CT采集和重建参数对肺结节影像组学特征稳定性的影响

目的：探讨CT重建算法、辐射剂量和对比剂3种参数对肺结节影像组学特征稳定性的影响。

方法：前瞻性选取96例患者共计150枚肺结节，对同一例患者的肺结节的原始图像重建后共获得6种不同参数的序列，使用ITK-SNAP软件分别对不同序列的肺结节进行分割，将所有扫描数据通过美国GE公司的A.K.（Analysis Kit）软件进行肺结节影像组学特征提取，通过组内相关系数选取复测信度好的影像组学特征，使用R语言统计学软件对特征参数进行分析，将不同序列的所有特征值配对比较，分别统计改变采集和重建参数导致变化的影像组学特征，比较不同采集和重建参数对肺结节影像组学特征稳定性的影响。

结果：150例肺结节共提取391个影像组学特征，选取经过复测组内相关系数大于0.75的320个特征纳入进一步分析。分别改变重建算法、辐射剂量和对比剂，发生变化的影像组学特征范围分别为25.6%~38.4%、47.2%~63.1%和46.9%~54.1%。对比剂使肺结节影像组学特征发生变化的数目略多于辐射剂量（P=0.08）；辐射剂量和对比剂使肺结节影像组学特征发生变化的数目均高于重建算法（P<0.01, P<0.01）。同时改变重建算法、辐射剂量和对比剂3种参数，66.6%（213/320）的影像组学特征发生了变化，其中包括26.7%（4/15）的形态学特征、95.0%（38/40）的直方图特征和64.5%（171/265）的纹理特征，直方图特征和纹理特征发生变化的百分比均高于形态学特征（P<0.01, P<0.01）。当辐射剂量与对比剂均改变时，FBP算法获得的肺结节影像组学特征发生变化的数目略少于50% ASiR-V算法（P=0.67）。

结论：CT重建算法、辐射剂量和对比剂会影响肺结节的CT影像组学特征，辐射剂量和对比剂对影像组学特征的稳定性影响大于重建算法。直方图特征和纹理特征较形态学特征更容易受到CT采集和重建算法的影响。当辐射剂量和对比剂参数同时改变时，FBP重建算法获得的影像组学特征的稳定性略优于50% ASiR-V算法。在影像组学分析中应充分考虑到这些参数的影响。

主题词：肺结节；影像组学；稳定性；计算机断层扫描；扫描参数

第二部分

不同密度类型肺结节CT影像组学特征稳定性的比较

目的：比较纯磨玻璃密度、部分实性和实性肺结节之间CT影像组学特征稳定性的差异。

方法：前瞻性选取纯磨玻璃密度肺结节、部分实性肺结节和实性肺结节各50枚，对同一例患者肺结节的原始图像重建后共获得6种不同参数的序列，使用ITK-SNAP软件分别对不同序列的三种密度肺结节进行分割，将所有扫描数据通过美国GE公司的A.K.（Analysis Kit）软件进行肺结节影像组学特征提取，通过组内相关系数选取复测信度好的影像组学特征。使用R语言统计学软件对特征参数进行分析，将不同序列的所有特征值配对比较，分别统计改变采集和重建参数导致不同密度肺结节变化的影像组学特征的数目，比较不同密度肺结节间影像组学特征稳定性的差异。

结果：（1）改变CT采集和重建参数对不同密度肺结节影像组学特征稳定性的影响 ①对于纯磨玻璃密度肺结节，分别改变重建算法、辐射剂量和对比剂使影像组学特征发生变化的范围为10.6%~40.3%、30.9%~46.3%和50.6%~64.1%。重建算法使纯磨玻璃密度肺结节发生变化的特征少于辐射剂量（P<0.01）；辐射剂量使纯磨玻璃密度肺结节发生变化的特征少于对比剂（P<0.01）；重建算法使纯磨玻璃密度肺结节发生变化的特征少于对比剂（P<0.01）。②对于部分实性肺结节，分别改变重建算法、辐射剂量和对比剂使影像组学特征发生变化的范围为10.0%~14.1%、8.4%~23.8%和13.8%~17.5%。重建算法使部分实性结节发生变化的特征少于辐射剂量（P<0.01）；对比剂使部分实性结节发生变化的特征少于辐射剂量（P<0.01）；重建算法使部分实性结节发生变化的特征略多于对比剂（P=0.91）。③对于实性肺结节，分别改变重建算法、辐射剂量和对比剂影像组学特征发生变化的范围分别为34.7%~47.7%、20.0%~42.2%和41.3%~55.9%。重建算法使实性肺结节发生变化的特征多于辐射剂量（P<0.01）；对比剂使实性肺结节发生变化的特征多于辐射剂量（P<0.01）；重建算法使实性肺结节发生变化的特征略少于对比剂（P=0.21）。（2）不同密度肺结节间影像组学特征稳定性的差异 改变重建算法、辐射剂量和对比剂参数的12组序列比较中有8组序列比较得到相同的结果，即在改变相同参数的情况下，纯磨玻璃密度肺结节和实性肺结节的影像组学特征稳定性相差不大（P>0.05），部分实性肺结节影像组学特征的稳定性优于纯磨玻璃密度肺结节和实性肺结节（P<0.05）。（3）改变参数后三种不同密度肺结节发生变化的影像组学特征类型比较 对结果一致的8组序列比较进行亚组分析，三种不同密度肺结节间稳定性差异性的主要表现为纹理特征的差异，且在参数变化时部分实性肺结节发生改变的纹理特征少于纯磨玻璃密度肺结节和实性肺结节。

