Automatic Sex Determination of Skulls Based on a Statistical Shape Model

Sex determination from skeletons is an important research subject in forensic medicine. Previous skeletal sex assessments are through subjective visual analysis by anthropologists or metric analysis of sexually dimorphic features. In this work, we present an automatic sex determination method for 3D digital skulls, in which a statistical shape model for skulls is constructed, which projects the high-dimensional skull data into a low-dimensional shape space, and Fisher discriminant analysis is used to classify skulls in the shape space. This method combines the advantages of metrical and morphological methods. It is easy to use without professional qualification and tedious manual measurement. With a group of Chinese skulls including 127 males and 81 females, we choose 92 males and 58 females to establish the discriminant model and validate the model with the other skulls. The correct rate is 95.7% and 91.4% for females and males, respectively. Leave-one-out test also shows that the method has a high accuracy. 1. Introduction Sex identification from skeletons is a vital work for a forensic anthropological analysis. Previous studies [1–4] indicate that pelvis is the most reliable indicator of sex assessment, and skull is the second one. However, not all forensic cases provide a complete skeleton due to breakage or postmortem destruction, while the skull can be well preserved in most cases since it is composed of hard tissue. So the skull is the most commonly used skeleton part in forensic anthropological analysis. Traditional skeletal sex assessments principally rely on visual assessments of sexually dimorphic traits. Rogers [5] achieves an accuracy of 89.1% for a historic skeletal collection by using visual morphological traits. By scoring the visual assessments of five cranial traits (glabella, mental, orbit, nuchal, and mastoid), Walker [6] achieves 90% accuracy for modern American skulls via a quadratic discriminant analysis model incorporating scores. According to Daubert [7] and Mohan [8] criteria, Williams and Rogers [9] assess 21 skull characteristics of 50 white European Americans (25 males and 25 females). For a characteristic, if the intraobserver error is no more than 10% and the accuracy is above 80% when it is used separately to identify the sex, the characteristic is defined as a high quality characteristic. They get six high quality characteristics like eyebrow, orbit, and so forth, and the accuracy reaches 94% by utilizing visual assessments of all these six characteristics. Visual assessments depend heavily on physical anthropologists’