Prediction of sex, based on skull CT scan measurements in Iranian ethnicity by machine learning-based model

Introduction: Identification of individuals is a crucial aspect of forensic medicine. Due to the durability of bones, they are regarded as an ideal investigative tool, particularly in complex cases where other body parts are highly degraded. Aim: This study aims to predict sex based on skull CT scan measurements in Iranian ethnicity by a machine learning-based model. We try to depict skull sexual differences and propose new analytic methods based on machine learning, to improve the efficacy of personal identification. Method: Eight variables were measured from skull CT images of 199 Iranians, including 118 males with a mean age of 56.4 years and 81 females with a mean age of 55.2 years. Craniometric data were analyzed by conventional logistic regression and the Gradient Boosting Decision Trees method. Results: According to statistical analysis utilizing a univariate logistic regression model, the LCB, LFCB, and BD indices had a statistically significant impact on the final sex prediction of the subject. With an AUC of 0.83, this model's overall accuracy for sex prediction was 83%. The gradient boosting model outperformed logistic regression, with AUC and accuracy values of 0.94 and 0.89, respectively, which were higher than those of logistic regression. In the gradient boosting model, LFCB, BD, and LCB were also the most important craniometrics. Conclusion: This study demonstrates sexual differences in the Iranian population and the high accuracy of the Gradient Boosting model in sex identification based on these differences.