본문 바로가기
  • Home

An Efficient Automatic Dental Prosthesis Design Strategy Based on Multimodal Fusion and Direct Mesh Generation using Transformer Architecture

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(6), pp.145~152
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : April 30, 2026
  • Accepted : June 9, 2026
  • Published : June 30, 2026

Se-hoon Park 1 Ho-Sung Choi 1

1박랩에이아이

Accredited

ABSTRACT

This study proposes a novel multimodal transformer framework for automatic dental prosthesis design, addressing the structural and functional limitations of existing computer-aided workflows. By robustly fusing 3D intraoral scans, 2D oral images, and optional facial scans, the architecture captures both topological constraints and broader esthetic contexts. A key innovation is the autoregressive mesh token decoder, which directly predicts vertices and faces, intentionally bypassing Signed Distance Field (SDF) or voxel intermediate representations. This eliminates the resolution-memory trade-off and preserves intricate details such as cusp tips. Furthermore, an inline occlusion validation layer is embedded during the generative sequence to dynamically mitigate collisions, thereby reducing the inefficiencies of post-hoc morphological revisions. To ensure clinical integration, a beam-search-based human-in-the-loop candidate management protocol is employed. Quantitative evaluations demonstrate significant improvements: post-design occlusal adjustments decreased from 2.3 to 0.5 per case, GPU memory consumption fell from 4.2 GB to 2.5 GB, and single prosthesis generation time was reduced from 180 to 120 seconds, while achieving a cusp-tip reconstruction accuracy of 90–96%.

Citation status

* References for papers published after 2024 are currently being built.