Department of Medical Physics, University of Crete, 71003 Iraklion, Crete, Greece.
Phys Med Biol. 2010 Feb 21;55(4):1127-40. doi: 10.1088/0031-9155/55/4/015. Epub 2010 Jan 28.
The purpose of this study was to develop and evaluate a semiautomatic method for left ventricular (LV) segmentation on cine MR images and subsequent estimation of cardiac parameters. The study group comprised cardiac MR examinations of 18 consecutive patients with known or suspected coronary artery disease. The new method allowed the automatic detection of the LV endocardial and epicardial boundaries on each short-axis cine MR image using a Bayesian flooding segmentation algorithm and weighted least-squares B-splines minimization. Manual editing of the automatic contours could be performed for unsatisfactory segmentation results. The end-diastolic volume (EDV), end-systolic volume (ESV), ejection fraction (EF) and LV mass estimated by the new method were compared with the reference values obtained by manually tracing the LV cavity borders. The reproducibility of the new method was determined using data from two independent observers. The mean number of endocardial and epicardial outlines not requiring any manual adjustment was more than 80% and 76% of the total contour number per study, respectively. The mean segmentation time including the required manual corrections was 2.3 +/- 0.7 min per patient. LV volumes estimated by the semiautomatic method were significantly lower than those by manual tracing (P < 0.05), whereas no difference was found for EF and LV mass (P > 0.05). LV indices estimated by the two methods were well correlated (r 0.80). The mean difference between manual and semiautomatic method for estimating EDV, ESV, EF and LV mass was 6.1 +/- 7.2 ml, 3.0 +/- 5.2 ml, -0.6 +/- 4.3% and -6.2 +/- 12.2 g, respectively. The intraobserver and interobserver variability associated with the semiautomatic determination of LV indices was 0.5-1.2% and 0.8-3.9%, respectively. The estimation of LV parameters with the new semiautomatic segmentation method is technically feasible, highly reproducible and time effective.
本研究旨在开发和评估一种在电影磁共振图像上半自动分割左心室 (LV) 并随后估计心脏参数的方法。研究组包括 18 例连续的已知或疑似冠状动脉疾病的心脏磁共振检查。新方法允许使用贝叶斯洪水分割算法和加权最小二乘 B 样条最小化自动检测每个短轴电影磁共振图像上的 LV 心内膜和心外膜边界。对于不满意的分割结果，可以手动编辑自动轮廓。通过新方法估计的舒张末期容积 (EDV)、收缩末期容积 (ESV)、射血分数 (EF) 和 LV 质量与手动跟踪 LV 腔边界获得的参考值进行比较。通过两位独立观察者的数据确定新方法的可重复性。不需要任何手动调整的心内膜和心外膜轮廓的平均数量分别超过研究中总轮廓数量的 80%和 76%。包括所需手动校正在内的平均分割时间为每位患者 2.3 +/- 0.7 分钟。半自动方法估计的 LV 容积明显低于手动跟踪的容积 (P < 0.05)，而 EF 和 LV 质量则没有差异 (P > 0.05)。两种方法估计的 LV 指数相关性良好 (r 0.80)。手动和半自动方法估计 EDV、ESV、EF 和 LV 质量的平均差值分别为 6.1 +/- 7.2 ml、3.0 +/- 5.2 ml、-0.6 +/- 4.3%和-6.2 +/- 12.2 g。半自动法确定 LV 指数的观察者内和观察者间变异性分别为 0.5-1.2%和 0.8-3.9%。使用新的半自动分割方法估计 LV 参数在技术上是可行的，高度可重复且节省时间。
