Sex Estimation Based on Tooth Measurements on Panoramic Radiographs with Classical and Machine-Learning Classifiers.

This study assessed sex estimation of Iranians according to maxillary left first molar measurements made on panoramic radiographs using classical and machine-learning classifiers. In this cross-sectional study, tooth length- and width-related variables were calculated for maxillary left first molars on 131 panoramic radiographs (65 males, 66 females; age range of 18-30 years). A subsample of the radiographs was selected and reevaluated by two examiners after 1 month. The intra-class correlation coefficient (ICC) was calculated to assess reliability. The regularized discriminant analysis (RDA), support vector machine (SVM), and cascade-forward and feed-forward neural network models were used for sex estimation. Comparisons were made with the Mann-Whitney and t tests. The intra-observer reliability was 0.9. SVM had the best performance on the test data in both classification schemes. The crown length at the cementoenamel junction (CEJL) and total crown length (CL) in the classification scheme I (sex estimation based on length and width variables), and CEJL/root length (RL), cementoenamel junction width (CEJW)/CEJL, and RL/total tooth length (TTL) in the classification scheme II (sex estimation based on the ratio of variables) were important variables for sex estimation determined by the SVM model. The CEJL had the highest discriminative potential with an area under the curve (AUC) of 78.8. The ratio of variables did not substantially improve sex estimation compared with single variables. CEJL is a reliable measure for sex estimation in Iranians with values higher than 6.25 indicating the male sex and other values indicating the female sex.