Department of Radiology, Seoul National University Hospital, 101 Daehangno, Jongno-gu, Seoul, 110-744, Republic of Korea.
Department of Radiology and Institute of Radiation Medicine, Seoul National University College of Medicine, 101 Daehang-no, Chongno-gu, Seoul, 110-744, Korea.
Eur Radiol. 2019 Aug;29(8):4008-4015. doi: 10.1007/s00330-018-5829-9. Epub 2018 Nov 19.
To determine the effects of different reconstruction algorithms on histogram and texture features in different targets.
Among 3620 patients, 480 had normal liver parenchyma, 494 had focal solid liver lesions (metastases = 259; hepatocellular carcinoma = 99; hemangioma = 78; abscess = 32; and cholangiocarcinoma = 26), and 488 had renal cysts. CT images were reconstructed with filtered back-projection (FBP), hybrid iterative reconstruction (HIR), and iterative model reconstruction (IMR) algorithms. Computerized histogram and texture analyses were performed by extracting 11 features.
Different reconstruction algorithms had distinct, significant effects. IMR had a greater effect than HIR. For instance, IMR had a significant effect on five features of liver parenchyma, nine features of focal liver lesions, and four features of renal cysts on portal-phase scans and four, eight, and four features, respectively, on precontrast scans (p < 0.05). Meanwhile, different algorithms had a greater effect on focal liver lesions (six in HIR and nine in IMR on portal-phase, three in HIR, and eight in IMR on precontrast scans) than on liver parenchyma or cysts. The mean attenuation and standard deviation were not affected by the reconstruction algorithm (p > .05). Most parameters showed good or excellent intra- and interobserver agreement, with intraclass correlation coefficients ranging from 0.634 to 0.972.
Different reconstruction algorithms affect histogram and texture features. Reconstruction algorithms showed stronger effects in focal liver lesions than in liver parenchyma or renal cysts.
• Imaging heterogeneities influenced the quantification of image features. • Different reconstruction algorithms had a significant effect on histogram and texture features. • Solid liver lesions were more affected than liver parenchyma or cysts.
确定不同重建算法对不同目标的直方图和纹理特征的影响。
在 3620 名患者中,480 名患者的肝实质正常,494 名患者患有局灶性实性肝脏病变(转移瘤=259;肝细胞癌=99;血管瘤=78;脓肿=32;胆管癌=26),488 名患者患有肾囊肿。使用滤波反投影(FBP)、混合迭代重建(HIR)和迭代模型重建(IMR)算法对 CT 图像进行重建。通过提取 11 个特征,进行计算机化的直方图和纹理分析。
不同的重建算法具有明显的、显著的影响。IMR 的影响大于 HIR。例如,IMR 对门静脉期肝实质的五个特征、局灶性肝病变的九个特征和肾囊肿的四个特征有显著影响,而对平扫期的影响分别为四个、八个和四个特征(p<0.05)。同时,不同的算法对局灶性肝病变(门静脉期 HIR 有六个,IMR 有九个;平扫期 HIR 有三个,IMR 有八个)的影响大于对肝实质或囊肿的影响。重建算法对平均衰减和标准差没有影响(p>0.05)。大多数参数具有良好或极好的观察者内和观察者间一致性,组内相关系数范围为 0.634 至 0.972。
不同的重建算法会影响直方图和纹理特征。重建算法在局灶性肝病变中的影响大于在肝实质或肾囊肿中的影响。
成像异质性影响图像特征的量化。
不同的重建算法对直方图和纹理特征有显著影响。
实性肝脏病变比肝实质或囊肿受影响更大。