Uncovering Heart Rate Response Patterns to Threat Pictures Through Deep Latent Representation Learning with a...

Area of Science:

• Psychophysiology
• Machine Learning
• Neuroscience

Background:

• Group-level averaging of psychophysiological data can obscure individual differences in responses.
• Understanding individual heart rate (HR) variability during threat is crucial for studying coping mechanisms.
• Machine learning, specifically variational autoencoders (VAEs), can uncover hidden patterns in complex data.

Purpose of the Study:

• To introduce a VAE-based method for characterizing individual HR responses to threat.
• To compare VAE-derived clustering with traditional waveform clustering.
• To demonstrate the VAE's utility in identifying distinct autonomic response profiles.

Main Methods:

• A simple VAE was developed to analyze HR responses from 165 participants exposed to threat stimuli.
• Simulations were used to validate the VAE's ability to separate HR waveforms.
• Latent space clustering was compared against clustering of raw HR waveforms.

Main Results:

• The VAE identified three distinct HR response profiles: strong decelerators, weak decelerators with late acceleration, and immediate accelerators.
• Clustering raw HR waveforms yielded only two groups.
• VAE-derived clusters were more coherent and internally consistent than those from raw data.
• The VAE successfully characterized HR patterns in a small fear-conditioning dataset.

Conclusions:

• A basic VAE model effectively categorizes psychophysiological response patterns, improving upon standard methods.
• This approach provides a framework for linking individual autonomic variability to affective and defensive behaviors.
• VAE-based analysis offers a powerful tool for exploring individual differences in psychophysiological responses.