Sex estimation from sacrum and coccyx with discriminant analyses and neural networks in an equally distributed population by age and sex

Sex estimation is an essential step in the process of the identification of the skeletal remains in forensic anthropology since it reduces the number of possible matches by half. In this study, sex estimation with 21 sacral and coccygeal metric parameters obtained from Computerized Tomography images of a Turkish population which consists of 480 patients that are equalized according to their sexes and ages, is performed. Univariate discriminant analysis, linear discriminant function analysis, stepwise discriminant function analysis, and multilayer perceptron neural networks are used in this study. A maximum of 67.1% accuracy for univariate discriminant analysis, 82.5% for linear discriminant function analysis, 78.8% for stepwise discriminant function analysis, and 86.3% for multilayer perceptron neural networks, were achieved. Although it does not reach an acceptable accuracy rate of 95% or more for sacrum and coccyx, sex estimation with neural networks is a promising field of research in corpses where identification is otherwise not possible, and further studies with other bones and with new techniques might give useful information. (C) 2019 Elsevier B.V. All rights reserved.