Deep learning for sex determination of the clavicle: A blind study on a Thai population

Abstract

Sex determination from bone is a primary step in biological identification. Clavicles are helpful in autopsies and identification that can lead to sex determination. We employed a previous deep learning method for the sex determination of the clavicle. We trained the model using a deep network designer of the GoogLeNet (a subset of the convolutional neural network) and received the best training model for the study results. This study's goal was to bring the optimal training model of each side view of the clavicle for a blind test and obtain an accurate blind test set on a Thai population. The total sample consisted of 50 pairs of clavicles as a test group (25 females, 25 males). For the deep learning approach, the clavicle was photographed, and each clavicle image was submitted to the training model for sex determination. Test groups of 50 samples were made. Images of the same size were input to test for blind study. The percentage of blind test accuracy was included in the statistical analysis using descriptive statistics. After training the model from GoogLeNet, we discovered the training model to test a blind dataset accuracy by picking the best of the training model from all experiments and bringing the model to test a blind dataset and get the result of blind test set accuracy. The results of this study found accuracies for a blind test set with the highest overall left inferior view of the clavicle with an accuracy of 92%. Accuracy from the test set of each view of the clavicle can demonstrate the forensic value of sex determination. Deep learning using a clavicle can determine the sex and is user friendly for forensic anthropology specialists.