Background – Deficient threat recognition is a critical cognitive impairment associated with various psychological and neurological disorders. Eye-tracking technology provides an objective method to assess attentional bias, yet its predictive power for threat recognition deficits remains underexplored. This study investigates whether eye-tracking metrics can serve as reliable predictors of attentional bias in individuals with deficient threat recognition.
Methods – A total of N participants (healthy controls and individuals with documented threat recognition impairments) completed a visual attention task while their gaze behavior was recorded using high-resolution eye-tracking technology. Metrics analyzed included fixation duration, saccade latency, pupil dilation, and dwell time on threat and non-threat stimuli. Statistical analyses involved generalized linear mixed models (GLMMs) to assess group differences and machine learning models (random forest and support vector machines) to determine predictive validity.
Results – Individuals with deficient threat recognition exhibited significantly shorter fixation durations and increased saccadic latency when viewing threat-related stimuli compared to controls (p < 0.001). Machine learning classifiers achieved 82.5% accuracy in distinguishing participants based on eye-tracking metrics alone, with fixation duration and pupil dilation emerging as the strongest predictors. Moreover, pupillary response variability correlated with self-reported anxiety levels (r = 0.68, p < 0.01), suggesting a potential neurocognitive link.
Conclusion – Eye-tracking metrics demonstrate robust predictive validity for attentional bias in individuals with threat recognition deficits. The findings highlight fixation duration and pupillary responses as key markers for cognitive impairments related to threat processing. These results support the development of eye-tracking-based diagnostic tools and targeted cognitive training interventions for at-risk populations. Future research should explore longitudinal studies and real-world applicability to refine predictive models further.