青少年脑MRI多尺度结构影像组学数据校准方法研究

Normalizing radiomics features from multiscale structural MRI of the adolescent brain

目的 探索联合N4偏置场校正、直方图匹配(histogram matching, HM)和ComBat协调的方法校准脑结构MRI影像组学特征“扫描仪效应”的价值。材料与方法 使用3台MRI扫描仪(Philip 1.5 T、Philip 3.0 T、GE 3.0 T)对健康志愿者进行脑结构MRI即三维容积T1加权成像(three-dimensional T1 weighted image, 3D-T1WI)、扩散张量成像(diffusion tensor imaging, DTI)检查,分别采用Computational Anatomy toolbox (CAT 12)工具包和FMRIB’s Software Library(FSL)软件进行预处理。对预处理后的T1WI和DTI影像依次进行N4偏置场校正和HM归一化。最后,利用LIFEx软件提取脑灰质和白质的影像组学特征并进行ComBat协调。利用Shapiro-Wilk检验进行影像特征测量值的正态性检验,采用方差分析及Tukey-HSD检验比较3台MRI扫描仪影像特征测量值有无差异,并进一步行Bartlett球形检验估计扫描仪之间的方差是否均匀。评估各校准环节影像组学特征数目和数值统计分布的设备间差异。结果 共纳入23例健康志愿者,其中男10例,女13例,男女性之间年龄(t=1.090,P=0.316)、受教育年限(t=-0.638,P=0.574)、流调用抑郁自评量表(Center for Epidemiological Survey, Depression Scale, CES-D)评分(t=-0.670,P=0.510)差异均无统计学意义(P>0.05)。3台MRI设备采集的原始图像中,强度直方图分布范围和峰值均未对齐。当使用5级(50次迭代)、全掩模进行N4偏置场校正时,3台设备之间脑组织区域的强度变异系数最低。N4偏置场校正锐化了强度峰值,HM归一化对齐了各强度峰,ComBat协调进一步对齐3台设备的图像强度分布范围及峰值。图像校准流程(N4偏置场校正、HM归一化与ComBat协调)对于3D-T1WI和DTI两个序列的影响具有相同的趋势,3台设备之间具有差异的影像组学特征数目百分比从N4偏置场校正前的88.6%(70/79)降低到ComBat协调后的3.8%(3/79),具有差异的特征数量逐渐减少。通过N4偏置场校正、HM归一化与ComBat协调相结合,脑灰质和白质感兴趣体积间有差异的影像组学特征百分比从N4偏置场校正前的43.0%(34/79)提高到ComBat协调后的84.8%(67/79)。结论 基于N4偏置场校正、HM归一化和ComBat协调的联合校准方法能够有效减少脑MRI结构图像及影像组学特征的“扫描仪效应”并保留特征的生物性差异,有助于跨扫描仪汇集多中心MRI影像学数据。

Objectives:To explore the value of combination of N4 bias field correction,histogram-matching(HM)normalization and ComBat harmonization to reduce the"scanner effect"of radiomics features from brain MRI.Materials and Methods:Three-dimensional T1 weighted image(3D-T1WI)and diffusion tensor imaging(DTI)of the brain was performed in 23 healthy volunteers with three MRI scanners(Philip 1.5 T,Philip 3.0 T,GE 3.0 T).Computational Anatomy toolbox(Cat 12)and FMRIB's software library(FSL)were used for preprocessing.Then,N4 bias field correction and HM normalization were performed on the preprocessed T1WI and DTI.Finally,LIFEx software was used to extract radiomics features of gray and white matter and then Combat harmonization were carried out.The Shapiro-Wilk test was used to exam the normality and the analysis of variance(ANOVA)and Tukey honestly significant difference(Tukey-HSD)test were used to compare the radiomics features of the three scanners,and the Bartlett spherical test was used to estimate whether the variance was uniform.The differences between scanners in the number of radiomics features and numerical statistical distribution in each processing were qualitatively and quantitatively evaluated.Results:A total of 10 males and 13 females were enrolled.There was no significant difference in age(t=1.090,P=0.316),education years(t=-0.638,P=0.574)and CES-D score(t=-0.670,P=0.510)between the males and females(P>0.05).In the original images acquired by the three MRI scanners,the distribution range and peak value of the intensity histogram were not aligned.When N4 bias field correction was performed using 5-level(50 iterations)full mask,the intensity variation coefficient of brain tissue among the three scanners was the lowest.N4 correction sharpened the intensity peak,HM normalized and aligned each intensity peak,and Combat harmonization further aligned the image intensity distribution range and peak of the three MRI.The process(N4 bias field correction,HM normalization and ComBat harmonization)had the same influence on T1WI and DTI sequences.Through the combination of N4 correction,HM normalization and Combat harmonization,the percentage of radiomics features with differences between scanners was reduced from 88.6%(70/79)before bias field correction to 3.8%(3/79)after ComBat harmonization.At the same time,the percentage of radiomics features with differences between VOI of gray and white matter increased from 43.0%(34/79)before bias field correction to 84.8%(67/79)after ComBat harmonization.Conclusions:The combination of N4 bias field correction,HM normalization and ComBat harmonization can effectively eliminate the"scanner effect"of the brain structural MRI and thereby help to incorporate multi-center MRI data across scanners.