Department of Biomedical Engineering, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, USA.
Med Phys. 2010 Feb;37(2):638-48. doi: 10.1118/1.3284976.
Quantitative measurements of wall thickness in human abdominal aortic aneurysms (AAAs) may lead to more accurate methods for the evaluation of their biomechanical environment.
The authors describe an algorithm for estimating wall thickness in AAAs based on intensity histograms and neural networks involving segmentation of contrast enhanced abdominal computed tomography images. The algorithm was applied to ten ruptured and ten unruptured AAA image data sets. Two vascular surgeons manually segmented the lumen, inner wall, and outer wall of each data set and a reference standard was defined as the average of their segmentations. Reproducibility was determined by comparing the reference standard to lumen contours generated automatically by the algorithm and a commercially available software package. Repeatability was assessed by comparing the lumen, outer wall, and inner wall contours, as well as wall thickness, made by the two surgeons using the algorithm.
There was high correspondence between automatic and manual measurements for the lumen area (r = 0.978 and r = 0.996 for ruptured and unruptured aneurysms, respectively) and between vascular surgeons (r = 0.987 and r = 0.992 for ruptured and unruptured aneurysms, respectively). The authors' automatic algorithm showed better results when compared to the reference with an average lumen error of 3.69%, which is less than half the error between the commercially available application Simpleware and the reference (7.53%). Wall thickness measurements also showed good agreement between vascular surgeons with average coefficients of variation of 10.59% (ruptured aneurysms) and 13.02% (unruptured aneurysms). Ruptured aneurysms exhibit significantly thicker walls (1.78 +/- 0.39 mm) than unruptured ones (1.48 +/- 0.22 mm), p = 0.044.
While further refinement is needed to fully automate the outer wall segmentation algorithm, these preliminary results demonstrate the method's adequate reproducibility and low interobserver variability.
对人体腹主动脉瘤（AAA）壁厚度进行定量测量，可能会为评估其生物力学环境提供更准确的方法。
作者描述了一种基于强度直方图和神经网络的算法，用于估算 AAA 中的壁厚度，该算法涉及对增强对比度的腹部 CT 图像进行分割。该算法应用于 10 个破裂和 10 个未破裂的 AAA 图像数据集。两名血管外科医生手动分割了每个数据集的管腔、内膜和外膜，并将参考标准定义为他们分割的平均值。通过将参考标准与算法自动生成的管腔轮廓和商业上可用的软件包进行比较，确定了可重复性。通过比较两名外科医生使用算法生成的管腔、外膜和内膜轮廓以及壁厚度，评估了可重复性。
对于管腔面积，自动和手动测量之间具有高度一致性（破裂和未破裂动脉瘤分别为 r = 0.978 和 r = 0.996），并且在血管外科医生之间也具有高度一致性（破裂和未破裂动脉瘤分别为 r = 0.987 和 r = 0.992）。与参考标准相比，作者的自动算法的结果更好，平均管腔误差为 3.69%，不到商用应用程序 Simpleware 与参考标准之间误差（7.53%）的一半。壁厚度测量也显示出血管外科医生之间的良好一致性，平均变异系数为 10.59%（破裂动脉瘤）和 13.02%（未破裂动脉瘤）。破裂动脉瘤的壁明显比未破裂动脉瘤厚（1.78 +/- 0.39 毫米），p = 0.044。
虽然进一步完善该算法以完全实现外膜自动分割仍有必要，但这些初步结果表明该方法具有足够的可重复性和较低的观察者间变异性。
