Application of deep convolutional neural network models in intra oral periapical radiograph for prediction of endodontic treatment outcomes

Introduction: Modern endodontic treatments are highly effective in saving teeth thatmight otherwise need to be extracted. However, like any medical procedure, there isalways a chance of failure. Predicting potential failure during the preoperative phase iscrucial to ensure that patients receive the most appropriate treatment. This study aimedto evaluate the performance of a deep learning-based segmentation model for predictingoutcomes of non-surgical endodontic treatment.Methods: Preoperative and 3-year postoperative periapical radiographic images of eachtooth from routine root canal treatments performed by endodontists from 2015 to 2021were obtained retrospectively from Thammasat University hospital. Preoperativeradiographic images of 1200 teeth with 3-year follow-up results (440 healed, 400healing, and 360 disease) were collected. Mask Region-based Convolutional NeuralNetwork (Mask R-CNN) was used to pixel-wise segment the root from other structuresin the image and trained to predict class label into healed, healing and disease. Threeendodontists annotated 1080 images used for model training, validation, and testing.The performance of the model was evaluated on a test set and also by comparison withthe performance of clinicians (general practitioners and endodontists) with and withoutthe help of the model on independent 120 images.Results: The performance of the Mask R-CNN prediction model was high with themean average precision (mAP) of 0.88 (95% CI 0.83-0.93) and area under the precisionrecall curve of 0.91 (95% CI 0.88-0.94), 0.83 (95% CI 0.81-0.85), 0.91 (95% CI 0.90-0.92) onhealed, healing and disease, respectively. The prediction metrics of general practitionersand endodontists significantly improved with the help of Mask R-CNN outperformingclinicians alone with mAP increasing from 0.75 (95% CI 0.72-0.78) to 0.84 (95% CI 0.81-0.87) and 0.88 (95% CI 0.85-0.91) to 0.92 (95% CI 0.89-0.95), respectively Conclusion: Deep learning-based segmentation model had the potential to predict nonsurgical endodontic treatment outcomes from periapical radiographic images and wereexpected to aid in endodontic treatment.