结论：重建算法、辐射剂量和对比剂，对不同密度肺结节影像组学特征产生影响不尽相同。对于纯磨玻璃密度肺结节，对比剂对其影像组学特征特征稳定性的影响最大，其次是辐射剂量，重建算法对其影像组学特征特征稳定性的影响最小；对于部分实性肺结节，辐射剂量对其影像组学特征稳定性的影响大于重建算法和对比剂；对于实性肺结节，重建算法和对比剂对影像组学特征稳定性的影响大于辐射剂量。在改变相同参数的情况下，纯磨玻璃密度肺结节和实性肺结节之间影像组学特征稳定性基本相同，而部分实性肺结节的影像组学特征稳定性要优于纯磨玻璃密度肺结节和实性肺结节，且三种不同密度肺结节间稳定性差异主要体现在纹理特征，提示未来在对亚实性肺结节的影像组学研究中，宜将纯磨玻璃密度肺结节和部分实性肺结节分开进行研究。

关键词：纯磨玻璃密度肺结节；部分实性肺结节；实性肺结节；影像组学；稳定性; 计算机断层扫描

PART Ⅰ

Influence of CT acquisition and reconstruction parameters on the stability of radiomic features in pulmonary nodules

Objective To investigate the influence of CT reconstruction algorithm, radiation dose, and contrast agent on the stability of radiomic features in pulmonary nodules.

Methods A total of 150 pulmonary nodules were prospectively selected from 96 patients. Six sequences with different parameters were obtained from the original image of the same patient. ITK-SNAP software was used to segment pulmonary nodules in different sequences. The radiomic features with good retest reliability were selected by intraclass correlation coefficient. The statistical software of R language was used to analyze the characteristic parameters. All the characteristic values of different sequences were matched and compared. The number of radiomic features changed by acquisition and reconstruction parameters was counted, respectively. The effects of varying acquisition and reconstruction parameters on the stability of radiomic features extracted from pulmonary nodules were compared.

Results A total of 391 radiomic features were extracted from 150 cases of pulmonary nodules, and 320 features with an intraclass correlation coefficient greater than 0.75 were selected for further analysis. When the CT reconstruction algorithm, radiation dose, and contrast agent were changed, the range of changed features was 25.6%-38.4%, 47.2%-63.1%, and 45.9%-54.1%, respectively. The number of changed radiomic features of pulmonary nodules caused by contrast agent was slightly more than that caused by radiation dose (P=0.08). The number of changed radiomic features of pulmonary nodules caused by radiation dose and contrast agent was higher than that caused by the reconstruction algorithm (P<0.01, P<0.01).Changing the three parameters of CT reconstruction algorithm, radiation dose, and contrast agent simultaneously made the change of radiomic features reach 66.6% (213/320). It included 26.7% (4/15) morphological features, 95.0% (38/40) histogram features and 64.5% (171/265) texture features. The percentage of changing histogram features and changing texture features were higher than morphological features (P<0.01, P<0.01). When the radiation dose and contrast agent were changed, the number of changed radiomic features of pulmonary nodules obtained by the FBP algorithm was slightly less than 50% ASiR-V (P=0.67).

Conclusions CT reconstruction algorithm, radiation dose, and contrast agent would affect the radiomic features extracted from pulmonary nodules. The influence of radiation dose and contrast agent on the stability of radiomic features is more significant than that of the reconstruction algorithm. Histogram and texture features are more easily affected by CT acquisition and reconstruction parameters than morphological features. When the radiation dose and contrast agent were changed simultaneously, the stability of the radiomic features obtained by the FBP reconstruction algorithm is slightly better than that obtained by the 50% ASiR-V algorithm. The influence of these parameters should be fully considered in the radiomics analysis.

[Subject words] Pulmonary nodules; Radiomics; Stability; Computed tomography; Acquisition and Reconstruction Parameters

PART Ⅱ

Stability comparison of CT radiomic features in different consistency pulmonary nodules

Objective To compare the stability of radiomic features among pure ground-glass density, part-solid, and solid pulmonary nodules.

