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

BACKGROUND

Sex classification based on the mandible is a challenging problem due to morphological similarities. In artificial intelligence classification studies, a single source of morphological information may not be sufficient.

OBJECTIVES

A hybrid and explainable machine learning approach combining different levels of morphological information about the mandible was developed in this study.

METHODS

Analyses were performed on 52 (31 female, 21 male) rat mandibles raised under the same conditions. A total of 24 osteometric parameters were calculated from two-dimensional photographs of the mandible. Furthermore, features were extracted from the mandibular photographs using handcrafted and pretrained deep learning. Machine learning models were first built separately using the features. Then, all features were combined to design a hybrid machine learning model. All models were tested using a 5-fold cross-validation method.

RESULTS

The logistic regression model trained with the hybrid feature set (Osteometric + Handcrafted + ResNet50) achieved the highest performance with 96.2% accuracy, 94.3% F1-score and 99.2% ROC-AUC value. SHAP explanatory AI analyses were performed to assess the interpretability of model decisions.

CONCLUSIONS

This study will contribute to the literature by classifying rat mandible images using artificial intelligence and explaining these results with XAI.