并行采集技术对颅脑MRI影像组学特征稳定性的影响  

Impact of parallel acquisition technology on the robustness of magnetic resonance imaging radiomic features

作　　者：李宁 张彪 王紫君 张娴 胡奎 LI Ning;ZHANG Biao;WANG Zi-jun(Department of Radiology,Hubei Cancer Hospital,Tongji Medical College,Huazhong University of Science and Technology,Wuhan 430079,China)

机构地区：[1]华中科技大学同济医学院附属湖北肿瘤医院放射科,武汉430079 [2]芜湖市繁昌区人民医院影像中心,安徽芜湖2412002

出　　处：《放射学实践》2023年第4期514-521,共8页Radiologic Practice

摘　　要：目的:探讨集成并行采集技术(iPAT)对颅脑磁共振影像组学特征稳定性的影响。方法:采用3.0T磁共振对自制体模和6例健康志愿者(被试组)行颅脑横轴面扫描,扫描序列包括T 1WI(S1)、T 1-FLAIR(S2)、T 2-FLAIR(S3)和T 2WI(S4),各序列分别采用不同缩减因子(R)的2种并行采集技术(GRAPPA和mSENSE)进行扫描。其中体模组4个序列使用的GRAPPA和mSENSE技术均分别采用了6种R值(R=none,2,3,4,5,6),共获得48组图像;被试组4个序列使用的GRAPPA和mSENSE技术均分别采用了3种R值(R=none,2,3),共获得24组图像。基于各组图像,提取4个体模ROI(P-ROI)或4个脑实质ROI(V-ROI)的影像组学特征。计算每个序列图像上每个P-ROI和V-ROI中提取的组学特征在6种(R=none,2,3,4,5,6)和3种缩减因子(R=none,2,3)之间的变异系数(CV=特征均值/标准差),基于CV值对特征的稳定性进行分级(分为十分稳定、稳定、一般和不稳定共4级)。首先,通过CV评价使用不同数量的加速因子对GRAPPA和mSENSE两种并行采集技术提取的组学特征稳定性的影响;然后,采用χ^(2)检验分析两种并行采集技术对不同序列和不同ROI之间组学特征稳定性影响的差异。结果:iPAT及缩减因子的不同组合对影像组学特征稳定性的影响显著。体模数据中,6种R(none,2,3,4,5,6)时,mSENSE和GRAPPA图像上提取的极稳定特征的占比均低于5%,mSENSE图像的特征稳定性较GRAPPA更差;同时包含3个R值(R=none,2,3)的mSNESE和GRAPPA共6组图像时,4个序列之间(χ^(2)=257.807,P<0.001)及4个P-ROI之间(χ^(2)=87.379,P<0.001)特征稳定性的差异均有统计学意义,FLAIR序列(S2和S3组)能提供更多极稳定的特征;与3组图像(R=none,2,3)相比,6组图像(R=none,2,3,4,5,6)时仅在4个P-ROI之间观察到特征稳定性的差异有统计学意义(χ^(2)=25.115,P<0.05)。在被试组中(R=none、2、3),仅在4个V-ROI之间发现特征稳定性的差异有统计学意义(χ^(2)=46.642,P<0.001)。结论:集成并行采集�Objective:To explore the impact of integrated parallel acquisition technique(iPAT)on the robustness of radiomics features extracted from brain MR images.Methods:A phantom and 6 healthy volunteers(HC group)were scanned at a 3.0T MR system using four sequences of T 1WI(S1),T 1-FLAIR(S2),T 2-FLAIR(S3)and T 2WI(S4),combined with one of the two types of iPAT:GRAPPA(generalized autocalibration partially parallel acquisitions)or mSENSE(modified sensitivity encoding).For phantom,four sequences(S1~S4)using GRAPPA or mSENSE with six reduction factors(R=none,2,3,4,5,6)were respectively scanned,thus 48 groups of images were obtained.For volunteers,four sequences(S1~S4)using GRAPPA or mSENSE with three R values(none,2,3)were respectively scanned,and 24 groups of images were obtained.The radiomics features of the four ROIs in the phantom(P-ROIs)or four ROIs in brain of each volunteer(V-ROIs)were extracted based on the images of each group.The coefficient of variation(CV=mean/SD)of each radiomics feature extracted from each P-ROI or V-ROI on the images of the 4 sequences(S1~S4)among the six or three R values was calculated,and CV was used to evaluate feature robustness.Firstly,CV was used to evaluate the changes of robustness of each radiomics feature extracted from GRAPPA or mSENSE images using 6 or 3 acceleration factors;χ^(2)-test was used to compare the differences in the robustness of radiomics features affected by iPAT between different sequences or between different ROIs.Results:iPAT and its acceleration factor R had a significant impact on the robustness of radiomics features.In phantom data(R=none,2,3,4,5,6),the proportion of extremely roust features in mSENSE or GRAPPA images was even less than 5%.GRAPPA delivered a more robust substrate for radiomic analyses than mSENSE.In phantom theχ^(2)-test result indicated that using mSENSE and GRAPPA with three R values(none,2,3),the difference of radiomics features robustness among the 4 sequences,and among the 4 P-ROIs were all significant(χ^(2)=257.807,P<0.001;χ^(2)=87.3

关 键 词：并行采集 影像组学 纹理特征 稳定性 变异系数 磁共振成像 

分 类 号：R445.2[医药卫生—影像医学与核医学] R735.7[医药卫生—诊断学]