conclusions:

• Heartbeat-related spectral perturbation (HRSP) is a viable and sensitive measure for reflecting dynamic interoceptive attention states.
• The developed CNN model effectively leverages HRSP features, particularly in the theta and alpha bands, for accurate classification of interoceptive focus.
• The findings suggest HRSP has transformative potential for clinical applications, enabling objective assessment of interoceptive processing and related disorders.