Validation of webcam-based eye-tracking for clinically relevant paradigms: Saccade, attention bias, and free-viewing tasks

Eye-tracking is a foundational methodology for studying human visual processing and cognitive behavior. In addition to these fundamental applications, it increasingly serves as a tool for quantifying behavioral biomarkers associated with neurodevelopmental and psychiatric conditions. However, the high cost of research-grade systems limits the scalability of such assessments. This study presents a pre-clinical technical validation of an affordable setup as a necessary first step. This study presents a pre-clinical technical validation of an affordable, AI-driven eye-tracking setup in a healthy, neurotypical cohort We benchmark this system against a gold-standard device to evaluate its reliability for structured, clinically relevant paradigms, including saccade/antisaccade, attention-bias, and free-viewing tasks. We conducted a comparative validation study with 38 participants. Gaze data were simultaneously recorded using a consumer webcam (Obsbot Tiny 2) running a custom computer vision algorithm and a reference system (EyeLink Portable Duo). Performance was assessed across three paradigms: saccade/antisaccade, attention bias, and free-viewing. The systems operated at different temporal resolutions (Obsbot: 25 Hz; EyeLink: 1000 Hz). To facilitate synchronized point-by-point comparison, the webcam data was interpolated to match the higher sampling rate of the EyeLink. Key metrics included reaction time ( τ S , τ A ), response accuracy, dwell time ( D f ), and image exploration area ( A ). To evaluate the system’s diagnostic utility, we used Pearson correlation on quantile-ranked data for continuous measures and concordance analysis for discrete measures to assess the preservation of participant ordering. The webcam system demonstrated strong fidelity in preserving participant rankings. We found high correlations for continuous metrics across all tasks (Pearson’s ρ ∈ [ 0 . 740 , 0 . 863 ] ) and near-perfect agreement for discrete classifications, such as antisaccade accuracy and first fixation location (concordance ϕ ∈ [ 0 . 893 , 0 . 945 ] ). Although the webcam data exhibited slightly higher measurement noise, it did not alter the relative ordering of individuals. An affordable, webcam-based eye-tracking system, when paired with a robust AI algorithm, can serve as a reliable and precise alternative to expensive hardware for structured clinical assessments. This validation suggests the potential for future investigation into the use of webcam-based eye-tracking in remote telehealth settings, provided further clinical testing is conducted. • Webcam-based gaze tracking (Obsbot Tiny 2) validated against EyeLink Portable Duo. • Saccade/antisaccade, attention bias, and free-viewing paradigms evaluated. • Reaction times, antisaccade accuracy, dwell time, first fixation, and exploration area assessed. • High agreement: Pearson ρ : 0.740–0.916; concordance ϕ : 0.893–0.945, preserving rank ordering. • Results support the potential of webcam eye tracking in telehealth and resource-limited settings.