Human and machine perception of sex-related information in external ear morphology

BACKGROUND: The external ear is important in forensic investigations because of its complex morphology and distinctive physical characteristics. Although ear morphology exhibits sex-related variation, its value for sex estimation remains uncertain. AIM: To compare the ability of human observers and a machine learning model to extract sex-related information from ear images. SUBJECTS AND METHODS: A balanced independent test set of 60 ear images (30 female, 30 male) was evaluated by 401 participants from the general population and by a machine learning model trained on a separate dataset of 500 images (250 female, 250 male). RESULTS: Human observers achieved 53.1% accuracy, with higher accuracy for female ears (60.2%) than male ears (39.8%). The machine learning model achieved 63.3% accuracy, with sensitivities of 83.3% for female ears and 43.3% for male ears, showing the same tendency towards higher performance for female ears. CONCLUSIONS: Under the conditions examined, isolated ear images provided limited discriminatory information for sex estimation. Although the machine learning model achieved numerically higher accuracy than untrained human observers, neither approach achieved sufficient accuracy to support the use of isolated ear images as a standalone method for forensic sex estimation.