Evaluation Of Multiple Prognostic Factors Of Invasive Breast Carcinoma Of No Specific Type With Intravoxel Incoherent Motions Imaging By Extracting The Histogram Metric

Objective:It was to explore the multiple prognostic factors(ER,PR,Her-2,Ki-67,CD31,D2-40,ALNM and molecular typing)of IBC-NST based on the extraction of IVIM histogram indicators to make an evaluation.Materials and Methods:A retrospective analysis was performed on 125patients with IBC-NST diagnosed by pathology in our hospital from September 20,2018 to November 10,2020,with a total of 125 lesions.All patients underwent breast MRI examination(including IVIM)before needle biopsy or surgery,the boundary of lesions was manually delineated layer by layer with Fire Voxel software to obtain the ADC,D,D*and f value of quantitative parameters,and the histogram analysis of each quantitative parameter was further carried out.The extracted histogram parameters included volume,mean,median,standard deviation,Inhomogeneity,skewness,kurtosis,entropy,minimum,maximum,percentile(5th～95th,from the 5th percentile,start every 5th percentile until the 95th percentile).All patients were grouped according to the expression or status of ER,PR,HER-2,Ki-67,CD31,D2-40,ALNM,and molecular typing.Kolmogorov-Smirnov test was performed on all ADC,D,D*and f value derived histogram parameters to verify the distribution,followed by Mann-Whitney U-test or Independent student t-test,to assess whether there were differences in IVIM derived histogram in the indicators among the each prognostic factor groups,so as to obtain candidate diagnostic indicators,and then explored the potential correlation between the IVIM histogram parameters and ER,PR,Her-2,Ki-67 through Spearman rank correlation analysis.Then,principal component analysis(PCA)was implemented to integrate candidate diagnostic indicators,and the principal components obtained from PCA were used to establish LG model based on Logistic regression,and the area under curve was calculated by ROC analysis to evaluate the diagnostic efficacy of IVIM histogram parameters for various prognostic factors.Result:1.Volume,ADC(Mean,Median,SD,Entropy,Kurtosis,Inhomogeneity,Min,Max,5th～45th,55th～95th),D(Mean,SD,Entropy,Skewness,Kurtosis,Inhomogeneity,Min,Max,5th～80th,90th,95th),f(Mean,SD,Kurtosis,Skewness,Inhomogeneity,5th,15th,20th,35th,85th)had different degrees of differences between groups in different prognostic factors,the differences were statistically significant(P<0.05).2.The correlation between ER and D 15th(r_s=-0.254,P=0.004),PR and D Max(r_s=0.266,P=0.046),Her-2 and D 5th(r_s=0.191,P=0.033),Ki-67 and ADC Max(r_s=0.300,P=0.001)were the strongest in each group respectively.3.The PCA analysis compressed 24,28,3,22,18,7,5,22,6,13 and 18candidate markers into 4,6,2,6,3,2,3,6,2,4 and 4 PCs,respectively,among the 11dichotomous prognostic factors.The cumulative contribution rates of PCs were88.821%,86.382%,86.802%,88.815%,91.901%,86.190%,74.494%,91.275%,75.996%,88.474%,86.882%.The results of ROC showed that LG model had a better diagnostic predictive value than any of single IVIM histogram parameters(AUC>0.5),among which LG predicted Luminal A type with the highest AUC(AUC=0.857,95%CI of AUC:0.751-0.962),where the lowest AUC was LG predicted Luminal B type(AUC=0.645,95%CI of AUC:0.546-0.743).Conclusion:Multiple prognostic factors of non-specific invasive breast cancer can be predicted based on the extraction of IVIM histogram indicators.Different IVIM histogram parameters had different diagnostic efficiency for various prognostic factors of IBC-NST,the ADC and D of the histogram parameters in predicting prognostic factors of this research were superior to the D*and f parameters.The optimal evaluation indicator was the integrated variable(LG model)of PCA analysis.The application value of LG model in IBC-NST was very considerable.