Sex estimation using sternum part lenghts by means of artificial neural networks. (August 2019)

• Record Type: Journal Article
• Authors: Oner, Zulal; Turan, Muhammed Kamil; Oner, Serkan; Secgin, Yusuf; Sahin, Bunyamin
• Is Part Of: Forensic science international. Volume 301(2019)
• Page Start: 6
• Page End: 11
• Publication Date: 2019-08
• Subjects: Sternum -- Computerized tomography -- Artificial neural network -- Multilayer perceptron classifier -- Stochastic gradient descent -- Sex identification; Medical jurisprudence -- Periodicals; Chemistry, Forensic -- Periodicals; Forensic Medicine -- Periodicals; Médecine légale -- Périodiques; Chimie légale -- Périodiques; Gerechtelijke geneeskunde; Gerechtelijke chemie; Gerechtelijke psychiatrie; Chemistry, Forensic; Medical jurisprudence; Electronic journals; Periodicals; Electronic journals; 614.1
• DOI: 10.1016/j.forsciint.2019.05.011
• Languages: English
• ISSNs: 0379-0738

Abstract

Highlights: The sternum is used for sex identification. Lengths of parts of sternum and sternal angle were greater in males. Accuracy value of network with two layers is more than a single layer. Maximum success of artificial neural network is 0.97. The highest accuracy rate was found in the set without sternal angle. Abstract: In addition to the pelvis, cranium and phalanges, the sternum is also used for postmortem sex identification. Bone measurements may be obtained on cadaveric bones. Alternatively, computerized tomography may be used to obtain measurements close to the original ones. Moreover, usage of artificial neural networks (ANNs) in the field of medicine has started to provide new horizons. In this study, we aimed to identify sex by an ANN using lengths of manubrium sterni (MSL), corpus sterni (CSL) and processus xiphoideus (XPL) and sternal angle (SA) from computerized tomography (CT) images brought to an orthogonal plane. This study used the thin-slice thoracic CT images of 422 cases (213 female, 209 male) with an age range of 27–60 years brought to the orthogonal plane. Measurements of MSL, CSL, XPL and SA were analyzed with a multilayer artificial neural network that used stochastic gradient descent (SGD) for optimization and two hidden layers. MSL, CSL and XPL were longer, and SA was wider in men (MSL p = 0.000, CSL p = 0.000, XPL p = 0.000, SA p = 0.02). In the case of the two hidden layers of the network with 20 and 14 neurons in the hidden layers, respectively, learning rate of 0.1 and momentum coefficient of 0.9, the accuracy (Acc) of sex prediction was 0.906. In order to define a more realistic performance of the network, bootstrap was run with the confidence interval of 94%. A sensitivity (Sen) value of 0.91 and a specificity (Spe) value of 0.90 were calculated. The success rates that were achieved in sex identification with measurements on the skeleton using ANN were observed to be higher than those achieved by linear models. Also, sometimes all parts of the bones may not be found or might be deformed. In this case, the number of parameters used for the estimation will be incomplete. The ANN has the strong advantage to be able to estimate despite the missing parameter.