Text Condition Embedded Regression Network for Automated Dental Abutment Design

TL;DR

The paper tackles the bottleneck of manually designing dental implant abutments by introducing TCEAD, a text-conditioned, mesh-based regression framework that leverages large-scale oral-scan pre-training and a text-guided localization module to accurately predict abutment parameters. By pre-training on mesh patches and employing a CLIP-derived text description to focus attention on the implant area, the method achieves superior IoU across transgingival, diameter, and height predictions compared to state-of-the-art approaches. The approach demonstrates robust performance gains, particularly in height regression, and offers a path toward automated, efficient, and clinician-aligned abutment design. The work highlights the potential of integrating text and mesh representations to accelerate digital dentistry workflows and reduces reliance on manual design processes.

Abstract

The abutment is an important part of artificial dental implants, whose design process is time-consuming and labor-intensive. Long-term use of inappropriate dental implant abutments may result in implant complications, including peri-implantitis. Using artificial intelligence to assist dental implant abutment design can quickly improve the efficiency of abutment design and enhance abutment adaptability. In this paper, we propose a text condition embedded abutment design framework (TCEAD), the novel automated abutment design solution available in literature. The proposed study extends the self-supervised learning framework of the mesh mask autoencoder (MeshMAE) by introducing a text-guided localization (TGL) module to facilitate abutment area localization. As the parameter determination of the abutment is heavily dependent on local fine-grained features (the width and height of the implant and the distance to the opposing tooth), we pre-train the encoder using oral scan data to improve the model's feature extraction ability. Moreover, considering that the abutment area is only a small part of the oral scan data, we designed a TGL module, which introduces the description of the abutment area through the text encoder of Contrastive Language-Image Pre-training (CLIP), enabling the network to quickly locate the abutment area. We validated the performance of TCEAD on a large abutment design dataset. Extensive experiments demonstrate that TCEAD achieves an Intersection over Union (IoU) improvement of 0.8%-12.85% over other mainstream methods, underscoring its potential in automated dental abutment design.

Figures (7)

• Figure 1: The process of manually determining abutment parameters. The blue part in 3D oral scan data represents the position of the cross-section, and the three 2D cross-sections represent the measurements of the transgingival, diameter and height, respectively.
• Figure 2: The tooth index according to the FDI World Dental Federation numbering system.
• Figure 3: Visualization of the remesh processing. First, perform manifold and simplification operations on the original mesh to process the oral scan data into a base mesh with 500 standard faces. Furthermore, refine the base mesh three times, each time refining one face into four faces. Finally, the refined mesh is projected as input. The refined mesh is divided into 500 patches, each of which contains 64 faces arranged in a fixed order after remesh processing.
• Figure 4: The overall framework of the proposed TCEAD method. The input mesh is first divided into multiple non-overlapping patches through the remeshing and then embedded through an MLP. In the pre-training stage, a random subset of patches is masked, and only the visible patches are processed by the mesh encoder, while the decoder reconstructs the original input using both the encoded visible embeddings and the masked embeddings. During the fine-tuning stage, the pre-trained weights are transferred to the network to support downstream tasks. Subsequently, the text-guided localization (TGL) module fuses the abutment position description with the mesh features and feeds them to the regression head. Finally, the regression head predicts the abutment parameters.
• Figure 5: Ablation experiment of the TGL component. The vertical axis represents the IoU performance, while the horizontal axis corresponds to value of each abutment parameter.