Differentiation Of High-grade Glioma And Solitary Brain Metastasis With Magnetic Resonance Imaging Using Histogram Analysis

Objective:To investigate the value of diffusion-weighted imaging(DWI),intravoxel incoherent motion(IVIM),and amide proton transfer-weighted(APTW)imaging in differentiating high-grade glioma(HGG)and solitary brain metastases(SBM)with the histogram feature analysis.Methods:Patients with newly diagnosed glioma(N=44,male/female: 25/19,age:20-79y)or SBM(N=21,male/female: 10/11,age:34-80y)were enrolled and conducted DWI,IVIM,and APTW imaging as well as T1 W,T2W,and enhanced T1 W imaging.The histogram features of apparent diffusion coefficient(ADC)from DWI,slow diffusion coefficient(Dslow),perfusion fraction(frac),fast diffusion coefficient(Dfast)from IVIM,and MTRasym(3.5ppm)from APTW imaging were extracted in the tumor parenchyma and compared between glioma and SBM.Parameters with significant differences were done with the logistics regression and receiver operator curves analysis to explore the optimal model.Then,magnetic resonance sequences suitable for further in-depth study were selected,and morphological features,such as location,degree of edema,and clear tumor boundaries,were added to explore a reliable model for the differentiation of HGG from SBM.Results:In studies of glioma and SBM identification,there were significant difference for ADCkurtosis(6.43±5.63 vs.4.54±3.23,p=0.02),MTRasym(3.5ppm)10(0.53±0.97 vs.1.00±0.75,p= 0.05),frac90(0.23±0.05 vs.0.29±0.06,p<0.01),fracentropy(0.56±0.34 vs.0.85±0.33,p<0.01),frackurtosis(6.76±4.08 vs.3.91±1.43,p=0.01),and fracmean(0.14±0.02 vs.0.17±0.04,p<0.01)between glioma and SBM.frackurtosis(OR=0.66,95% CI:0.48-0.92,p=0.02),and fracmean(OR=1.44,95% CI:1.16-1.18,p<0.01)were independent factors for SBM differentiation.The AUC of model combined with age,frackurtosis and fracmean was0.83(accuracy:0.82,sensitivity:0.57,specificity:0.93,PPV:0.80,NPV:0.82),while the AUC of model without age with improved sensitivity was 0.81(accuracy:0.80,sensitivity:0.62,specificity:0.89,PPV:0.72,NPV:0.83).No significant difference existed between the two models with the De Long ‘test(p=0.58).In studies using IVIM to further differentiate HGG from SBM,tumor margins in SBM were clearer compared with HGG(p= 0.017).For HGG,the kurtosis of Dfast(p=0.022)and frac(p=0.077),the kurtosis(p=0.019)and frac(p=0.025)of Dslow,and the entropy of Dslow(p=0.005)and frac(p=0.001)were significantly reduced,and the frac mean values of ETR(p=0.007)and PTEZ(p=0.017)were significantly reduced.In ROC analysis,the entropy of frac in ETR showed the best differentiation performance with AUC 0.7419(sensitivity = 0.88,specificity = 0.55).When integrated with the skewness of frac in ETR,clear tumor edges help improve performance at an optimal AUC of 0.7928(sensitivity = 0.81,specificity =0.71).Conclusion:fmean and fkurtosis in enhanced tumor region differ between glioma and SBM and their combined model could help the differentiation.Skewness(Dfast,frac),kurtosis(Dslow,frac),entropy(Dslow,frac)in tumor region,and mean(frac)in tumor and peritumoral edema zone showed significant differences between HGG and SBM.Clear tumor region in T2 W improved the differentiation with optimal AUC 0.7928 when integrated with skewness of frac.