中国实用口腔科杂志

中国实用口腔科杂志 ›› 2024, Vol. 17 ›› Issue (2): 166-173.DOI: 10.19538/j.kq.2024.02.007

• 论著 • 上一篇    下一篇

基于人工智能的正畸病例客观评级系统临床应用效果研究

王    巍1，王宇凤2，龚蓓文1，王少烽1，乔清晨1，孙雅溪1，左飞飞2，王亚杰2，谢贤聚1   

  1. 1. 首都医科大学附属北京口腔医院正畸科，北京 100050；2. 北京朗视仪器股份有限公司，北京  100084
  • 出版日期:2024-03-30 发布日期:2024-03-30
  • 基金资助:
    北京市自然科学基金‑海淀原始创新联合基金（L222024）；北京市医院管理中心临床医学发展专项（ZLRK202330）

  • Online:2024-03-30 Published:2024-03-30

摘要： 目的    研究基于人工智能的美国正畸专家认证委员会（American Board of Orthodontics，ABO）客观评级系统（objective grading system，OGS）临床应用效果。方法    收集2016年1月至2023年7月于首都医科大学附属北京口腔医院正畸科结束正畸治疗的38例患者治疗后石膏模型行回顾性分析，使用iTero Element®扫描仪生成数字化模型，应用人工智能ABO-OGS测量系统自动测量评分，记为人工智能组；1名主治医师和1名主任医师基于3D打印数字化模型使用ABO测量尺行手动测量评分，分别记为手动测量主治组和手动测量主任组。共测量2次，间隔2周，记录测量时间和各项指标评分。结果    人工智能组、手动测量主治组和手动测量主任组测量评分时间依次为（0.18 ± 0.00）、（292.94 ± 27.10）、（210.00 ± 21.21）s，差异具有统计学意义（F = 980.288，P < 0.001）；3组均具有较好的测量评分一致性，ABO-OGS总分的ICC值分别达0.897、0.869、0.929；仅人工智能组各项指标2次测量评分比较，差异均无统计学意义（均P > 0.05）。人工智能组在颊舌向倾斜度指标2次测量评分的差值低于手动测量主治组，但边缘嵴高度指标差值高于手动测量主任组，差异均有统计学意义（均P < 0.05）。基于2次测量评分的均值进行组间比较分析时，人工智能组的大部分指标测量评分偏高，与手动测量主治组相比，其覆盖、咬合接触指标及总分较高；与手动测量主任组相比，其排齐、边缘嵴高度、颊舌向倾斜度、覆盖和咬合接触指标及总分较高，差异均有统计学意义（均P < 0.05）。结论    人工智能ABO-OGS测量系统重复性较好且用时短，但其测量评分普遍高于手动评分，仍需进一步调整优化。

关键词: 客观评级系统, 人工智能, 正畸病例, 可重复性

Abstract: Objective    To study the clinical application effect of the American Board of Orthodontics （ABO） objective grading system（OGS）based on artificial intelligence（AI）. Methods    Post-treatment plaster models of 38 patients，who completed orthodontic treatment at the Department of Orthodontics，Beijing Stomatological Hospital Affiliated to Capital Medical University from January 2016 to July 2023，were collected for retrospective analysis. The iTero Element® scanner was used to generate digital models. An AI-based ABO-OGS measurement system was then employed to perform automatic assessment based on digital models，which was recorded as the AI group；1 attending physician and 1 senior chief physician used the ABO measuring ruler to perform manual measurement based on the 3D printed digital model，which were recorded as the manual measurement attending（MMA）group and the manual measurement chief（MMC）group，respectively. Measurements were performed twice with an interval of 2 weeks as a washout period. The scores and time consumption were recorded. Results    The measurement time-consumption of the AI group，MMA group and MMC group was（0.18 ± 0.00）s，（292.94 ± 27.10）s，and（210.00 ± 21.21）s，respectively，showing a statistically significant difference（F = 980.288，P < 0.001）. All three groups had decent measurement consistency，with the ICC values of the ABO-OGS total score reaching 0.897，0.869 and 0.929，respectively. Additionally，only in the AI group，there was no statistical difference in the comparison of the two measurement scores of various indexes（all P > 0.05）. The difference between the two measurement scores of the buccolingual inclination index in the AI group was lower than that in the MMA group，but the difference in marginal ridges index was higher than that in the MMC group，and the differences were all statistically significant（both P < 0.05）. Comparative analysis among groups was conducted based on the average score of the two measurements，and the AI group displayed higher measurement scores for most indexes. Compared with the MMA group，the overjet and occlusal contact indexes scores and total score in AI group were higher. Compared with the MMC group，the alignment，marginal ridges，buccolingual inclination，overjet and occlusal contact indexes scores and total score in AI group were higher，and the differences were statistically significant（all P < 0.05）. Conclusion    The AI-based ABO-OGS exhibits decent repeatability in measurements with higher time efficiency；however，its measurement scores are generally higher than manual scores，which still needs further optimization.

Key words: objective grading system, artificial intelligence, orthodontic cases, repeatability