协调多CT设备间肝脏影像组学特征的一致性

Harmonizing the consistency of liver radiomic features across multiple CT

目的评估健康肝脏影像组学特征在多种CT扫描仪中的一致性,并探究基于图像和基于特征的协调方法对结果的影响。方法回顾性收集2015年1月1日~2023年1月1日赣州市人民医院4台CT设备的243例健康成年人的腹部CT检查结果(Optima CT660:n=83,Revolution 512 CT:n=56,Emotion 16:n=69,Definition AS+CT:n=35)。对每位患者在肝门层面选择肝实质勾画直径为30 mm的三维感兴趣区域,使用Pyradiomics提取93个影像组学特征。采用均值中心化、Z-Score、重采样、直方图匹配和ComBat方法协调设备间的差异。使用Mann-Whitney U检验比较两台扫描仪之间以及应用协调方法前后特征的一致性,并计算Cohen's d值,以评估不同方法的影响大小。结果肝脏影像组学特征总体一致性为55.38%,其中在相同制造商设备中为87.10%,在不同制造商设备为39.52%;经均值中心化、Z-Score、重采样、直方图匹配和ComBat方法协调后,总体保持一致的特征比例分别为:44.82%、68.82%、66.49%、76.52%和100%;d值分别为-0.57、0.62、0.57、0.78、1.59。结论肝脏CT影像组学特征在不同设备之间的一致性较差,仅在同一制造商的扫描仪中保持良好。基于图像和特征的协调方法能有效减少由不同CT制造商和设备型号引起的特征变异;其中,重采样和直方图匹配依赖于具体的参数设置和参考图像的选取;基于特征的ComBat协调表现最佳,使得所有特征在不同设备之间均能保持一致,对未来跨设备或跨中心的研究具有积极影响。

Objective To evaluate the consistency of radiomic features of the healthy liver across multiple CT scanners and to investigate the impact of image-based and feature-based harmonization methods on the results.Methods Abdominal CT examinations of 243 healthy adults(Optima CT660:n=83,Revolution 512 CT:n=56,Emotion 16:n=69,Definition AS+CT:n=35)were retrospectively collected from January 1,2015 to January 1,2023 at four CT scanners of Ganzhou People's Hospital.For each patient,a 30 mm diameter three-dimensional region of interest was delineated in the liver parenchyma at the portal vein level,and 93 radiomic features were extracted using Pyradiomics.Mean centering,Z-Score,resampling,histogram matching,and ComBat methods were used to harmonize inter-device differences.The Mann-Whitney U test was used to compare the consistency of features between two different scanners before and after applying harmonization methods,and Cohen's d value was calculated to assess the effect size of different methods.Results The overall consistency of liver radiomic features was 55.38%,with 87.10% consistency among devices from the same manufacturer and 39.52% among devices from different manufacturers.After harmonization with mean centering,Z-Score,resampling,histogram matching,and ComBat methods,the proportions of consistent features were 44.82%,68.82%,66.49%,76.52%,100%,respectively;the d values were-0.57,0.62,0.57,0.78,1.59,respectively.Conclusion The consistency of liver CT radiomic features is poor between different devices,maintaining good consistency only among scanners from the same manufacturer.Image-based and feature-based harmonization methods can effectively reduce feature variation caused by different CT manufacturers and device models.Among these methods,resampling and histogram matching depend on specific parameter settings and the selection of reference images.The feature-based ComBat harmonization performs best,ensuring that all features remain consistent across different devices,which positively impacts future cross-device or cross-center studies.