International validation study of AI-guided Craniofacial Superimposition in a contemporary population sample

Abstract

Reliance on primary identifiers (DNA, fingerprints) alone, can delay or hamper forensic human identification. At the same time, the use of techniques with unknown accuracy can compromise identification efforts. Craniofacial Superimposition has faced scrutiny due to limitations in reliability studies, such as small samples, non-representative case selections and unrealistic experimental setups. These studies often lacked comprehensive cross-comparisons and disregarded recommended practices.

Building upon insights from a previous validation study, which highlighted limitations caused by material quality and quantity, a blind validation study focusing on a contemporary sample with superior photographic quality has been conducted. Ten practitioners with different levels of expertise addressed the identification scenario under blind conditions employing a novel AI-guided Craniofacial Superimposition technology, which automates tasks within the application of the technique. Two conclusions are reached: (i) the combination of human expertise and AI assistance can lead to reliable identification outcomes. In particular, it is noteworthy that experienced participants achieved a 100 % correct decision rate, suggesting that prior experience and expertise in the technique contributes to improved decision-making ability, even when aided by AI tools; (ii) the utilization of an automatic AI-based ranking tool significantly reduced the workload for participants by providing a prioritized list of potential matches, narrowing down the number of comparisons from 1175 to 141. Additional constraints and sources of error have been identified. The study was conducted with Skeleton‑ID v4.2, a commercial AI tool (Panacea Cooperative Research, Spain); authors’ competing interests are disclosed in full.