| 项明娣,胡万龙,曹佳红,闫芸,谭洁,冯斌,陈理杰,钱天仕,吴福理.基于深度学习的二维头影测量定点分析系统的研究[J].口腔材料器械杂志,2024,33(2):100-106. |
| 基于深度学习的二维头影测量定点分析系统的研究 |
| Research on two-dimensional cephalometric landmark point positioning and analysis system based on deep learning |
| 投稿时间：2024-04-02 修订日期：2024-04-12 |
| DOI：10.11752/j.kqcl.2024.02.06 |
| 中文关键词: 人工智能 深度学习 卷积神经网络 头影测量 |
| 英文关键词:Artificial intelligence Deep learning Convolutional neural network Cephalometric Analysis |
| 基金项目:浙江省教育厅项目(编号:Y202249954) |
|
| 摘要点击次数: 322 |
| 全文下载次数: 272 |
| 中文摘要: |
| 目的 探讨基于深度学习的图像高斯金字塔和卷积神经网络方法，构建和训练一种高精准度的自动化二维头影测量标志点定位与分析系统模型。方法 收集2021年1月至12月期间本院所摄的400张年龄在18至50周岁且无牙列缺损的头颅侧位影像，在3D slicer（NIH美国）上完成每张44个牙颌和颅面软硬组织结构标志点的标注，并构建和训练基于图像高斯金字塔和卷积神经网络的自动化二维头影测量标志点定位和分析系统。结果 运用图像高斯金字塔和卷积神经网络方法能高精准获取44个牙颌和颅面的软硬组织结构标志点，在 2.0 mm、2.5 mm、3.0 mm、4.0 mm 精度范围内预测的平均准确率分别为 85.61 %、90.72 %、93.82 %、96.34 %；44个牙颌和颅面软硬组织结构标志点的平均误差为 1.22 mm，平均标准差为 1.27 mm；常见头影测量项目（ANB、SNA、SNB、ODI、APDI、FHI、FMA、MW）的平均预测准确率为 85.00 % 。结论 运用图像高斯金字塔和卷积神经网络方法能高精准获取牙颌和颅面的软硬组织结构标志点，并且对牙颌和颅面形态分析诊断具有良好的准确性，该技术将有助于推进自动化头影测量的临床运用。 |
| 英文摘要: |
| Objective This study aims to explore the application of deep learning techniques, including image Gaussian pyramid and convolutional neural networks, to construct and train a highly accurate automated 2D cephalometric landmark localization and analysis system model. Methods A total of 400 lateral cephalometric images of individuals aged 18 to 50 years without dentition defects were collected from January to December 2021 in the hospital. For each 2D cephalometric image, 44 landmark points of dental, maxillofacial, and craniofacial soft and hard tissue structures were annotated in a 3D slicer, and an automated 2D cephalometric landmark point localization and analysis system based on an image Gaussian pyramid and convolutional neural network was constructed and trained. Results The application of the image Gaussian pyramid and convolutional neural networks achieved high accuracy in obtaining the 44 soft and hard tissue landmarks of the dental, maxillofacial, and craniofacial structures. The average accuracy of prediction in the 2.0 mm, 2.5 mm, 3.0 mm, and 4.0 mm were 85.61 %, 90.72 %, 93.82 %, and 96.34 %, respectively. The average error and standard deviation for the 44 landmarks were 1.22 mm and 1.27 mm, respectively. The average prediction accuracy of common cephalometric measurements (ANB, SNA, SNB, ODI, APDI, FHI, FMA, MW) was 85.00 %. Conclusion The application of the image Gaussian pyramid and convolutional neural networks can obtain the soft and hard tissue landmarks of the dental, maxillofacial, and craniofacial structures with high precision and good accuracy for diagnosing dentition and craniofacial morphology. This technique can contribute to the clinical application of automated cephalometric measurements. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|