Key points are not available for this paper at this time.
Background: Artificial Intelligence (AI) significantly enhances the diagnosis of pericoronal radiolucency by accurately interpreting dental radiographs. Through advanced algorithms, AI can identify early signs of abnormalities near unerupted teeth. This helps clinicians differentiate between benign and malignant conditions, leading to more informed decisions; improved treatment plans, ultimately benefiting patient care and outcomes. Method: ology: A total of 2500 radiographs were screened of which 1070 radiographs were used in the study. 315 images of pericoronal radiolucency in mandibular third molars and 755 images of the normal mandibular third molars were included. The AI algorithms employed in the study were Logistic regression and Naive Bayes. Accuracy, sensitivity, specificity, precision, recall, F1, AUC-ROC curve were used for performance evaluation. Results: This study found that Logistic regression model showed slightly higher accuracy than Naive Bayes model in predicting peri coronal radiolucency. In performance prediction for logistic regression model in predicting pericoronal radiolucency in third molars in 315 images, showed a slightly higher rate of prediction of 58.3 %, whereas, Naive Bayes model showed a comparatively lower prediction of pericoronal radiolucency, 52.2 %. During performance evaluation, Logistic regression performed better in CA, F1, and Recall, and Naive Bayes performed better in AUC and Precision model. Conclusion: The current study demonstrated that Logistic regression have slightly highest accuracy in detecting pericoronal radiolucency in digital orthopantomogram images, which is consistent with the normal radiographic evaluation. Also, the Naive Bayes algorithm showed a fairly considerable performance in the classification of pericoronal radiolucencies.
Madhumitha et al. (Sun,) studied this question.