OBJECTIVE: The aim of this study was to examine the success of deep learning-based convolutional neural networks (CNN) in the detection and differentiation of amalgam, composite resin, and metal-ceramic restorations from bitewing and periapical radiographs. METHOD AND MATERIALS: Five hundred and fifty bitewing and periapical radiographs were used. Eighty percent of the images were used for training, and 20% were left for testing. Twenty percent of the images allocated for training were then used for validation during learning. The image classification model was based on the application of CNN. The model used Resnet34 architecture, which is pre-trained on the ImageNet dataset. Average sensitivity, receiver operating characteristic (ROC) curve, and area under the curve (AUC) were calculated for performance evaluation of the model. RESULTS: The model training loss was 0.13, and the validation loss was 0.63. The independent test group result was 0.67. Amalgam AUC was 0.95, composite AUC was 0.95, and metal-ceramic AUC was 1.00. The average AUC was 0.97. The false positive rate in the validation set was 18, the false negative rate was 18, the true positive rate was 60, and the true negative rate was 138. The true positive rate was 0.82 for amalgam, 0.75 for composite, and 0.73 for metal-ceramic. CONCLUSION: Deep learning-based CNNs from periapical and bitewing radiographs appear to be a promising technique for the detection and differentiation of restorations.
OBJECTIVE: The aim of this study was to examine the success of deep learning-based convolutional neural networks (CNN) in the detection and differentiation of amalgam, composite resin, and metal-ceramic restorations from bitewing and periapical radiographs. METHOD AND MATERIALS: Five hundred and fifty bitewing and periapical radiographs were used. Eighty percent of the images were used for training, and 20% were left for testing. Twenty percent of the images allocated for training were then used for validation during learning. The image classification model was based on the application of CNN. The model used Resnet34 architecture, which is pre-trained on the ImageNet dataset. Average sensitivity, receiver operating characteristic (ROC) curve, and area under the curve (AUC) were calculated for performance evaluation of the model. RESULTS: The model training loss was 0.13, and the validation loss was 0.63. The independent test group result was 0.67. Amalgam AUC was 0.95, composite AUC was 0.95, and metal-ceramic AUC was 1.00. The average AUC was 0.97. The false positive rate in the validation set was 18, the false negative rate was 18, the true positive rate was 60, and the true negative rate was 138. The true positive rate was 0.82 for amalgam, 0.75 for composite, and 0.73 for metal-ceramic. CONCLUSION: Deep learning-based CNNs from periapical and bitewing radiographs appear to be a promising technique for the detection and differentiation of restorations.
Entities:
Keywords:
artificial intelligence; dental restorations; digital radiology