Comparison among artificial intelligence-based age estimation from morphological analysis of the pubic symphysis versus experienced and novice practitioners using a new atlas for component labeling.

Traditional age estimation methods based on macroscopic observation has been criticized for being excessively dependent on the observer's experience. The aim of this technical note is to propose a new atlas to assist the forensic practitioner in labelling pubic symphysis components. Furthermore, intra- and inter-observer evaluation was conducted using both novice and experienced practitioners. Two experienced and two novice practitioners have used this atlas to label 1,127 identified pubes from autopsies. Furthermore, they have considered the phases of Todd's method (1920) to estimate the age of each pubis. A previously published, semi-automatic artificial intelligence rule-based method based on the C4.5 algorithm has also been used to recommend a specific age-at-death estimation from the human-defined labels, to be compared with the macroscopic age estimation performed by all observers. Linear weighted kappa coefficients indicate that the intra- and inter-observer error when using the new atlas is higher for novice practitioners (Kappa < 0,6) than for experienced practitioners (Kappa > 0,6). Component labeling produces less error than phase assignment following the traditional method only in the case of experienced practitioners. In addition, the artificial intelligence method achieves a global percentage of correct estimates similar to what the four practitioners can achieve. The proposed atlas can be thus considered an effective tool for component labeling. Besides, explainable machine learning techniques could help automate age estimation methods through component analysis. These techniques reduce subjectivity, but it is important that researchers engage in the process to ensure the replicability of the method. Nevertheless, these results must be regarded as preliminary until they are subjected to a more extensive evaluation by a larger cohort of observers.