Diffusion-weighted Imaging Based On Multiple Mathematical And Models In Assessing The Response To Neoadjuvant Chemotherapy Radiation Therapy In Locally Advanced Rectal Cancer

ObjectiveIn this study,we explored the diagnostic performance of multiple mathematical models(ADC,IVIM,and SEM)according to histogram analyses derived from the whole-tumor volume combined with prognostic factors in order to assess the response of LARC to CRT by using the tumor regression grade(TRG)as the reference standard.Materials and Methods1.Clinical informationSixty patients who were clinically diagnosed as primary rectal cancer and confirmed by surgery and pathology from October 2015 to November 2017 were enrolled in this single-institution study.MRI examination was performed before NCRT in all patients.Among them,46 were male patients and 14 were female patients,the ratio of male to female was about 2.24:1,the age was 27-78 years old,with an average age of 54.63±8.5 years.2.Methods2.1 EquipmentAll the patients were performed MRI with the Philips Achieva 3.0 Tesla magnetic resonance,with abdominal 16-channel phased-array surface coil.2.2 Main sequence and parameterAxial DWI 9b values(0,10,20,40,80,150,300,600,1000 s/mm2);Other sequences(refer to main document).2.3 Image analysisDelineation of whole-tumor volume of interest(VOI):two radiologists manually delineated tumor boundaries on axial HR-T2WI with commercial software(TK-SNAP).Similarly,VOIs was delineated on DWI(b =1000s/mm2)with reference to HR-T2WI.What’s more,VOIs delineation on HR-T2WI sequence was used for tumor volume calculation.2.4.1 9b-value DWI data were imported into MATLAB software package developed internally(MathWorks Inc.,Natick,MA,USA)to calculate the parameters of multiple models and compare the predictive value of histogram analysis of each parameter for the efficacy of radiotherapy and chemotherapy for primary rectal cancer.1.Mono-exponential model:S(b)/S(0)= exp(-b x ADC),where b is the level of diffusion weighting.ADC was calculated from a mono-exponential fit using 9b values(0,10,20,40,80,150,300,600,and 1000 s/mm2).2.Bi-exponential model:S(b)/S(0)= f x exp(-b × Dfast)+(1-f)× exp(-b × Dslaw),where f is the diffusion fraction linked to microcirculation,which is also called the perfusion fraction,Dfast is the fast diffusion rate constant,and Dslow is the slow diffusion rate constant.3.SEM:S(b)/S(0)= exp{(-b x s)α },where DDC is a measure of the mean intravoxel diffusion rate in the presence of heterogeneity and the heterogeneity index a characterizes the deviation of the signal attenuation from mono-exponential form,which is limited to values between zero and one.2.4.2 Firstly,the data of 9 b-value DWI were imported into MATLAB software package developed internally(MathWorks Inc.,Natick,MA,USA)and ADC values using the data of 2 b-value(0,1000 s/mm2)and 9 b-value(0,10,20,40,80,150,300,600,1000 s/mm2)for calculation,and the calculation model is a Mono exponential model:s(b)/s(0)= exp(-b × ADC).To compare the predictive value of 2b value and 9b value of ADC value histogram analysis parameters in the efficacy of radiotherapy and chemotherapy for primary rectal cancer.2.5 Statistical analysisPostoperative pathological results were used as the grouping criteria,and the pathological complete response was pCR group and the pathological incomplete response was non-pCR group.SPSS 20.0 statistical software package was used for statistical analysis.(1)intra-class correlation coefficient(ICC)was applied to evaluate the inter-observer reliability of ADC,IVIM,SEM and tumor volume parameters.(2)To compare the predictive value of histogram analysis parameters of 2b value and 9b value of ADC value in the efficacy of radiotherapy and chemotherapy for primary rectal cancer.Using t test or U test evaluated ADC2b and ADC9b histogram parameters statistical difference between the pCR group non-pCR group respectively,and use the receiver operating characteristic curve(the ROC curve)and the Area Under the curve(AUC)to compare statistically significant difference histogram parameters of ADC9b and ADC2b in predicting diagnostic efficacy of radiotherapy and chemotherapy for primary rectal cancer.(3)t test or U test were used to evaluate whether there were statistical differences in HR-MRI morphological indexes and clinical indexes(CRM,EMVI,CEA,leukocyte value,tumor size)between the pCR group and the non-pCR group respectively,and ROC curve and AUC were used to compare the diagnostic efficacy of parameters with statistical differences.(4)t test or U test were used to evaluate whether there were statistical differences in the histogram analysis parameters of IVIM and SEM models between the pCR group and the non-pCR group respectively,then ROC curve and AUC were used to compare the diagnostic efficacy of histogram analysis parameters which were statistical differences.(5)multivariate logistic regression analysis was used to screen the morphological and clinical parameters of ADC,IVIM,SEM and HR-T2WI with statistical differences,and the selected parameters were combined by binary logistic regression analysis.Finally,ROC curve and AUC were used to calculate the diagnostic efficacy of the combined parameters.Results1.Inter-observer reliability results:In terms of the reproducibility,the ICCs for the ADC and SEM histogram parameters obtained from pre-CRT had good or excellent correlations.However,the ICCs for IVIM mostly indicated fair correlations.2.Among the morphological of HR-MRI and clinical parameters,only tumor volume was statistically different between the two groups.3.The mean,median,SD,and 10th,25th,75th,90th,and 95th percentiles of the ADC maps were lower in the pCR group compared with the non-pCR group(p =0.000-0.013),the AUC value for the ADC mean was highest(AUC = 0.890;p = 0.000;95%CI = 0.806-0.974).For the histogram analysis parameters in the IVIM model,only the Dslow median differed significantly between the two groups(p = 0.032).The mean,median,skewness,kurtosis,and 25th,75th,90th,and 95th percentiles for the DDC maps were lower in the pCR group compared with the non-pCR group(p =0.000-0.013).When these eight parameters were tested separately by ROC curve analysis,the AUC value for the DDC median(0.866,p = 0.000,95%CI= 0.771-0.960)was highest,followed by that of the DDC mean,and the AUC value was lowest for the DDC 95th percentile.4.Among the combined diagnosis results of parameters,the ADC mean in histogram parameters combined with tumor volume had the highest diagnostic performance(AUC=0.908),with a sensitivity of 100%and a specificity of 81%.5.Only the ADC2b-values skewness differed significantly between the two groups(p = 0.035).The result shows that mean,median,SD,and 10th,25th,75th,90th,and 95th percentiles of ADC9b-values were significantly lower in the pCR group than in the non-pCR group(p = 0.001-0.025).The highest diagnostic performance was obtained for the ADC9b-values mean(AUC = 0.890;p = 0.000;95%CI= 0.806-0.974)with a sensitivity of 100%and specificity of 70.2%.Conclusion1.We found that combining the whole-tumor ADC mean with the tumor volume was highly accurate at predicting the response of LARC to CRT.2.Furthermore,our preliminary results indicated that the mean value was better than or equal to histogram analysis,and thus the mean value may be suitable for use in everyday clinical practice.3.Finally,according to our results,IVIM models were inferior to mono-exponential modeling and SEM in terms of predicting pCR as well as their reproducibility.