Comparison of Three Gaze Position Calibration Techniques in First Purkinje-Image Based Eye Trackers

Abstract

Significance: This study highlights potential differences that can arise in gaze-position estimates from first Purkinje image-based eye trackers based on how individual Hirschberg ratios (HRs) are calculated.

Purpose: The purpose of this study was to evaluate the accuracy and repeatability of eccentric-viewing, prism-based, and theoretical techniques that are routinely used to calibrate HR in first Purkinje image-based eye trackers.

Methods: Hirschberg ratios of 28 participants (18 to 40 years old) were obtained using the PlusOptix PowerRef 3 photorefractor and eye tracker. In the gold standard eccentric-viewing technique, participants viewed eccentric targets (±12°, 4° steps) at 2 m. In the prism-based technique, 4 to 16Δ-D base-out and base-in prisms were placed in 4Δ-D steps before an eye occluded with an infrared filter; the fellow eye fixated a target at 1 m. Each participant's HR was calculated as the slope of the linear regression of the shift in Purkinje image relative to the pupil center for each target eccentricity or induced prism power. Theoretical HR was calculated from the participant's corneal curvature and anterior chamber depth measures. Data collection was repeated on another visit using all three techniques to assess repeatability. Data were also obtained from an Indian cohort (n = 30, 18 to 40 years old) using similar protocols.

Results: Hirschberg ratio ranged from 10.61 to 14.63°/mm (median, 11.90°/mm) in the eccentric-viewing technique. The prism-based and theoretical techniques demonstrated inaccuracies of 12 and 4% relative to the eccentric-viewing technique. The 95% limits of agreement of intrasubject variability were ±2.00, ±0.40, and ±0.30°/mm for the prism-based, eccentric-viewing, and theoretical techniques, respectively (P > .05). Intraclass correlation coefficients (95% confidence interval) were 0.99 (0.98 to 1.00) for eccentric, 0.99 (0.99 to 1.00) for theoretical, and 0.88 (0.74 to 0.94) for prism-based techniques. Similar results were found for the Indian cohort.

Conclusions: The prism-based and theoretical techniques both demonstrated relative inaccuracies in measures of HR compared with the eccentric-viewing technique. The prism-based technique exhibited the poorest repeatability.