基于深度学习重建的卵巢高分辨单次激发快速自旋回波MRI在卵泡计数中的应用研究  

Application value of high-resolution single-shot fast spin-echo ovarian MRI based on deep learning reconstruction in follicle counting

作　　者：杨仁杰[1] 邹宇洁[2] 刘薇音 刘昌盛[1] 文之 李亮[1] 查云飞[1] YANG Renjie;ZOU Yujie;LIU Weiyin;LIU Changsheng;WEN Zhi;LI Liang;ZHA Yunfei(Department of Radiology,Renmin Hospital of Wuhan University,Wuhan 430060,China;Reproductive Medicine Center,Renmin Hospital of Wuhan University,Wuhan 430060,China;GE Healthcare(China)Co.,Ltd.,Beijing 100176,China)

机构地区：[1]武汉大学人民医院放射科,武汉430060 [2]武汉大学人民医院生殖医学中心,武汉430060 [3]通用电气医疗(中国)有限公司,北京100176

出　　处：《磁共振成像》2024年第10期50-55,61,共7页Chinese Journal of Magnetic Resonance Imaging

基　　金：国家自然科学基金(编号:81601461)。

摘　　要：目的与经阴道超声(transvaginal ultrasonography,TVUS)以及常规算法重建(conventional reconstruction,CR)的高分辨单次激发快速自旋回波(single-shot fast spin-echo,SSFSE)和螺旋桨(periodically rotated overlapping parallel lines with enhanced reconstruction,PROPELLER)T2WI图像进行比较,评估深度学习重建(deep learning reconstruction,DLR)的高分辨SSFSE图像在卵泡计数中的应用价值。材料与方法前瞻性招募临床确诊或疑似多囊卵巢综合征的患者进行卵巢MRI检查,无性生活史的患者同时接受TVUS检查。MRI扫描序列包括三平面高分辨PROPELLER和SSFSE序列,SSFSE序列同时采用DLR和CR,分别生成SSFSE-DLR和SSFSE-CR图像。SSFSE-DLR、SSFSE-CR以及PROPELLER图像间定性指标(模糊伪影、主观噪声、卵泡清晰性)的比较采用Wilcoxon秩和检验。采用组内相关系数(intra-class correlation coefficient,ICC)和Bland-Altman法进行卵泡计数的重复性评估,同时采用配对t检验比较SSFSE-DLR的观察者内和观察者间差异的绝对值与SSFSE-CR和PROPELLER之间的显著性。基于SSFSE-DLR和TVUS的卵泡计数间比较采用配对t检验。结果24例患者接受了MRI检查,其中18例同时接受了TVUS检查。除SSFSE-DLR和PROPELLER的主观噪声评分差异无统计学意义(P>0.05)之外,观察者1对SSFSE-DLR的定性评价均优于SSFSE-CR和PROPELLER(P<0.05)。观察者2对SSFSE-DLR的定性评价均优于SSFSE-CR和PROPELLER(P<0.05)。SSFSE-DLR具有最高的ICC、最窄的95%一致性限度以及最小的观察者内和观察者间差异的绝对值(P<0.05),表现出最佳的卵泡计数可重复性。而且,SSFSE-DLR相比TVUS可以检出更多的卵泡(P<0.001)。结论深度学习重建的SSFSE图像能够显著改善卵泡显示的清晰性,从而提高卵泡计数的可重复性,可以作为识别多囊卵巢形态的更可靠的成像方法。Objective:To investigate the application value of high-resolution single-shot fast spin-echo(SSFSE)acquisition with deep learning reconstruction(DLR)in follicle counting compared to transvaginal ultrasonography(TVUS),conventional reconstruction(CR)SSFSE and periodically rotated overlapping parallel lines with enhanced reconstruction(PROPELLER)images.Materials and Methods:Participants with clinically confirmed or suspected polycystic ovary syndrome(PCOS)were prospectively recruited and underwent ovarian MRI.Those with no history of sexual activity also underwent ovarian TVUS.High-resolution PROPELLER and SSFSE T2-weighted sequences were obtained on three matched planes.The SSFSE sequences implemented both DLR and CR,generating SSFSE-DLR and SSFSE-CR images respectively.Qualitative indicators including blurring artifacts,subjective noise,and conspicuity of follicles were compared using Wilcoxon signed-rank tests.Follicle counting was performed by two observers,with repeatability assessed using intraclass correlation coefficient(ICC)and Bland-Altman method.Absolute values of intra-observer and inter-observer differences were compared using a paired t-test.Follicle count between SSFSE-DL and TVUS was also compared using a paired t-test.Results:Twenty-four participants underwent MRI,with 18 of them also undergoing TVUS.Observer 1 assigned higher subjective scores to SSFSE-DLR in comparison to SSFSE-CR and PROPELLER(P<0.05),despite the similar subjective noise observed between SSFSE-DLR and PROPELLER(P>0.05).Observer 2 also rated SSFSE-DLR higher than SSFSE-CR and PROPELLER(P<0.05).Furthermore,SSFSE-DLR demonstrated the best repeatability for follicle counting,achieving the highest ICC,narrowest 95%limits of agreement,and the lowest absolute values of intra-observer and inter-observer differences(P<0.05).Moreover,SSFSE-DL detected more follicles than TVUS(P<0.001).Conclusions:SSFSE-DLR images significantly improved the display of ovarian morphology and the repeatability of follicle counting,thereby fortifying the relia

关 键 词：多囊卵巢综合征 多囊卵巢 卵泡计数 深度学习重建 单次激发快速自旋回波 磁共振成像 

分 类 号：R445.2[医药卫生—影像医学与核医学] R711.75[医药卫生—诊断学]