(PO10) Predicting Endodontic Outcomes with Deep Convolutional Neural Networks

Abstract: Convolutional Neural Networks (CNNs) are under study for diagnostic tasks in the field of medical imaging. Our previous study (Lee et al., 2023) presented the applicability of CNNs in predicting the prognosis of endodontic treatments. The aim of our research is to develop a prognosis prediction model with a high level of accuracy and versatility. A database was prepared with all types of teeth that received endodontic treatment or retreatment by endodontists, with a follow-up period of one year (n=2,603). We altered the previous model structure, replacing average pool with attention pool. The model was trained, validated, and tested to predict the one-year endodontic prognosis by analyzing preoperative periapical radiographs as input. The outcome was compared to that of the previous study. Compared to the previous model, accuracy increased from 67.0% to 72.0%; sensitivity increased from 40.7% to 65.8%; precision increased from 44.3% to 70.2%; and the F1 score increased from 42.0% to 69.0%. However, specificity decreased from 78.3% to 68.5%. Additionally, the area under the average receiver-operating-characteristic of the model increased from 0.638 to 0.72.Deep convolutional neural networks can effectively predict endodontic outcomes using radiographic clinical features, providing an opportunity for enhanced diagnostic accuracy and more informed clinical decision-making.