Methods 50 pure ground-glass pulmonary nodules, 50 part-solid pulmonary nodules, and 50 solid pulmonary nodules were prospectively selected. The original images of pulmonary nodules in the same patient were reconstructed, and six sequences with different parameters were obtained. The ITK-SNAP software was used to segment the pulmonary nodules with three different densities in different sequences. All the scanning data were extracted by A.K. (Analysis Kit) software of GE company. The radiomic features with good retest reliability were selected by the intraclass correlation coefficient. R language statistical software was used to analyze the characteristic parameters. All the feature values of different sequences were paired and compared. The number of radiomic features changed by the acquisition and reconstruction parameters was counted. The stability of radiomic features of three kinds of density pulmonary nodules was compared when CT acquisition and reconstruction parameters changed.

Results (1) Influence of changing CT acquisition and reconstruction parameters on the stability of radiomic features extracted from different consistency pulmonary nodules ① For pure ground-glass pulmonary nodules, changing the reconstruction algorithm, radiation dose, and contrast agent parameters could influence the radiomic features in the range of 10.6%-40.3%, 30.9%-46.3%, and 50.6%-64.1%, respectively. The features of pure ground-glass pulmonary nodules caused by the reconstruction algorithm were less than those caused by radiation dose (P<0.01). The features of pure ground-glass pulmonary nodules caused by radiation dose were less than those caused by contrast agent (P<0.01). The features of pure ground-glass density pulmonary nodules caused by the reconstruction algorithm were less than those caused by the contrast agent (P<0.01). ② For part-solid pulmonary nodules, changing the reconstruction algorithm, radiation dose, and contrast agent parameters could influence the radiomic features in the range of 10.0%-14.1%, 8.4%-23.8%, and 13.8%-17.5%, respectively. The features of part-solid pulmonary nodules caused by the reconstruction algorithm were less than those caused by radiation dose (P<0.01). The features of part-solid pulmonary nodules caused by contrast agent were less than those caused by radiation dose (P<0.01). The features of part-solid pulmonary nodules caused by the reconstruction algorithm were slightly more than those caused by the contrast agent (P=0.91). ③ For solid pulmonary nodules, changing the reconstruction algorithm, radiation dose, and contrast agent parameters could influence the radiomic features in the range of 34.7%-47.7%, 20.0%- 42.2%, and 41.3%- 55.9%, respectively. The features of solid pulmonary nodules caused by the reconstruction algorithm were more than those caused by radiation dose (P<0.01). The features of solid pulmonary nodules caused by contrast agent were more than those caused by radiation dose (P<0.01). The features of solid pulmonary nodules caused by the reconstruction algorithm were slightly less than those caused by contrast agent (P=0.21). (2) The difference of stability of radiomic features between different consistency pulmonary nodules After sorting out the results of 12 sequences comparison, 8 sequences comparison got consistent results. Those were, the changed radiomic feature extracted from pure ground-glass pulmonary nodules and solid pulmonary nodules had no significant difference (P>0.05). The changed radiomic feature extracted from part-solid pulmonary nodules was lower than that of pure ground-glass pulmonary nodules and solid pulmonary nodules (P<0.05). (3) Comparing radiomic features of three different consistency pulmonary nodules after changing parameters After subgroup analysis of 8 sequences comparison with consistent results, the main difference of stability among the three pulmonary nodules with different densities was texture features. When the parameters changed, the number of texture features of part-solid pulmonary nodules changed was lower than that of pure ground-glass pulmonary nodules and solid pulmonary nodules.

Conclusion Reconstruction algorithm, radiation dose, and contrast agent have different effects on radiomic features of different consistency pulmonary nodules. For pure ground-glass pulmonary nodules, contrast agent has the most significant impact on the stability of radiomic features, followed by radiation dose. The reconstruction algorithm has the least impact on the stability of radiomic features. For part-solid pulmonary nodules, the effect of radiation dose on the stability of radiomic features is greater than that of the reconstruction algorithm and contrast medium. For solid pulmonary nodules, the influence of reconstruction algorithm and contrast agent on the stability of radiomic features is more significant than that of radiation dose. In changing the same parameters, the stability of the radiomic features of pure-ground glass density pulmonary nodules and solid pulmonary nodules is similar. The stability of the radiomic features of part-solid pulmonary nodules is better than that of pure ground-glass density pulmonary nodules and solid pulmonary nodules. Moreover, the difference in stability among the three different pulmonary nodules is mainly reflected in the texture features. In the follow-up radiomic studies of subsolid pulmonary nodules, pure ground-glass density pulmonary nodules and part-solid pulmonary nodules should be studied separately.

Subject words Pure ground-glass density pulmonary nodules; part-solid pulmonary nodules; solid pulmonary nodules; radiomics; Stability; Computed tomography