Determination of Gender By Machine Learning Algorithms, Through Using Craniocervical Junction Parameters and Dimensions of the Cervical Spinal Canal

Abstract

Gender determination is the first step for biological identification. With the widespread use of machine learning algorithms (MLA) for diagnosis, the significance of applying them also in gender determination studies has become apparent. This study has therefore aimed at determining gender from the parameters obtained out of magnetic resonance images (MRI) of the cranio-cervical junction and cervical-spinal canal by using MLA. MRI of the craniocervical junction and cervical-spinal canal of 110 men and 110 women were included in this study. The 15 parameters were tested with Decision Tree (DT), Random Forest (RF), Logistic Regression (LR), Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA) algorithms. Accuracy (Acc), Specificity (Spe), Sensitivity (Sen), F1 score (F1), Matthews-correlation coefficient (Mcc) values were used as performance criteria. The Acc, Spe, Sen, F1, and Mcc were found to be 1.00 in the LR, LDA, QDA and RF algorithms. The ratios of the Acc, Spe, Sen, and F1 were 0.98, and of the Mcc was 0.96 in the DT algorithm. It was found that the ratio between the SHAP analyzer of the RF algorithm and the belt of the ratio between the arch of the atlas and the anterior-posterior distance of the dens (R3) parameter had a higher contribution to the estimation of gender compared to other parameters. It was concluded that the LDA, QDA, LR, DT and RF algorithms applied to the parameters acquired from the MRI of the craniocervical junction and cervical-spinal canal, could determine the gender with very high accuracy.