Scholarlink

Objectives <p>This study introduces an open-source intelligent anterior tooth restoration method utilizing a point cloud completion network and evaluates its accuracy compared to traditional manual design.</p> Materials and methods <p>A total of 209 digital dental models from patients treated at Beijing Stomatological Hospital (Jan 2024–Jan 2025) were analyzed. Missing maxillary right central incisors were simulated using Geomagic Wrap. The study was divided into two groups: manual design (MD) and AI-generated (AI-G). The AI-G group utilized an AdaPoinTr-based model. Reconstruction accuracy was quantified via L1/L2 Chamfer Distance (CD), F-Score, and Hausdorff Distance (HD). Statistics: Paired t-test, Wilcoxon signed-rank tests (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="784_2025_6618_Article_IEq1.gif" Format="GIF" Height="10" Rendition="HTML" Resolution="72" Type="Linedraw" Width="14" /> </InlineMediaObject> <EquationSource Format="TEX">\(\alpha\)</EquationSource> </InlineEquation>=0.05).</p> Results <p>The MD group exhibited superior local accuracy [L1-CD, MD (9.9171 ± 1.4654) vs. AI-G (15.2144 ± 0.2917), <i>P</i>&lt;0.001] and global matching [L2-CD, MD (2.9709 ± 1.3945) vs. AI-G (6.7731 ± 0.3179, <i>P</i>&lt;0.001) compared to AI-G. Microstructural retention (F-Score) also favored MD [MD 58.4955 (IQR: 11.8238) vs. AI-G 34.6080 (IQR: 0.6612), <i>P</i>&lt;0.001]. However, extreme error control (HD) showed no significant difference [MD 0.0490 mm (IQR: 0.1094 mm) vs. AI-G 0.1264 mm (IQR: 0.0073 mm), <i>P</i>=0.058]. AI-G demonstrated more stable error distribution (IQR: 0.0073 mm vs. 0.1094 mm).</p> Conclusions <p>This study introduces an innovative AdaPoinTr network for reconstructing the missing central incisor. Though slightly less accurate than manual design, the AI approach exhibits better stability and consistency, promising for intelligent prosthodontics.</p> Clinical relevance <p>This study validates the feasibility of AI-driven point cloud completion networks in 3D reconstruction tasks for missing central incisors, providing new research directions for the intelligent development of digital prosthodontics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Feasibility and accuracy assessment of AI-driven incisor restoration using 3D point cloud data

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Xiaohan Ma,
  • Guofen Lin,
  • Shengnan Zhao,
  • Xuechao Zou,
  • Huixin Xie,
  • Min Tang

摘要

Objectives

This study introduces an open-source intelligent anterior tooth restoration method utilizing a point cloud completion network and evaluates its accuracy compared to traditional manual design.

Materials and methods

A total of 209 digital dental models from patients treated at Beijing Stomatological Hospital (Jan 2024–Jan 2025) were analyzed. Missing maxillary right central incisors were simulated using Geomagic Wrap. The study was divided into two groups: manual design (MD) and AI-generated (AI-G). The AI-G group utilized an AdaPoinTr-based model. Reconstruction accuracy was quantified via L1/L2 Chamfer Distance (CD), F-Score, and Hausdorff Distance (HD). Statistics: Paired t-test, Wilcoxon signed-rank tests ( \(\alpha\) =0.05).

Results

The MD group exhibited superior local accuracy [L1-CD, MD (9.9171 ± 1.4654) vs. AI-G (15.2144 ± 0.2917), P<0.001] and global matching [L2-CD, MD (2.9709 ± 1.3945) vs. AI-G (6.7731 ± 0.3179, P<0.001) compared to AI-G. Microstructural retention (F-Score) also favored MD [MD 58.4955 (IQR: 11.8238) vs. AI-G 34.6080 (IQR: 0.6612), P<0.001]. However, extreme error control (HD) showed no significant difference [MD 0.0490 mm (IQR: 0.1094 mm) vs. AI-G 0.1264 mm (IQR: 0.0073 mm), P=0.058]. AI-G demonstrated more stable error distribution (IQR: 0.0073 mm vs. 0.1094 mm).

Conclusions

This study introduces an innovative AdaPoinTr network for reconstructing the missing central incisor. Though slightly less accurate than manual design, the AI approach exhibits better stability and consistency, promising for intelligent prosthodontics.

Clinical relevance

This study validates the feasibility of AI-driven point cloud completion networks in 3D reconstruction tasks for missing central incisors, providing new research directions for the intelligent development of digital prosthodontics.