Comparative Evaluation of Artificial Intelligence Models for Gender Prediction Using Coronoid Process Morphology on Digital OPG Images: A Study of Logistic Regression, Random Forest, and k-Nearest Neighbours

Abstract

Aim: This study evaluates the efficacy of artificial intelligence (AI) in determining gender based on the morphological characteristics of the coronoid process observed in digital orthopantomogram (OPG) images.

Methodology: A dataset of 200 Orthopantomograms (OPG), comprising 100 male and 100 female images, was collected from individuals aged 18 to 45 years. Low-resolution images and OPGs with extensive dental surgeries were excluded. Coronoid processes were classified into different shapes according to Shakya et al. (2013) and further divided by gender. The dataset was split into 80% for training and 20% for testing. Preprocessing, including normalization and noise removal, was performed using Picsart software to ensure clean input data.

Model Training and Evaluation: Three machine learning models—Logistic Regression, Random Forest, and k-Nearest Neighbors (k-NN)—were trained and evaluated using accuracy, precision, recall, specificity, and LogLoss.

Results: Logistic Regression demonstrated the highest AUC (0.619) and Specificity (0.871), indicating strong performance in distinguishing gender. Despite a lower recall (0.230), it showed balanced performance across key metrics, outperforming both Random Forest and k-NN. Random Forest exhibited the lowest recall (0.190) and highest LogLoss (7.421), while k-NN had moderate performance but the highest LogLoss (16.695).

Conclusion: Logistic Regression proved to be the most effective model for gender determination based on coronoid process morphology. Future studies should expand datasets and integrate advanced imaging techniques to improve model generalizability and clinical applicability. This research highlights AI’s potential in enhancing diagnostic accuracy in dental imaging.