The subpubic angle and palpable pelvic parameters in sex estimation using machine learning algorithms in a Turkish population.

PURPOSE: This retrospective cross-sectional study aimed to estimate sex in the Turkish adult population using the multidimensional characteristics of pelvic morphometry derived from computed tomography (CT) images analyzed through machine learning algorithms (MLAs) based on palpable parameters.

MATERIAL AND METHODS: Abdominal CT scans of 301 adults (150 females, 151 males) were retrospectively evaluated. Six measurements were obtained from the pelvis, including five palpable anatomical landmarks representing the distances between the right and left anterior superior iliac spines, posterior superior iliac spines, the highest posterior points of the iliac crests, the most lateral points of the iliac crests, the ischial tuberosities and the subpubic angle. Independent samples t-test and multivariate logistic regression were used to assess sex differences. Supervised learning algorithms Logistic Regression (LR), Classification and Regression Tree (CART), Multilayer Perceptron (MLP) and the unsupervised K-means clustering algorithm (KM) were applied for sex estimation (p < 0.05).

RESULTS: The LR model achieved 94.7% overall accuracy. All variables except the distance between the right and left anterior superior iliac spines, significantly contributed to sex estimation. MLP achieved the highest accuracy (100% for females, 96% for males), while CART reached >93% accuracy using only subpubic angle. KM classified sexes with 92% accuracy.

CONCLUSION: CT-based pelvic morphometry analyzed by MLAs provides a highly accurate and objective approach for sex estimation. The developed model can be applied for sex estimation in the Turkish population, particularly in forensic studies, involving living individuals or partially preserved skeletal remains.