基于轮廓匹配的正交双平面X线图像重建胸段脊柱3D模型

Three-dimensional Reconstruction of Thoracic Spine from Biplanar Radiographs Based on Contour Matching

目的由正交双平面X线图像重建胸段脊柱三维模型。方法通过估计椎体模型的方向和位置重建脊柱模型。首先基于对极几何原理确定椎体模型的位置。然后将三维椎体模型进行分层投影后,基于Hough变换匹配投影轮廓和X线图像中对应椎体边缘,以估计椎体模型的方向。结果三位测试者对16例胸段脊柱正、侧位相X线图像重建结果比较显示:重建可重复度位置误差在2.5 mm内,方向误差在2.6°内;和传统重建方法比较位置误差为3.1 mm,方向误差为3.0°。重建手工操作时间在3 min内,计算机平均运算时间为115 min。结论 Hough变换匹配边界能够估计椎体方向,提出的重建方法有较好的可重复度和准确度,可望用于辅助医生从X线图像对三维脊柱畸形进行诊断。

Objective To reconstruct the three-dimensional （3D） thoracic spine model from biplanar radio- graphs. Methods The spine was reconstructed by estimating vertebral orientations and locations. First, verte- bral locations were determined based on the epipolar geometry. Vertebral orientations were then estimated by matching the projected contours of the standard 3D primitives with the edges of the corresponding vertebrae in biplanar radiographs based on the Hough transform technique with a hierarchical scheme. Results The com- parison of the reconstructions from three operators on 16 cases of biplanar radiographs of the thoracic spine showed that precision was within 2.5 mm for location and 2.6~ for orientation, and accuracy was within 3.1 mm for location and 3.0~ for orientation. The user interaction time for reconstruction was within 3 min. The average computation time by computer was 115 rain. Conclusion The vertebral orientations can be estimated by contour matching using the Hough transform. The proposed reconstruction method has believable repeat- ability and accuracy and may help orthopedic surgeons diagnose the 3D spine deformity from radiographs.