An automatic method for skeletal patterns classification using craniomaxillary variables on a Colombian population

Area of Science:

• Forensic Anthropology
• Biometrics
• Computer Science

Background:

• Mandibular bone loss complicates forensic facial reconstruction.
• Current methods often fail to account for non-Class I skeletal patterns, particularly in Colombian populations (24.5%).
• Craniomaxillary measurements lack parametric trends, necessitating non-parametric approaches.

Purpose of the Study:

• To classify skeletal patterns using craniomaxillary variables via an automatic non-parametric method.
• To simulate natural mandibular position in a contemporary Colombian sample.
• To address limitations in current forensic facial reconstruction techniques.

Main Methods:

• Collected lateral cephalograms (n=229) from Colombian young adults.
• Utilized landmark coordinates to derive craniomaxillary variables.
• Trained and evaluated a Support Vector Machine (SVM) classifier with a linear kernel, selecting the top 10 variables for accuracy.

Main Results:

• Achieved 74.51% classification accuracy using specific angular variables (e.g., Pr-A-N, PNS-A-Pr).
• Demonstrated effective distinction between skeletal Class II and Class III patterns through Class Precision and Class Recall metrics.
• Identified key craniomaxillary variables for skeletal pattern classification.

Conclusions:

• Support Vector Machines provide a robust model for classifying skeletal patterns using novel craniomaxillary variables.
• This approach is applicable to the significant portion of the Colombian population with non-Class I skeletal patterns.
• The method enhances forensic facial reconstruction capabilities by simulating mandibular position more accurately.