Estimation of Gender by Using Decision Tree, a Machine Learning Algorithm, With Patellar Measurements Obtained From MDCT Images

Abstract

Aim: The present study aimed to analyze whether gender could be determined with the decision tree (DT) method, a machine learning algorithm, based on patellar multidetector computed tomography (MDCT) image measurements. Material and Methods: The study was conducted on 219 male and 131 female MDCT images. The patellar anteroposterior (Ap), craniocaudal (Cc), transverse (Trv) length and volume (Vol), adjusted on the orthogonal plane by the radiologist, were calculated. In patellar length measurements, initially linear discriminant outliers were detected to clear the data for gender prediction. Accuracy (Acc), Sensitivity (Sen), Specificity (Spe), F1-Score (F1) and Matthew’s Correlation Coefficient (Mcc) criteria were taken as the performance criteria for DT. Results: It was determined that male Ap, Trv, Cc, and Vol values were higher when compared to the female values and there was a significant difference between these values based on gender (pAp, Trv, Cc, Vol = 0.000). Using the above-mentioned measurements, it was calculated that the prediction rate for male individuals was 98.2% and for female individuals, it was 98.4%. Conclusion: DT analysis based on patella morphometry provided a simple, adequate and highly accurate approach for gender estimation. Furthermore, it was determined that it would provide an advantage for researchers in gender prediction using only branching and cut-off values on the tree structure without the need to use a computer.