Convolutional Neural Network Approach In Forensic Odontology For Sex Estimation

In the reconstructive phase of medico-legal human identification, sex estimation is vital for building the biological profile of victims, mainly for age estimation. Traditional methods are subjective, making convolutional neural networks (CNN) a promising alternative for more accurate sex estimation. This study assesses the reliability of the VGG16 model as a forensic sex prediction tool and evaluates its performance using orthopantomography (OPGs). This study included 1050 OPGs from patients aged between 16 and 30 years old, including male and female individuals. Using Python, the OPGs were pre-processed, resized and similar copies were created using data augmentation methods. The model was evaluated for precision, sensitivity, F1-score and accuracy, and heatmaps were created. The study was approved by the Ethics Committee with approval number 216/23. A discrepancy between the validation and training loss values was discovered during the training, and the model demonstrated a general balance between sexes in the general test, achieving F1-scores of 0.89. Contrary to expectations, the 16–20 age group had the highest accuracy rate (90%) for the model in the age group test. Analysis of the heatmaps revealed that the model did not concentrate on anatomically significant regions outside of the mandibular symphysis, perhaps as result of inadequate image extraction techniques. Results indicate that CNNs are accurate in classifying human remains based on the generic factor sex for medico-legal identification, achieving an overall accuracy of 89%. This accuracy provides a reliable foundation for later age estimation, enhancing overall forensic analysis.