Sex Determination of 3D Skull Based on a Novel Unsupervised Learning Method

Area of Science:

• Forensic Anthropology
• Computational Biology
• Medical Imaging

Background:

• Sex determination from skull morphology is crucial for identifying individuals from unidentified human remains in forensic investigations.
• Existing supervised learning methods face limitations with unlabeled data or imbalanced sample sizes in skull datasets.
• There is a need for robust, unsupervised methods for accurate sex determination from skeletal remains.

Purpose of the Study:

• To propose a novel unsupervised classification technique for sex determination from the skull morphology of the Han Chinese ethnic group.
• To develop and evaluate a stable and efficient unsupervised algorithm for classifying skull data.
• To address the limitations of supervised learning in forensic anthropological applications.

Main Methods:

• Utilized 78 landmarks from 3D skull models derived from computed tomography scans.
• Constructed a skull dataset comprising 40 interlandmark measurements.
• Developed and applied a novel unsupervised algorithm, MKDSIF-FCM (Multi-Kernel Dual-Space Iterative Fuzzy C-Means), for classification.

Main Results:

• The proposed MKDSIF-FCM algorithm demonstrated high sex determination accuracy: 98.0% for females and 93.02% for males.
• The unsupervised method outperformed all attempted classification techniques in accuracy and stability.
• The algorithm proved effective even with potential data imbalances or lack of prior labels.

Conclusions:

• The MKDSIF-FCM algorithm offers a highly accurate and stable unsupervised method for sex determination from skull morphology.
• This approach holds significant potential for application in forensic investigations and archaeological studies involving unidentified human remains.
• Unsupervised learning provides a valuable alternative to supervised methods, particularly in scenarios with limited or unlabeled forensic data.