| Objective: Primary hepatocellular carcinoma(HCC)is the main malignant tumor of the liver,of which more than 90% of HCC patients are accompanied by different degrees of cirrhosis,and 1/3 of patients with cirrhosis will develop into HCC.The liver cirrhosis is developed after a long period of chronic liver disease,at which point the risk of liver cancer is very low.Accurate HCC differential diagnosis and evaluation of liver cirrhosis degree,grading,is expected to retain the remaining hepatocyte active agent and liver biological function.With the development of computer technology,MR diffusion weighted imaging(DWI),as an attractive imaging technique,has been widely used in the clinical diagnosis of abdominal diseases.In order to maximize the use of MR on the diagnosis of liver disease and to carry out high quality and effective examination,imaging genomics based on MR-DWI has been proposed in recent years,and reliable and meaningful radiomics features are very important for the identification of tumor phenotypes.The purpose of this study was to study the effects of diffusion-sensitive factor(b value)on the radiomics features extracted from patients in normal control group,cirrhosis and HCC group.Methods: In this study,MR scans were performed for 3 different groups of patients with cirrhosis,HCC and healthy volunteers.And different sequence images were obtained,including T1 weighting(gradient echo T1 weighting,GRE T1),T2 weighted(fast spin echo T2 weighting,FSE T2),multi-stage dynamic enhancement sequence(multi phase contrast-enhanced MR),diffusion weighted imaging(DWI).Among them,the B values in DWI are set to 0,20,50,100,200,400,800,1000,1200,1500s/mm2.The 74 extracted radiomics features were evaluated by 3D Slicer Radiomics Software.For the normal control group and the cirrhosis group,we defined three VOIs in the two consecutive layers in the DWI with b=0s/mm2 with their diameter 2cm and height 14 mm in different liver segments,avoiding the inferior vena cava,portal vein and portal vein area.For the HCC group,we sketched the lesion contour manually.The percentage variation coefficient(percentage coeffcient of variation,%cov),fitting coefficient,and Spearman correlation coefficient were used to evaluate the stability of the texture characteristics of imaging groups between different groups with the change of B value.The concordance correlation coefficient(CCC)was used to evaluate the repeatability of radiomics-based features between similar B values,with %COV < 30 indicating small variability,30?%COV?50 indicating moderate one and 50?%COV indicating large one.Using exponential and polynomial fitting,we investigated the changing trend of each radiomics feature with the change of the b value,with the R2 indicating good fitting.Spearman correlation coefficient(r)indicates the correlation between the b values and radiomics features,with the || < 0.2 indicating negligible correlation,0.2 ? || < 0.4 indicating weak one,0.4 ? || < 0.7 indicating moderate one,0.7 ? || <0.9 indicating strong correlation,0.9 ? || < 1 indicating very strong one.Results: This study showed that in the radiomics-based texture features extracted from MR-DWI,the histogram features and texture features of the first order intensity have a strong correlation with the change of diffusion-sensitive factors.For the normal control group,the radiomics features with mild,moderate,and large variability accounted for 28%,19%,53% of the total features,respectively.For the liver cirrhosis group,the radiomics features with moderate,and large variability accounted for 49%,27%,and 24%,respectively.And for the HCC group,the radiomics features with moderate,and large variability accounted for 26%,28%,and 46%,respectively.A total of 18(24%)features in normal control group,cirrhosis group and HCC group showed lower variability,while 11(15%)of these 18 variables had statistically significant differences in between-group comparisons.Exponential and polynomial fitting revealed that there were 66(89%)and 48(65%)features with good degree of fitting(R2>0.9)in the HCC group and normal control group repsectively.And there were 48(65%)features with good degree of fitting in both HCC and normal control goup.The number of features with very strong correlation,strong correlation,moderate correlation,weak correlation and negligible in HCC group and normal control group were: 0(0%)/0(0%),0(0%)/63(85%),65(88%)/8(11%),2(3%)/3(4%)and 7(9%)/0(0%),respectively.There were 20(27%)/21(28%)? 45(61%)/50(68%)features with positive(r>0.4)and negative(r<-0.4)correlations with the increasing b values in the HCC group and normal control groups.We found that the radiomics features extracted in the b=0,20,50,100,200,400 s/mm2?b=800,1000,1200,1500 s/mm2 showed high repeatability(CCC?0.8)in the liver cirrhosis group.And the radiomics features extracted in the b=0,20,50,100,200 s/mm2?b=1 000 s/mm2 showed high repeatability in the HCC group.Finally,11 radiomics features(GLRLM-SRE and GLRLM-R.Percent,GLRLM-RLNUN,GLSZM-Z.Entropy,GLRLM-RLNU,GLCM-IMC2,GLCM-Entropy,GLRLM-R.Entropy,GLRLM-LRE,GLSZM-SZNUN,GLCM-ID))were screened and could be used in further research for HCC and liver cirrhosis.Optimized B-value combination: b=0,20,50,100,200,1000s/mm2,could provide reference for clinical diagnosis and radiomics study of MR-DWI.Conclusion and significance: The radiomics features tested by this study had large variance because of the influence of different diffusion sensitive factor B values.Most features were unstable and had a significant correlation with the change of B value in DWI.At the same time,this study showed that the radiomics features extracted from the similar B-value were reproducible. |
PDF Full Text Request