能谱迭代算法优化光谱CT胸部虚拟平扫图像质量的应用价值

Application value of spectral iteration algorithm to optimize image quality of chest virtual non-contrast in spectral CT

目的:比较能谱迭代算法不同迭代等级CT虚拟平扫(VNC)图像质量与iDose4算法常规平扫(TNC)图像质量的差异,探讨能谱迭代算法优化胸部VNC图像质量的应用价值。方法:前瞻性收集采用双层探测器光谱CT行胸部平扫联合增强扫描的患者54例。TNC使用iDose4算法4级重建图像,将动、静脉期增强扫描数据分别使用能谱迭代算法0~6级进行重建,得到14组VNC图像。测量VNC和TNC图像的主动脉、竖脊肌内的CT值和噪声(SD值),并计算SNR和CNR,同时由2位放射科医师对VNC图像质量进行评分。采用单因素方差分析及Kruskal-Wallis检验分析TNC与VNC图像的客观评价指标,使用Bland-Altman散点图分析2种图像各组织CT值的一致性,采用Spearman相关分析评价图像SNR、CNR与能谱迭代等级的相关性。采用Wilcoxon检验比较VNC图像质量的主观评分。结果:与TNC的CT值相比,动脉期VNC(主动脉)差异有统计学意义(P<0.05),动脉期VNC(竖脊肌)和静脉期VNC(主动脉、竖脊肌)差异均无统计学意义(均P>0.05)。不同迭代等级的同一期VNC图像之间,各组织的CT值差异均无统计学意义(均P>0.05)。动、静脉期能谱迭代等级为3级的VNC图像客观指标(噪声、SNR、CNR)与TNC差异均无统计学意义(均P>0.05)。Bland-Altman分析显示,动脉期VNC(主动脉)CT值在一致性界限为15 HU时界外数据点占比为3.7%,其他组织VNC与TNC之间CT值均有较好的一致性。动脉期VNC及静脉期VNC图像的SNR、CNR与能谱迭代等级均呈正相关(r=0.792,0.263,0.761,0.255;均P<0.01)。静脉期VNC能谱迭代等级3、4级图像的主观评分最高,差异无统计学意义(P>0.05)。2位医师对14组VNC图像的主观评分一致性均较高(均K≥0.898)。结论:在双层探测器光谱CT的胸部扫描中,能谱迭代等级为3级的静脉期VNC图像客观指标与TNC最接近,图像评分最高,可替代TNC进行使用。

Objective:To compare the difference of image quality between virtual non-contrast(VNC)with different iteration levels of spectral iteration algorithm and true non-contrast(TNC)with iDose4 algorithm,which to explore the application value of spectral iteration algorithm on optimizing the image quality of chest VNC.Methods:54 patients with chest plain scan combined with enhanced scan by dual-layer spectral detector CT were prospectively collected.TNC was reconstructed using iDose4 algorithm to iterate level 4.The spectral iteration algorithm was used to reconstruct 14 groups of VNC images with level 0 to level 6 from the arterial and venous enhanced scan data.The CT value and noise in the aorta and erector spinae of VNC and TNC images were measured and used to calculate SNR and CNR.VNC image quality was evaluated by two radiologists.One-way ANOVA analysis and Kruskal-Wallis test were used to analyze the objective evaluation indexes of TNC and VNC images.Bland-Altman plots were used to analyze the consistency of CT value in TNC and VNC images.Spearman correlation analysis was used to analyze the correlation between SNR,CNR and spectral iteration levels.Wilcoxon test was used to compare the subjective scores of VNC.Results:Compared with TNC,the CT value of the aorta in VNC-a had a statistical difference(P<0.05),while the CT value of the erector spinae in VNC-a and the CT values of the aorta and erector spinae in VNC-v had no statistical differences(all P>0.05).There were no statistical differences in the CT value of each tissue between the same phase VNC images of different iteration levels(all P>0.05),and no statistical differences of noise,SNR or CNR at spectral iteration level 3 in the arterial phase or venous phase between VNC and TNC(all P>0.05).Bland-Altman analysis showed that the proportion of out-of-boundary data points in the aortic CT value of VNC-a was 3.7% when the consistency limit was 15 HU,and the CT values of VNC and TNC in other tissues had a good consistency.The SNR and CNR of VNC-a and VNC-v images were positively correlated with the spectral iteration levels(r=0.792,0.263,0.761,0.255;all P<0.01).The subjective scores of VNC-v images of spectral iteration level 3 and level 4 were the highest,with no statistical difference(P>0.05).The consistencies of subjective scores of 14 groups of VNC images were high(all K≥0.898).Conclusions:In the chest scan of the dual-layer spectral detector CT,the objective index of the VNC-v image with the spectral algorithm iteration level 3 is the closest to the TNC image,and the image score is the highest,which can replace TNC for routine use.