Machine learning-based sex estimation using palatal morphometry acquired from digital impressions

Area of Science:

• Forensic anthropology
• Bioinformatics
• Medical imaging

Background:

• Accurate sex estimation is crucial for identifying human remains.
• The human palate offers structural integrity and sexual dimorphism, making it valuable for forensic analysis.
• Innovative technologies like intraoral scanners (IOS) and machine learning (ML) enhance forensic identification accuracy and efficiency.

Purpose of the Study:

• To estimate sex from palatal morphometric features in the Iraqi subpopulation.
• To integrate intraoral scanning and machine learning for forensic sex determination.
• To assess the efficacy of ML models in predicting sex from palatal measurements.

Main Methods:

• Extraction of palatal morphometric features from digital impressions of 200 individuals (100 male, 100 female) using specialized software.
• Analysis of features including palatal width in the canine area (WPC) and palatal area (APA).
• Application of six different machine learning (ML) models for sex prediction, with evaluation using nested cross-validation (NCV).

Main Results:

• Palatal width (WPC) and palatal area (APA) showed significant discriminative power (AUC 0.983 and 0.868, respectively).
• Support Vector Classifier (SVC) and Logistic Regression (LR) models achieved the highest sex prediction accuracy at 95% via NCV.
• The study identified key morphometric features for effective sex estimation.

Conclusions:

• The developed methodology for palatal sex estimation is highly promising for forensic applications.
• Further research with larger, diverse populations is recommended to validate and strengthen the findings.
• This approach offers a reliable tool for sex determination from human remains.