Assessment Of Hypoxia In Pancreatic Ductal Adenocarcinoma Based On CT Manifestations And Texture

[Objective]To explore the predictive value of computed tomography(CT)manifestations and texture features for hypoxia in pancreatic ductal adenocarcinoma(pancreatic ductal adenocarcinoma,PDAC).[Methods]Forty-three patients with postoperative pathologically confirmed pancreatic ductal adenocarcinoma and preoperative enhanced CT abdomen were retrospectively collected.The expression levels of Hif-1α,CD 105 and collagen fibers in the tumors were evaluated by immunohistochemical staining and Masson staining.The CT features of tumor size,location,regularity of morphology,presence of cystic necrosis,vascular infiltration,fatty infiltration,pancreatic duct dilatation,peripheral lymph node enlargement,ascites and enhancement of tumor parenchyma were evaluated visually.The correlation between the above imaging features and pathological indices was calculated and analyzed,and the categorical data were expressed as the number of cases(percentage),and the χ2 test was used for comparison between groups,and Spearman correlation analysis was used for continuous data.The patients’ clinical,imaging,and pathological data were further uploaded to the Radcloud platform(Wisepoint Medical Technology Co.,Ltd.),and the enhanced CT scan images of the lesions were outlined as regions of interest(ROI),and the pathological data were used as labels,and the Variance Threshold method and The best texture features of arterial phase,venous phase and delayed phase images were filtered by the SelectKbest method in descending order,the stepwise regression method was selected to filter the variables and construct logistic regression models,and the ROC curves were used to evaluate the predictive efficacy of the best texture features for intra-tumor hypoxia.[Results]There was a statistically significant correlation(P<0.05)between the three featu res of cystic necrosis(P=0.011),vascular infiltration(P=0.008),and pancreatic du ct dilatation(P=0.023)and the degree of differentiation of PDAC among the imagin g indexes assessed visually,and there was no statistical correlation between the other imaging features and the degree of differentiation and between each imaging feature a nd the collagen fiber area ratio,HIF-lα expression,and MVD significant correlation.The results of the Spearman correlation analysis between the degree of strengthening(TPER),the degree of differentiation,the collagen fiber area ratio,the HIF-la expressi on score,and the MVD in the arterial,venous,and delayed stages showed that only t he degree of strengthening in the arterial stage of the tumor parenchyma was statistica lly significantly positively correlated with the MVD(P value 0.033),and there was no significant correlation between the degree of strengthening and the pathological index es in the other stages(P>0.05).There was no significant correlation(P>0.05)bet ween the remaining stages of enhancement and pathological indices.The AUC value o f the ROC curve for predicting MVD by the degree of arterial stage enhancement wa s 0.608.For the collagen fiber area ratio grouping label,seven features were screened by the variance threshold method and SelectKBest method in the arterial phase.For these seven features,the stepwise regression method was selected to screen the variables to avoid multicollinearity,and after six methods of forward-conditional,forward-LR,for ward-Wald,backward-conditional,backward-LR,and backward-Wald,the final common screening was HighGrayLevelZoneEmphasis_glszm_wavelet-LHL,Kurtosis_firstorder_ex ponential,plotting ROC curve,AUC values of 0.680,0.647 respectively;the same met hod venous period finally common screening is SmallAreaHighGrayLevelEmphasis-glsz m_wavelet-LHL,plotting ROC curve,AUC value of 0.734;delayed period finally com mon screening is SizeZoneNonUniformity_glszm_wavelet-HHH,plotting ROC curve Th e AUC value was 0.702.For the HIF-1α expression grouping label,five features were screened by the vari ance threshold method and SelectKBest method in the arterial period,and for these fi ve features,the stepwise regression method was selected to screen the variables,and t he final common screening was TotalEnergy_firstorder_wavelet-HLH,HighGrayLevelZo neEmphasis_glszm_wavelet-HHL,plotted ROC curve,AUC value of 0.624,0.691;the same method venous period finally jointly screened is Kurtosis_firstorder_wavelet-LHL,plotted ROC curve,AUC value was 0.761;the final screening results in the delayed phase were inconsistent and P>0.05.For the MVD grouping label,the arterial phase was screened for 7 features by th e variance threshold method and SelectKBest method,and for these 7 features,the ste pwise regression method was selected to screen the variables,and the final common s creening was LargeAreaLowGrayLevelEmphasis_glszm_wavelet-LHL,and the ROC curv e was plotted,with an AUC value of 0.816;the same method venous period finally c ommon screening is LargeDependenceHighGrayLevelEmphasis-gldm_wavelet-LLH,Gray LevelNonUniformity-glszm_wavelet-LHH,plotting ROC curve,AUC value 0.781 and 0.716,respectively;the final common screening in the delay period is LargeDependence LowGrayLevelEmphasis_gldm_wavelet-HLH,and the ROC curve is plotted with an A UC value of 0.686.[Conclusions]There is a statistically significant correlation between three features of cystic necrosis,vascular infiltration,and pancreatic duct dilatation and the degree of differentiation of PDAC among the imaging indexes evaluated visually,however no significant correlation is found between other imaging features and the degree of differentiation and between each imaging feature and collagen fiber area ratio,HIF-la expression,and MVD.A statistically significant positive correlation is found between the degree of enhancement in the arterial stage of tumor parenchyma and MVD,however there is no significant correlation between the degree of enhancement in the remaining stages and the pathological indexes.Its predictive efficacy is poor without enough assessment value on tumor MVD.Among the three stages,the image features in the venous stage have the best predictive efficacy for tumor collagen fiber area and tumor hypoxia,and the image features in the arterial stage have the best predictive efficacy for tumor MVD.