Comparison of mandibular morphometric parameters in digital panoramic radiography in gender determination using...

Area of Science:

• Forensic Anthropology
• Radiology
• Machine Learning

Background:

• Gender determination is crucial in forensic and clinical contexts.
• Mandibular morphometric analysis from radiographs offers a potential non-invasive method.
• Machine learning algorithms can analyze complex patterns in radiographic data.

Purpose of the Study:

• To evaluate the efficacy of machine learning algorithms in gender determination using morphometric features from digital panoramic radiographs.
• To identify key mandibular morphometric features that contribute most to accurate gender classification.

Main Methods:

• Extraction of twelve morphometric measurements from 200 digital panoramic radiographs.
• Application of six machine learning algorithms (k-NN, decision trees, SVM, Naive Bayes, LDA, neural networks).
• Rigorous validation using tenfold cross-validation repeated 10 times for each classification process.

Main Results:

• An overall accuracy of 82.6% was achieved when all 12 features were used.
• Coronoid height (80.9%), condyle height (78.2%), and ramus height (77.2%) were the most accurate individual features.
• The Naive Bayes algorithm yielded the highest classification accuracy at 84.0%.

Conclusions:

• Machine learning effectively determines gender from mandibular morphometric structures on panoramic radiographs.
• Combined analysis of multiple morphometric features, particularly those selected by algorithms like MRMR, maximizes classification accuracy.
• This approach provides a reliable tool for gender determination in forensic and anthropological studies.