摘要
目的:评估基于深度学习重建的半傅立叶采集单次激发快速自旋回波(HASTEDL)序列在肝脏检查中的应用价值。方法:使用3.0T MR对36例患者(男28例,女8例)行上腹部MRI扫描,扫描序列包括单次屏气HASTEDL和多次屏气刀锋伪影校正(BLADE)序列。由两位医师分别对肝脏成像质量(锐利度和伪影)进行五分制主观评分。分别在肝门水平肝脏的肝右叶和肝左叶、病灶显示最大层面及其相应层面同一相位方向的右侧背景区放置感兴趣区,测量两组图像上肝脏的信号强度(SI)及其标准差(SD,作为背景噪声),计算图像的信噪比(SNR)和对比噪声比(CNR)。测量病灶最大径(Dmax),观察和记录图像错层率及病灶检出率。对两组图像上肝右叶和肝左叶内肝实质的SNR、CNR,病灶的SNR、CNR、D值,以及图像错层率、图像质量评分结果分别使用Wilcoxon检验或卡方检验进行组间比较。结果:两位医师对两组图像(HASTEDL和BLADE序列)的主观评分和客观测量数据的一致性均为良好(Kappa和ICC值均大于0.75)。两组之间图像锐利度主观评分的差异无统计学意义(4.62±0.55 vs. 4.27±0.65,P=0.289),HASTE-DL组图像伪影的主观评分显著高于BLADE组(4.78±0.48 vs. 4.14±0.98,P<0.001)。HASTEDL组肝左叶和肝右叶内肝组织的SNR、病灶的SNR和CNR均显著高于BLADE组(P<0.001)。两组之间病灶Dmax的差异无统计学意义(P=0.978)。BLADE组的图像错层率明显高于HASTEDL组(P=0.014)。两组中病灶检出率均为100%。结论:基于深度学习重建的单次屏气HASTE序列能有效提高肝脏T2WI图像质量而不会遗漏病灶,并可显著缩短扫描时间,优化肝脏扫描效率,有较好的临床应用前景。
Objective:To evaluate the clinical value of deep learning reconstruction combined with accelerated half-fourier acquisition single-shot turbo-spin-echo(HASTE DL)sequence in liver imaging.Methods:36 patients(28 males and 8 females)were scanned on the upper abdomen by single breath-holding HASTE DL and multiple breath-holding BLADE sequences using a 3.0T MR scanner.The liver image quality(sharpness and artifacts)was scored on a 5-point scale by two observers,and the areas of interest were delimited in the anterior and posterior segments of the right lobe of the liver,the inner and outer segments of the left lobe of the liver,the focus and the right background area in the same phase direction of the corresponding plane of the liver at the level of the liver hilum.The signal intensity(SI)of the liver and erector spine muscle in the two groups of images were measured,standard deviation(SD)of the background noise,and signal to noise ratio(SNR)and contrast to noise ratio(CNR)were calculated,respectively.The maximum diameter(D max)of the lesion was measured.The staggered rate of the image was observed.Wilcoxon test and Kappa test were used to analyze the SNR,CNR,D,splitter rate and image quality scores of the two groups.Results:Both the objective measurement and the subjective score of the two groups of images were in good agreement between the two observers(Kappa and ICC values were greater than 0.75).There was no significant difference in subjective sharpness score between BLADE group and HASTE DL group(4.62±0.55 vs.4.27±0.65,P=0.289).The subjective of HASTE DL group was significantly higher than that of BLADE group(4.78±0.48 vs.4.14±0.98,P<0.001).The SNRs of the hepatic tissue in left lobe and right lobe,the SNR and CNR of the focal lesion in HASTE DL group were significantly higher than those in BLADE group(P<0.001).There was no significant difference in the D max of the lesion between the two groups(P=0.978).The ration of slice discontinuity with HASTE DL was significantly lower than that with BLADE group(P=0.014).Conclusion:The single breath holding HASTE sequence based on deep learning reconstruction can effectively improve the quality of liver T 2WI images without mis-sing lesions.It can significantly shorten the scanning time and optimize the efficiency of liver scanning,and has a good clinical application prospect.
作者
张楠
刘锴
傅彩霞
陈财忠
孙海涛
曾蒙苏
ZHANG Nan;LIU Kai;FU Cai-xia(Department ofRadiology,Zhongshan Hospital of Fudan University,Shanghai Institute of ImagingMedicine,Shanghai 200032,China)
出处
《放射学实践》
CSCD
北大核心
2023年第12期1611-1616,共6页
Radiologic Practice
