EEG-based detection of threat perception from event-Related Potentials (ERPs)

Introduction: In high-stakes environments, where rapid responses to threats are critical, reducing the latency between threat perception and reaction could save lives. Towards this goal, a system that acts on the neural response to threat perception may reduce the latency of threat reactivity. Here we present a detailed electroencephalography (EEG) -based study to identify early neural correlates of threat perception from the event-related potentials (ERPs) and determine the accuracy of detecting threatening versus non-threatening, or distractor (novel) versus non-threatening stimuli on a single-trial basis. We aimed to determine the feasibility of integrating a Brain-Computer Interface (BCI) to enable real-time detection and response to threats. A functional connectivity (FC) analysis was included, which focused on nodes in the Default Mode Network (DMN) and major attention networks, to identify possible connectivity patterns that enable discrimination between threat and non-threat. Through an investigation of both local neural responses and the dynamic interplay between associated brain networks, this study offers a comprehensive approach to the early detection of threat perception.
Methods: Twenty-eight participants were exposed to two Rapid Serial Visual Presentation (RSVP) tasks, which presented visual stimuli with offset latencies between 2.6Hz to 10Hz (i.e., at presentation speeds of 100-175ms, 200-275ms and 300-375ms). The first RSVP protocol presented images without a background and the second RSVP protocol presented the same images with the background included. Stimulus categories were ‘direct’, ‘faces’, and ‘objects and scenes’. Standard machine learning methods were applied to establish distinct classifiers for each stimulus type and presentation rate.
Results: The findings revealed that threat-related ERPs, particularly those elicited by direct threat stimuli, could be classified with high accuracy (p < 0.001). Notable FC difference were observed foremost in the theta-band, with the highest reproducible connectivity (ICC, 0.84) in response to threatening faces. Regarding the direct image category, the most robust significant difference in FC between exposure to threat and non-threat stimuli, was observed between the left precuneus and the left posterior cingulate cortex (PCC, ICC, 0.7), key nodes in the DMN, that are understood to be tonically active, and involved in monitoring the external and internal environments.
Conclusion: The results of this study indicate the potential of EEG-based BCIs to enhance human threat detection in real-time – by bridging the gap between the speed of neural perception and the slower process of physical response, particularly in situations where decision-accuracy is critical. Key findings include the detection of the neural correlates of threat perception on a single-trial basis within less than 375ms, with threat-associated ERPs being more robustly separable compared to distractor-associated ERPs – additionally, direct or immediate personal threats resulted in improved accuracy, and a functional connectivity analysis revealed distinct connectivity patterns for both threat and distractor stimuli. Taken together, the research findings presented here demonstrate the potential for a threat perception detection based Brain-Computer Interface to offer a temporal advantage in dangerous environments.