Comparing five different iterative reconstruction algorithms for computed tomography in an ROC study
Comparing five different iterative reconstruction algorithms for computed tomography in an ROC study
复制标题
DOI：
10.1007/s00330-014-3333-4
复制
发表时间：
2014-12-01
影响因子：
5.9
通讯作者：
Fosse, Erik
中科院分区：
文献类型：
--
作者：
Jensen, Kristin;Martinsen, Anne Catrine T.;Fosse, Erik
本研究的目的是评估四家CT供应商在三种不同剂量水平下使用五种不同的迭代重建技术所获得的病变显著性。比较了迭代算法和滤波反投影（FBP）算法在系统间和系统内的应用。采用四套不同供应商的CT系统对拟人化肝脏幻影进行检查。CTDIvol水平分别为5 mGy、10 mGy和15 mGy。利用FBP和迭代算法对图像进行重构。图像由四名观察员独立解译，并计算ROC曲线下面积（auc）。测量噪声和噪声对比比（CNR）。与FBP（0.70、0.86和0.93）相比，一种迭代算法在所有剂量水平下均增加了AUC(0.79、0.95和0.97)（p < 0.001和p = 0.047）。另一种算法将AUC从FBP时的0.78提高到5mgy时的0.84 （p = 0.007）。10和15 mGy的差异无统计学意义（p值：0.084-0.883）。与FBP相比，三种算法的AUC没有差异（p值：0.008-1.000）。所有算法都降低了噪声（10- 71%），提高了CNR。尽管所有算法都显示出降噪效果，但只有两种算法改进了病变检测。迭代重建算法在不同剂量水平下对病灶检测的影响不同。与滤波后的反投影相比，迭代算法提高了病灶的可检测性。三种算法没有显著提高病变的可检出性。euro cent One算法在最低剂量水平下提高了病灶的可检测性。
The purpose of this study was to evaluate lesion conspicuity achieved with five different iterative reconstruction techniques from four CT vendors at three different dose levels. Comparisons were made of iterative algorithm and filtered back projection (FBP) among and within systems.An anthropomorphic liver phantom was examined with four CT systems, each from a different vendor. CTDIvol levels of 5 mGy, 10 mGy and 15 mGy were chosen. Images were reconstructed with FBP and the iterative algorithm on the system. Images were interpreted independently by four observers, and the areas under the ROC curve (AUCs) were calculated. Noise and contrast-to-noise ratios (CNR) were measured.One iterative algorithm increased AUC (0.79, 0.95, and 0.97) compared to FBP (0.70, 0.86, and 0.93) at all dose levels (p < 0.001 and p = 0.047). Another algorithm increased AUC from 0.78 with FBP to 0.84 (p = 0.007) at 5 mGy. Differences at 10 and 15 mGy were not significant (p-values: 0.084-0.883). Three algorithms showed no difference in AUC compared to FBP (p-values: 0.008-1.000). All of the algorithms decreased noise (10-71 %) and improved CNR.Only two algorithms improved lesion detection, even though noise reduction was shown with all algorithms.aEuro cent Iterative reconstruction algorithms affected lesion detection differently at different dose levels.aEuro cent One iterative algorithm improved lesion detectability compared to filtered back projection.aEuro cent Three algorithms did not significantly improve lesion detectability.aEuro cent One algorithm improved lesion detectability at the lowest dose level.