Artificial Intelligence Breakthrough in Differentiating External Cervical Resorption (ECR) from Caries

IRAN: Researchers from Sharif University of Technology in Tehran, Iran, have published a study presenting a significant stride in the application of Artificial Intelligence (AI) for the detection of External Cervical Resorption (ECR). The study, published on November 15, 2023, outlines their innovative use of label-efficient self-supervised learning (SSL) to train a model capable of discerning ECR from caries with remarkable accuracy.

Leveraging SSL for ECR Detection

The primary objective of the research was to harness label-efficient self-supervised learning to develop a model proficient in detecting External Cervical Resorption and distinguishing it from dental caries. The methodology involved collecting periapical (PA) radiographs of teeth exhibiting ECR defects. Board-certified endodontists independently reviewed the PA radiographs and cone beam computed tomographic (CBCT) images to establish the ground truth regarding the presence of ECR. The radiographic data were then categorized into three regions of interest (ROIs): healthy teeth, teeth with ECR, and teeth with caries.

AI Advances Dental Caries Detection, Study Finds

Comprehensive Model Assessment

To assess the effectiveness of their approach, the researchers implemented nine contrastive SSL models, including SimCLR v2, MoCo v2, BYOL, DINO, NNCLR, SwAV, MSN, Barlow Twins, and SimSiam. Additionally, seven baseline deep learning models, namely ResNet-18, ResNet-50, VGG16, DenseNet, MobileNetV2, ResNeXt-50, and InceptionV3, were employed for comparison. The evaluation utilised a 10-fold cross-validation strategy and a hold-out test set.

Promising Results and Superior Performance

Results from the 10-fold cross-validation revealed that most SSL models surpassed the transfer learning baseline models. Notably, the DINO model achieved the highest mean accuracy at 85.64% ± 4.56, outperforming 13 other models significantly (p<.05). DINO also exhibited the highest test set accuracy at 84.09%, while MoCo v2 demonstrated the highest recall (77.37%) and F1-score (82.93%).

Implications for Clinical Practice

The study’s findings indicate a substantial breakthrough in the realm of dental diagnostics, demonstrating that AI can significantly assist clinicians in detecting External Cervical Resorption and effectively differentiate it from caries. Furthermore, the introduction of label-efficient self-supervised learning in this context suggests that SSL-based models can outperform traditional transfer learning approaches, potentially reducing the dependence on large, labeled datasets.

New Approach Offers Hope in Battle Against Dental Caries