A deep-learning-based workflow to assess taxonomic affinity of hominid teeth with a test on discriminating Pongo and Homo upper molars.

摘要: Objectives Convolutional neural network (CNN) is a state-of-art deep learning (DL) method with superior performance in image classification. Here, CNN-based workflow proposed to discriminate hominid teeth. Our hope that this could help confirm otherwise questionable records of Homo from Pleistocene deposits where there standing risk mis-attributing molars Pongo Homo. Methods and materials A two-step was designed. The first step converting the enamel-dentine junction (EDJ) into EDJ card, is, two-dimensional conversion three-dimensional surface. In step, researchers must carefully orient teeth according cervical plane. second training CNN learner labeled cards. sample consisting 53 fossil (modern human Neanderthal) adopted generate cards, which were then separated set (n = 84) validation 22). To assess feasibility workflow, Pongo-Homo classifier trained aforementioned card set, used predict taxonomic affinities six samples (test set) von Koenigswald's Chinese Apothecary collection. Results show cards are classified accurately by learner. More importantly, predictions for specimens test match well diagnosis results deduced multiple lines evidence, implying great potential method. Discussion This paves way future studies using address complexity (e.g., distinguishing Asia). Further improvements include visual interpretation extending applicability moderately worn