Machine learning methods for sex estimation of sub-adults using cranial computed tomography images

Area of Science:

• Forensic Anthropology
• Medical Imaging
• Machine Learning

Background:

• Accurate sex estimation is crucial in forensic anthropology and developmental studies.
• Cranial computed tomography (CCT) provides detailed anatomical information for analysis.
• Machine learning (ML) offers potential for automating complex classification tasks.

Purpose of the Study:

• To compare the classification accuracy of three ML methods (Random Forest, Support Vector Machines, Linear Discriminant Analysis) for sex estimation in sub-adults.
• To evaluate the performance of ML models using craniometric parameters derived from CCT scans.
• To introduce the first AI-based classification model for sex estimation in sub-adults using CCT data.

Main Methods:

• Analysis of 521 CCT scans from Malaysian sub-adults (0-20 years) using Mimics software.
• Measurement of 14 craniometric parameters via a plane-to-plane protocol.
• Development and optimization of RF, SVM, and LDA classification models using GridSearchCV.

Main Results:

• Random Forest (RF) achieved the highest testing accuracy of 73% with optimal hyperparameters.
• Support Vector Machines (SVM) obtained 67% accuracy, and Linear Discriminant Analysis (LDA) achieved 65% accuracy.
• RF demonstrated superior performance compared to SVM and LDA for sex estimation in this cohort.

Conclusions:

• Random Forest (RF) is the most effective ML method among those tested for sex estimation in sub-adults using CCT scans.
• This study presents a novel AI-based approach for sex estimation in sub-adults, with implications for forensic and clinical applications.
• Further research can explore larger datasets and diverse populations to refine these ML models.