组织工程与重建外科杂志 ›› 2024, Vol. 20 ›› Issue (6): 605-.

• • 上一篇    下一篇

基于人工智能技术的颅颌面畸形自动化头影测量研究

  

  • 出版日期:2024-12-02 发布日期:2025-01-03

Automated cephalometric analysis of craniomaxillofacial deformities based on artificial intelligence technologies

  • Online:2024-12-02 Published:2025-01-03

摘要:

 目的 开发一种新的自动标记点检测框架,用于严重颅颌面畸形(Craniomaxillofacial deformities,CMFdeformities)的诊断和治疗,解决其数据量少、形态差异大的问题。方法 本研究基于三维点云变形模型和深度学习网络的方法,首先使用正常人数据通过变形模拟严重 CMF 患者数据进行数据增强,然后通过三维点云卷积神经网络(Convolutional neural network,CNN)语义分割模型进行标记点的粗略定位,再根据标记点是否位于骨缺损区域,分别使用不同的模型进行精细定位。结果 上述方法在正常标记点和缺损标记点的检测上均优于现有技术,CT扫描下的正常标记点和缺陷标记点的平均误差分别为1.19 mm和1.13 mm,CBCT扫描下分别为0.91 mm和0.94 mm。结论 新方法能有效提高严重CMF畸形标志点检测的准确性,对临床手术设计和患者治疗具有重要意义。

关键词:

Abstract:

Objective To develop a new automatic landmark detection framework for the diagnosis and treatment of patients with severe craniomaxillofacial (CMF) deformities, addressing the issues of limited data quantity and large morphological differences. Methods This study proposed a method based on a three-dimensional (3D) point cloud deformation model and deep learning networks. First, normal human data was deformed to simulate severe CMF patient data for data augmentation. Then, a coarse-to-fine strategy was adopted, where initial coarse localization of landmarks was performed using a 3D point cloud convolutional neural network (CNN) semantic segmentation model, followed by fine localization using different models based on whether the landmarks are located in bone defect areas. Results The experiments demonstrated that the proposed method outperformed existing technologies in the detection of both normal and defective landmarks. The average errors for normal landmarks and defective landmarks detected under CT scanning were 1.19 mm and 1.13 mm, respectively, and under CBCT scanning were 0.91 mm and 0.94 mm, respectively. Conclusion The new method can effectively improve the accuracy of landmark detection for severe CMF deformities, which is significant for clinical surgical design and patient treatment.

Key words: