Different machine learning methods based on maxillary sinus in sex estimation for northwestern Chinese Han population

Area of Science:

• Forensic Anthropology
• Medical Imaging Analysis
• Machine Learning in Forensics

Background:

• Sex estimation is crucial in forensic science.
• The maxillary sinus is a resilient anatomical structure suitable for forensic analysis.
• Machine learning enhances accuracy in predicting sex from anatomical data.

Purpose of the Study:

• To evaluate the efficacy of maxillary sinus linear measurements for sex estimation.
• To develop and compare machine learning models (logistic, KNN, SVM, RF) for sex prediction.
• To assess the impact of age on sex estimation accuracy.

Main Methods:

• Collected Cone-Beam Computed Tomography (CBCT) data from 477 individuals (Han population, northwest China).
• Measured linear dimensions of maxillary sinuses (length, width, height) and inter-sinus distance.
• Developed and validated logistic, KNN, SVM, and random forest models using 80% training and 20% testing data splits.

Main Results:

• The random forest model achieved 77.78% overall accuracy for individuals over 18.
• Accuracy was slightly higher for males (78.12%) than females (77.42%).
• Incorporating age as a variable improved accuracy, reaching 88.46% in the 18-27 age group, with all variables showing linear correlation with age.

Conclusions:

• Linear maxillary sinus measurements are valuable for adult sex estimation (≥18 years).
• A robust random forest model was developed for sex estimation in the Han population of northwest China.
• Age is a significant predictive variable for improving sex estimation accuracy.