Dental dimorphism can be used for discriminating sex in forensic contexts. Geometric morphometric analysis (GMA) allows the evaluation of the shape and size, separately, of uneven 3D objects. This study presents experiments using a novel combination of GMA and an artificial neural network (ANN) for sex classification, applied to premolars of Caucasian Italian adults (50 females and 50 males). General Procrustes superimposition (GPS) and the partial least square (PLS) method were performed, respectively, to study the shape variance between sexes and to eliminate landmark variations. The "set-aside" approach was used to assess the accuracy of the proposed neural networks. As the main findings of the pilot study, the proposed method applied to the first upper premolar correctly classified 90% of females and 73% of males of the test sample. The accuracy was 0.84 and 0.80 for the training and test samples, respectively. The sexual dimorphism resulting from GMA was low, although statistically significant. GMA combined with the ANN demonstrated better sex classification ability than previous odontometric or dental morphometric methods. Future research could overcome some limitations by considering a larger sample of subjects and other kinds of teeth and experimenting with the use of computer vision for automatic landmark positioning.
Keywords: 3D dental images; artificial neural network; dental sexual dimorphism; forensic odontology; geometric morphometric analysis; multilayer perceptron; sex estimation.