Morphological Sex Classification in Rat Mandibles Using a Hybrid and Explainable Machine Learning

Area of Science:

• Comparative anatomy
• Artificial intelligence in biology
• Machine learning for biological classification

Background:

• Sex classification from mandibles is challenging due to subtle morphological differences.
• Single sources of morphological data may be insufficient for accurate AI classification.
• Developing robust AI models requires integrating diverse feature types.

Purpose of the Study:

• To develop a hybrid and explainable machine learning (ML) approach for sex classification using rat mandibles.
• To combine multiple levels of morphological information for improved classification accuracy.
• To enhance the interpretability of AI models in biological sex determination.

Main Methods:

• Analysis of 24 osteometric parameters from 52 rat mandibles (31 female, 21 male).
• Feature extraction using handcrafted methods and deep learning (ResNet50) from mandibular images.
• Development of individual and hybrid ML models, evaluated with 5-fold cross-validation.

Main Results:

• The hybrid logistic regression model integrating osteometric, handcrafted, and deep learning features achieved 96.2% accuracy.
• The model demonstrated high performance with a 94.3% F1-score and 99.2% ROC-AUC.
• SHapley Additive exPlanations (SHAP) analysis provided insights into the model's decision-making process.

Conclusions:

• This research successfully classifies rat mandible images using a hybrid AI approach.
• The study contributes explainable AI (XAI) methods to the field of biological sex classification.
• The findings highlight the potential of integrating diverse morphological data for accurate and interpretable AI-driven biological analyses.