Fourth Clinical Division, School and Hospital of Stomatology, Peking University, Beijing, China.
Ling. AI, Beijing, China.
Med Image Anal. 2021 Feb;68:101904. doi: 10.1016/j.media.2020.101904. Epub 2020 Nov 18.
Cephalometric analysis is a fundamental examination which is widely used in orthodontic diagnosis and treatment planning. Its key step is to detect the anatomical landmarks in lateral cephalograms, which is time-consuming in traditional manual way. To solve this problem, we propose a novel approach with a cascaded three-stage convolutional neural networks to predict cephalometric landmarks automatically. In the first stage, high-level features of the craniofacial structures are extracted to locate the lateral face area which helps to overcome the appearance variations. Next, we process the aligned face area to estimate the locations of all landmarks simultaneously. At the last stage, each landmark is refined through a dedicated network using high-resolution image data around the initial position to achieve more accurate result. We evaluate the proposed method on several anatomical landmark datasets and the experimental results show that our method achieved competitive performance compared with the other methods.
头影测量分析是一种广泛应用于正畸诊断和治疗计划的基本检查。其关键步骤是检测侧位头颅片中的解剖标志点，这在传统的手动方法中非常耗时。为了解决这个问题，我们提出了一种新的方法，使用级联的三阶段卷积神经网络自动预测头影测量标志点。在第一阶段，提取颅面结构的高级特征来定位有助于克服外观变化的侧面部区域。接下来，我们处理对齐的面部区域，同时估计所有标志点的位置。在最后一个阶段，使用初始位置周围的高分辨率图像数据通过专用网络细化每个标志点，以获得更准确的结果。我们在几个解剖标志数据集上评估了所提出的方法，实验结果表明，与其他方法相比，我们的方法具有竞争力。
