Gradient boosting based prediction of gender based on gonial angle measurements

Abstract

Objective

Forensic odontology is a field of dentistry that applies dental science to the legal system, examining and analyzing dental evidence in criminal investigations, civil litigation, and other legal proceedings. It plays a crucial role in determining the gender of individuals whose bones have been significantly deformed in a catastrophic event. Forensic dentists use morphological and genetic studies to determine an individual's gender, relying on sexual dimorphism and remnants. The mandible and gonial angle, formed by the lower mandible and ramus, aid in gender determination and age estimation. The study aimed to assess the comparison of neural networks and gradient boosting in the prediction of gender based on gonial angle measurements.

Methods

Two hundred CBCT and 200 OPG images were retrieved from Oral and Maxillofacial Radiology archives, involving 100 males and 100 females. Before evaluation, CBCT scans underwent manual reorientation for standardization. The coronal view was adjusted by aligning the software's vertical reference line with the median sagittal plane. The axial reconstruction line was aligned with the mandibular body. The sagittal reconstruction image thickness was increased to 35 millimeters, with two lines for demarcation of the Gonion point. After obtaining the dataset, outliers were removed and normalized, and data were split into 80 % percent and 20 % percent test data and subjected to gradient boosting and neural networks.

Result

The study compares Neural Networks' and gradient-boosting models' performance on a task, finding that the Neural Network outperformed the latter with an Area Under the Curve (AUC) of 0.922 and a higher F1 score (Harmonic mean of Precision and Recall).

Conclusion

The study demonstrates that the gonial angle, a mandibular measure, can accurately determine gender, with conventional statistical methods and machine learning models predicting it, but with limitations