Retinal Vascular Biometrics for Forensic Personal Identification: A Pilot Validation Study in Palembang, Indonesia

Abstract

Background. Identification of the dead after mass-fatality events can fail when friction ridges are destroyed, dental records are missing or DNA is degraded. The retinal vascular pattern is protected and individually distinctive, but local feasibility data are scarce.

Objective. To evaluate the feasibility and preliminary discriminative performance of a non-mydriatic fundus imaging and vessel-matching pipeline in adults in Palembang, Indonesia.

Methods. In this prospective cross-sectional pilot validation study, one right-eye fundus image was acquired from each of 100 adult volunteers (aged 20–65 years; 52 men, 48 women). Vessels were segmented with Gaussian matched filters, skeletonized and summarized as a five-feature vector; a fused score combined correlation and feature distance. Reference comparisons were 100 self-matches and impostor comparisons numbered 9,900. Age groups were compared by analysis of variance.

Results. Extraction plus matching took 1.50 s per comparison (1.76 s with pre-processing). Recomputed from the tabulated scores, 21% of self-matches scored below 0.80 and 16% of impostor comparisons above it; error rates crossed at about 18.5%, the area under the ROC curve was about 0.89 and d′ was 2.55. Self-match scores did not differ significantly among age groups (F = 0.962, p = 0.386; n = 25–40 per group).

Conclusion. The pipeline was feasible and fast, but discrimination was moderate, and self-matched reference trials, one image per person and no postmortem data cannot support claims of identification accuracy or age invariance. Independent repeat captures and cadaveric validation are needed.