MOSAIC: Unifying Methods of Sex, Stature, Affinity, & Age for Identification through Computational Standardization

To aid in the identification of unknown human remains, forensic anthropologists are often called upon to create a biological profile, or an estimation of the unknown individual's age, biological sex, population affinity (i.e., ancestry), and stature.  The data collected to develop a biological profile typically includes both metric and nonmetric/morphological data from across the entire skeleton, and these data are then analyzed using a variety of analytical methods.  These methods are each based on different subsets of data (e.g., cranial metrics for population affinity estimitation versus innominate morphology for sex estimation), with multiple methods available to estimate each parameter of the biological profile.  However most currently available methods analyze only a single date modality (metric v. nonmetric), focus on only a single skeletal element (e.g., cranium v. innominate), and can only estimate a single biological profile parameter.  This leaves forensic anthropologists with a bewildering array of methods that may produce idiosyncratic results, with no evidence-based way to determine how differing results should be weighed or reported.  Further, fragmenting methods by individual biological profile parameters ignores the realities of human skeletal variation, creates a priori assumptions about which data provide information for which parameter, and fails to account for interactions between sex, population affinity, age, and stature.

MOSAIC, or Methods of Sex, Stature, Affinity, & Age for Identification through Computational Standardization, is a proposed non-proprietary computer program that will produce a holistic estimation for all biological profile parameters.  This program will use a machine learning (ML), artificial intelligence (AI) system built using a large database of contemporary skeletal data.  The innovatons of MOSAIC are (1) the creation of a dataset that encompasses trait and measurement data from the entire skeleton that spans multiple biological profile methods ("matched data") and (2) the use of a robust ML/AI approach to uncover new patterns of interacton within the data, while minimizing variability between practitioners.  Preliminary data show that the alpha version of MOSAIC significantly outperforms the most commonly used methods to individually estimate sex and population affinity, as well as a combined sex and population affinity estimation.  The use of a reference sample of matched data and non-partitioned approach addresses pitfalls in traditional approaches, while also standardizing the analytical process of biological profile estimation. CA/NCF