Texture Analysis-based Multiparametric MRI For Assessing Renal Injury In Animal Model And Renal Dysfunction In A Clinical Study

Part ? Assessment of kidney injury using texture analysis-based multiparametric MRI in animal modelObjective To compare the potential value of multiparametric MRI based on texture analysis in assessing the severity grade of acute kidney injury(AKI).Materials and Methods Seventy-six healthy New Zealand white rabbits were injected intramuscularly with nephrotoxic drugs(gentamicin,n=62)or saline(n=14).From 2 to 7 days after the administration,the multiparametric MRI,including 3 imaging methods listed as diffusion weighted imaging(DWI)where the quantitative analysis is performed on the apparent diffusion coefficient(ADC),blood oxygen level-dependent MRI(BOLD)and susceptibility weighted imaging(SWI),was performed on the left kidney of the rabbits.Then,blood was collected to measure the serological creatinine(Cr)and urea nitrogen(SUN)level.A histological examination was performed on the left kidney removed.The severity grade of AKI was assessed by histological analysis,being classified into 5 grades,including grade 0,1,2,3 and 4,where grade 0 is non-renal injury,and grade 5 is at the most serious level.The MRI image analysis of left kidney is performed using both the traditional manual method of region of interest and the method of texture analysis.There are many imaging parameters,and the selected features must meet the following screening criteria:1)correlation with the histological score>0.3 or<-0.3(and P<0.05);2)one of the highly-correlated texture features(r? 0.9)was retained for each imaging method;3)Inter-rater intraclass correlation coefficient>0.75,being acceptable for repeatability analysis.Their efficacy of the selected imaging parameters for differentiating the severity of AKI was compared using the receiver operating characteristic curve(ROC).Results Sixty-six cases fulfilled all multi-parameter MRI,serological and histological examinations,and were included in statistical analysis,including 28 cases with renal injury of grade 0,17 cases with grade 1,8 cases with grade 2,4 cases with grade 3 and 9 cases with grade 4.Among the nine features measured using the manual method,ADCmedulla and BOLDcortex met the screening criteria mentioned above.Among the texture features,10 features met the screening criteria mentioned above,including ADCCorrelation,ADCKurtosis,BOLDHomogeneity,BOLD0.75Quantile,B OLD Variance,BOLDEnergy,BOLD ClusterP rominence,SWIHomogeneity,SWIEnergy and SWIMAD.In the rabbits injected with saline,the 95%range of Cr and SUN were 35-75 ?mol/l and 3.3-9.2 mmol/l,respectively.Therefore,renal injury was predicted when Cr>75 ?mol/l or SUN>9.2 mmol/1.Because of the small sample size in some renal injury grades,the 66 cases were classified into the non-renal injury group(non-RI)composed of the cased with renal injury of grade 0(n=28),the mild-moderate renal injury group(mRI)composed of the cases with renal injury of grade 1 and 2(n=25),and severe renal injury group(sRI)composed of the cases with renal injury of grade 3 and 4(n=13).The areas under ROC(AUC)of BOLD0.75Quantile,BOLDEnergy,BOLDHomogeneity,BOLDvariance,SWIEnergy,SWIHomogeneity,Cr and SUN were larger than 0.65 for differentiating the mRI from non-RI,and no statistical difference was found between their AUC(all P>0.05).The AUC of ADCmedulla,SWIMAD,Cr and SUN for differentiating the sRI from mRI were larger than 0.65,and the discriminant efficiency of ADCmedulla was comparable to that of Cr and SUN(all P>0.05).Conclusion The DWI based on the traditional method of imaging analysis is suitable for differentiating the sRI from mRI,and its discriminant efficacy is comparable to that of Cr and SUN.The BOLD and SWI based on texture analysis are suitable for differentiating the mRI from non-RI,and the discriminant efficacy is comparable to that of Cr and SUN.Part ? Evaluation of renal dysfunction using texture analysis-based multiparametric MRI in a clinical studyObjective To explore the value of texture analysis-based multiparametric MRI,including diffusion weighted imaging(DWI),blood oxygen level-dependent MRI(BOLD),and susceptibility weighted imaging(SWI)in evaluating renal dysfunction.Materials and Methods Seventy-two patients(mean age 53.72 ± 13.46 years)underwent MRI examination consisting of DWI,BOLD,and SWI.Their clinical data used to calculate the glomerular filtration rate were prepared.According to their estimated glomerular filtration rate(eGFR),the patients were classified into either the severe renal function impairment group(sRI,eGFR<30 mL/min/1.73 m2),the non-severe renal function impairment group(non-sRI,eGFR? 30 mL/min/1.73 m2 and<80 mL/min/1.73 m2),or the control group(CG,eGFR? 80 mL/min/1.73 m2).Thirteen texture features were extracted after the texture analysis performed on each imagimg method,and then were analysed to select the most valuable features for discerning the three groups for each imaging method,respectively.A random forest plot and a linear correlation were performed to select and rank the importance of texture features.The most valuable features had to meet the following criteria:(a)the correlation coefficient with eGFR of>0.3 or<-0.3(with P<0.05),which was considered as being of potential value in assessing renal dysfunction and(b)one of the top five important features listed in the mean decrease accuracy and also in the mean decrease Gini index for each imaging method.A receiver operating characteristic curve was performed to compare the capacities of the features to differentiate non-sRI from sRI or CG.Results Six features proved to be the most valuable for assessing renal dysfunction were DWI0.25Quantile,DWI0.5Quantile,DWIHomogeneity,BOLDEntropy,SWISkewness,and SWICorrelation.Three features derived from DWI(DWI0.25Quantile,DWI0.5Quantile,and DWIHomogeneity)were smaller in sRI than that in non-sRI group(all P<0.05).BOLDEntropy and SWICorrelation were smaller in non-sRI than that in CG group(all P<0.05).DWI0.25Quantile,DWI0.5Quantile,and DWIHomogeneity showed similar capacities for differentiating sRI from non-sRI group(all P>0.05).Similarly,BOLDEntropy and SWICorrelation showed an equal capacity for differentiating non-sRI from CG group(P>0.05).Conclusion Texture analysis-based DWI,BOLD,and SWI may assist in assessing renal dysfunction,and texture analysis-based BOLD and SWI may be suitable for assessing renal dysfunction during early stages.