深度学习重建算法联合低剂量增强CT对肝脏低对比度病灶显示的影响

Effect of Deep Learning Reconstruction Algorithm Combined with Low-dose Contrast-Enhanced CT on Liver Low-Contrast Lesion

目的:通过门静脉期肝脏图像质量评价,探究不同级别深度学习(DL)重建算法在低剂量CT增强对肝脏低对比度病变显示的影响。方法:前瞻性收集2022年4月—2022年8月行全腹部增强扫描的患者62例,随机分为A组和B组:A组(n=27)为常规辐射剂量组,管电压120 kV,自动管电流(剂量调制3级),重建Karl 5级图像;B组(n=35)为低辐射剂量组:管电压120 kV,自动管电流(剂量调制2级),重建DL(1~4)4个等级图像,记为B1~B4。记录A、B组剂量长度乘积(DLP),并计算有效辐射剂量(ED)。在轴位图像上测量肝实质、门静脉、病灶以及同层面竖脊肌的CT值和SD值,计算信噪比(SNR)和对比度噪声比(CNR);统计A、B组病灶检出数量并测量病灶的最大直径。2名观察者采用5分法评估A、B组图像质量以及病灶的显示情况。结果:A、B组患者性别、年龄及身高、体重及体重指数(BMI)差异均无统计学意义;B组ED相较于A组降低了33.96%(P<0.05);A、B各组CT值均无统计学差异(P>0.05)。B组组内肝实质、门静脉SNR、CNR随着DL等级升高逐渐升高(P<0.05);A、B两组比较,仅B3组肝实质、门静脉的SD值、SNR及CNR与A组无统计学差异,仅B4组病灶CNR与A组有统计学差异(P<0.05)。2名观察者对图像主观评分一致性较好(Kappa值为0.824~0.878,P<0.05),B3组与A组主观评分无统计学差异(P>0.05),其余各组均低于A组(P<0.05)。结论:DL算法可显著减少低剂量图像噪声,保证肝脏低对比度病灶的清晰显示,DL 3为推荐的最佳重建等级。

Purpose:To explore the effect of different levels of deep learning(DL)reconstruction algorithm on liver low-contrast lesions in low-dose contrast-enhanced CT through evaluating the image quality of the liver in the portal venous phase.Methods:A total of 62 patients who underwent abdominal contrast-enhanced CT scans from April 2022 to August 2022 were prospectively collected and randomly divided into two groups:group A and group B.Group A(n=27)was the conventional radiation dose group:tube voltage 120 kV,automatic tube current modulation(level 3),and Keep artifact really low(Karl)reconstruction algorithm level 5 were used.Group B(n=35)was the lowdose group:tube voltage 120 kV,automatic tube current modulation(level 2),and DL level 1-4 were used,denoted as B1-B4.The dose length product(DLP)was recorded,and the effective dose(ED)was calculated for groups A and B.On the axial images,the CT and SD values of the liver parenchyma,portal vein,hepatic lesion,and erector spinae at the same slice were measured,and the signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)were calculated.The number of lesions detected in group A and group B was counted and the maximum diameter of lesions was measured.Two observers used a five-point scale to evaluate the image quality and the lesion visualization in groups A and B.Results:There was no statistically significant difference in gender,age,height,weight and body mass index(BMI)between group A and group B.The ED in group B decreased by 33.96%compared with that in group A(P<0.05).No significant difference was observed in the CT values among groups A and B(P>0.05).The SNR and CNR of liver parenchyma and portal vein within group B increased gradually with the increase of DL level(P<0.05).The intergroup comparison between A and B showed that only group B3 had no statistically significant difference in the SD value,SNR and CNR of liver parenchyma and portal vein with those of group A,whereas only group B4 had statistically significant differences in lesion CNR with that of group A(P<0.05).The two observers had good agreement regarding the subjective scores(Kappa=0.824-0.878,P<0.05).There was no statistically significant difference in the subjective scores between group B3 and group A(P>0.05),while the scores of other groups were lower than those of group A(P<0.05).Conclusion:The DL reconstruction algorithm can significantly reduce image noise in low-dose CT,and ensure the clear visualization of low-contrast hepatic lesions.Among the DL levels,DL3 is the most recommended reconstruction strength.