A study on sex estimation by using machine learning algorithms with parameters obtained from computerized tomography...

Area of Science:

• Forensic Anthropology
• Medical Imaging
• Machine Learning

Background:

• Sexual dimorphism in skeletal structures like the cranium and mandible is well-documented.
• Computerized tomography (CT) provides detailed 3D imaging of skeletal morphology.
• Machine learning (ML) offers powerful tools for analyzing complex datasets and pattern recognition.

Purpose of the Study:

• To evaluate the efficacy of ML algorithms in sex prediction using craniofacial CT parameters.
• To identify which ML algorithms and parameters yield the highest accuracy in sex determination.
• To assess the potential of CT-based skeletal analysis in forensic sex estimation.

Main Methods:

• CT images from 300 individuals (150 males, 150 females) were analyzed.
• 25 skeletal parameters from the cranium and mandible were extracted.
• Various ML algorithms were applied to these parameters, with performance evaluated using Accuracy, Specificity, Sensitivity, F1 score, and Matthews Correlation Coefficient.

Main Results:

• The Logistic Regression (LR) algorithm achieved the highest performance metrics: 0.90 Accuracy, 0.90 Specificity, 0.90 Sensitivity, and 0.90 F1 score.
• LR demonstrated a high predictive capability, correctly classifying 27 out of 30 males and 27 out of 30 females in a subset analysis.
• Other ML algorithms showed accuracies ranging from 0.81 to 0.88.

Conclusions:

• ML algorithms, particularly LR, can accurately predict sex from craniofacial skeletal parameters derived from CT images.
• This approach holds significant promise for applications in forensic anthropology and identification.
• CT imaging combined with ML provides a reliable, non-invasive method for sex determination from skeletal remains.