Prediction of gender from radiographic condylar and coronoid measurements using elastic net and random forests.

Aim: This study aims at the prediction of gender from radiographic condylar and coronoid measurements using random forest and elastic net algorithms. Background: Artificial intelligence (AI) has the potential to revolutionise the process of determining gender from skeletal remains by enhancing objectivity, efficiency, and accuracy. AI systems can be trained to automatically assess skeletal features relevant to gender identification, such as the size of the pelvis, skull, and specific mandibular traits. Materials and Methods: A total of 200 digital panoramic radiographs were collected, out of which 100 were males and 100 were females. The average age range of the samples was 20-40 years. Coronoid height and condylar height were measured using Planmeca Romexis Viewer Software version 2.9.2.R (Planmeca OY, Helsinki, Finland). Random forest and elastic net algorithms were employed in the study. Results: The 20-30 years group had an average age of 25.68 years, while the 31-40 years group had an average age of 35.32 years. The 20-30 years group had a lower range and variability compared to the 31-40 years group. Both age groups had similar median values, but the 20-30 years group had slightly higher variability. In elastic net algorithms, the true positive rate was 0.925, indicating high accuracy in identifying positive cases. The random forest model's performance metrics included a precision of 0.7368, recall of 0.875, and F1-score of 0.79, indicating its effectiveness in predicting genders. A high AUC of 0.952 was observed. Conclusion: The study shows that machine learning models can achieve high accuracy in gender prediction. However, future research should expand the sample size, explore additional features, and conduct cross-validation for applicability.