Automatic margin line extraction using 3D deep learning on digital surface models of prepared teeth for crown generation

We propose a personalized digital solution to create margin lines on prepared teeth automatically and consistently during a dental crown design. Using artificial intelligence (AI), the proposed framework disrupts the current nonrepeatable manual margin line design procedure. We used supervised deep learning on 3D scanned prepared teeth to segment them in two regions such that margin lines coincide with the boundaries separating the two regions. A real-world dataset composed of 1113 digital surface scans of prepared teeth was labeled based on crown bottoms generated by dental professionals. The training set contained 1035 cases and the test set contained 78 cases, with both covering all teeth positions. A majority voting classifier was used to combine results of 5-folds of the trained model to enhance the segmentation performance. A novel post-processing procedure was implemented to correct segmentation results and extract the margin lines. The predicted margin lines were compared with the ground truth margins produced by dental technicians. Out of 78 test cases, the proposed model extracted 70 successful margin lines based on a ground truth-to-predicted maximum cut-off distance of 200 μm, and 62 for a threshold of 100 μm. Our margin line segmentation post-processing procedure was compared with the traditional graph-cut technique and showed superior increase in accuracy, DSC = 0.986 versus 0.974, respectively (p-value 0.024). The proposed end-to-end AI framework is a consistent and time-efficient digital tool for automatic margin line extraction from 3D surface models of prepared teeth.