Convolutional Neural Network Model for Sex Determination Using Femur Bones

Abstract

Forensic anthropology is the critical discipline that applies physical anthropology in forensic education. One valuable application is the identification of the biological profile. However, in the aftermath of significant disasters, the identification of human skeletons becomes challenging due to their incompleteness and difficulty determining sex. Researchers have explored alternative indicators to address this issue, including using the femur bone as a reliable sex identifier. The development of artificial intelligence has created a new field called deep learning that has excelled in various applications, including sex determination using the femur bone. In this study, we employ the Convolutional Neural Network (CNN) method to identify the sex of human skeleton shards. A CNN model was trained on 91 CT-scan results of femur bones collected from Universiti Teknologi Malaysia, comprising 50 female and 41 male patients. The data pre-processing involves cropping, and the dataset is divided into training and validation subsets with varying percentages (60:4, 70:30, and 80:20). The constructed CNN architecture exhibits exceptional accuracy, achieving 100% accuracy in both training and validation data. Moreover, the precision, recall, and F1 score attained a perfect score of 1, validating the model's precise predictions. The results of this research demonstrate excellent accuracy, confirming the reliability of the developed model for sex determination. These findings demonstrate that using deep learning for sex determination is a novel and promising approach. The high accuracy of the CNN model provides a valuable tool for sex determination in challenging scenarios. This could have important implications for forensic investigations and help identify victims of disasters and other crimes.