LOME-Laboratory of Optics and Experimental Mechanics, INEGI-Institute of Mechanical Engineering and Industrial Management, Faculty of Engineering, University of Porto, Porto, Portugal,
Int J Comput Assist Radiol Surg. 2013 Sep;8(5):711-21. doi: 10.1007/s11548-012-0802-6. Epub 2012 Dec 2.
Medical imaging and in particular digital radiographic images offer a great deal of information to dentists in the clinical diagnosis and treatment processes on a daily basis. This paper presents a new method aimed to produce an accurate segmentation of dental implants and the crestal bone line in radiographic images. With this, it is possible computing several measures to biomechanical and clinical evaluation of dental implants positioning and evolution.
The proposed segmentation method includes two major steps: (1) the preprocessing that combine denoising filters, morphological operations and histogram threshold techniques and (2) the final segmentation involving made-to-measure adjusted and trained active shape models for detecting the precise location of the intended structures.
Resulting measurements were compared to manual measurements made by experts on representative radiographs from patients. The calculated intraclass correlation coefficient was 0.75 and showed good reliability of the method, and the Bland-Altman analysis showed 95% of the values within the limits of agreement. The mean of the differences between the manual and method-driven measurements was 0.049 mm ([Formula: see text]) 95% CI, inferior to the established limit (0.15mm).
It was demonstrated that the method achieved a precise segmentation of the intended structures. The validation process on standardized periapical radiographs showed good agreement between the manual measurements and the ones produced by the new method. Future work will be focused on making the method more robust to densitometry changes and to validate the method on non-standardized radiographs.
医学成像，特别是数字射线照相图像，在日常临床诊断和治疗过程中为牙医提供了大量信息。本文提出了一种新方法，旨在对射线照相图像中的牙种植体和牙槽嵴骨线进行精确分割。通过这种方法，可以计算出几个用于评估牙种植体定位和演变的生物力学和临床指标。
所提出的分割方法包括两个主要步骤：（1）预处理，包括去噪滤波器、形态操作和直方图阈值技术；（2）最终分割，涉及定制调整和训练的主动形状模型，以检测目标结构的精确位置。
所得测量值与专家在患者代表性射线照片上进行的手动测量值进行了比较。计算的组内相关系数为 0.75，表明该方法具有良好的可靠性，Bland-Altman 分析显示 95%的值在可接受范围内。手动和方法驱动测量值之间的差异平均值为 0.049mm（[公式：见文本]）95%置信区间，低于既定限制（0.15mm）。
结果表明，该方法实现了对目标结构的精确分割。在标准化根尖射线照片上的验证过程表明，手动测量值与新方法产生的测量值之间具有良好的一致性。未来的工作将集中于使该方法对密度测量变化更具鲁棒性，并在非标准化射线照片上验证该方法。
