Prediction Of Ki-67 Expression Level In Peripheral Lung Cancer Based On CT Radiomics

Objective: to explore the feasibility and performance of radiomics based imaging model for predicting Ki-67 expression in peripheral lung cancer.Materials and Methods:417 cases of peripheral lung cancer confirmed by pathology and immunohistochemistry from January 2016 to October 2019 were collected and divided into training cohort(n=293)and verification cohort(n=124).All patients underwent biphasic CT dynamic contrast-enhanced examination,and 396 histological features were extracted from the arterial phase and venous phase images of each patient.The least absolute shrinkage and selection operator regression model(LASSO)dimensionality reduction analysis suitable for high-dimensional data is used for feature screening,meaningful features are selected and imaging tags are established in the training queue.Multivariate regression analysis was used to establish three models(radiomics model,clinicopathological model,radiomics nomogram).Finally,the ROC curve and the area under the curve(AUC)were used to evaluate the prediction efficiency of the three models,and the decision curve analysis was used to evaluate the clinical practicability of the radiomics nomogram in the training queue and verification queue.Delong test is used to compare whether there is a significant difference in prediction performance among the three models.The correction curve is used to evaluate the correction effect of radiomics nomogram.Finally,Hosmer-Lemeshow test is used to analyze whether there is a significant difference between risk rate prediction and observation probability.The clinicopathological features included age,sex,smoking history,histological type,tumor stage and CT signs,and CT signs included tumor maximum diameter(Dmax),lobulation sign,spiculation sign,cavity,liquefied necrosis,pleural indentation,vacuole sign,air bronchus sign and vascular convergence sign.The difference of Ki-67 expression level in clinicopathological features was analyzed by Wilcox-on rank sum test or Fisher test,and univariate and multivariate Logistic regression analysis was performed.To analyze the correlation between the expression level of Ki-67 and clinicopathological features and to establish a clinicopathological model.There was a statistical significance when P was less than 0.05.Results: In this study,there were 73 cases(24.9%)in the high expression group and220cases(75.1%)in the low expression group in the training cohort.There were significant differences in sex,smoking,histological type and tumor stage between Ki-67 high expression group and low expression group(P<0.05).High expression of Ki-67 is more common in smoking,male and adenocarcinoma,while low expression of Ki-67 is more common in early patients.In the verification cohort,there were 31 cases in the highexpression group of Ki-67(25.0%)and 93 cases in the low expression group(75.0%).There were significant differences in sex,histological type and tumor stage between Ki-67 high expression group and low expression group(P<0.05).High expression of Ki-67 is more common in men and adenocarcinomas,while low expression of Ki-67 is more common in early patients.In the training group,Ki-67 was highly expressed in patients with cavities,liquefaction necrosis,larger tumor and lobulation sign(p<0.05).In the verification cohort,Ki-67 was highly expressed in patients with liquefaction necrosis,large tumor and pleural indentation sign(p<0.05).In arterial phase,univariate analysis showed that there were significant differences between Ki-67 expression level and sex,smoking status,tumor clinical stage,histological type,maximum diameter of tumor(Dmax),air bronchus sign,lobulation sign,spiculation sign and vascular convergence sign.Multivariate logistic regression analysis showed that histological subtypes,maximum tumor diameter((Dmax)),air bronchus sign,vascular convergence sign and lobulation sign were independent risk factors for predicting Ki-67 expression in peripheral lung cancer(p<0.05).In the venous phase,univariate analysis showed that the expression of Ki-67 was significantly correlated with smoking status,tumor clinical stage,histological subtype,maximum tumor diameter(Dmax),liquefaction and necrosis,lobulation sign and cavity(p<0.05).Multivariate logistic regression analysis showed that sex,clinical stage,histological subtype,(Dmax),liquefaction necrosis,lobulation sign and spiculation sign were independent risk factors for predicting Ki-67 expression in peripheral lung cancer.The taxonomic tags composed of related taxonomic characteristics have good predictive effect in the training cohort(arterial phase AUC,0.76;95%CI,0.70-0.82);venous phase AUC,0.76;95%CI,0.69-0.82 andverification cohort(arterial phase AUC,0.76;95%CI,0.67-0.86;venous phase AUC,0.76;95%CI,0.67-0.85).In the training cohort,the arterial phase imaging nomogram was composed of histological subtype,tumor maximum diameter(Dmax),lobulation sign,vascular convergence sign,air bronchial sign and imaging label.Venous phase radiomics nomogram is composed of sex,tumor clinical stage,histological subtype,tumor maximum diameter(Dmax),lobulation sign,spiculation sign and imaging label.Compared with other models,it has the best diagnostic efficacy(arterial phase AUC,0.86;95%CI,0.81-0.91;venous phase AUC,0.81;95%CI,0.75-0.87),which is higher than that of clinical model(arterial phase AUC,0.84;95%CI,0.79-0.89;venous phase AUC,0.77;95%CI,0.71-0.83).In the verific-ation cohort,the predictive performance of radiomics nomogram(arterial phase AUC,0.79;95%CI,0.69-0.89;venous phase AUC,0.81;95%CI,0.72-0.90)was also better than that of clinical models(arterial phase AUC,0.78;95%CI,0.68-0.88;venous phase AUC,0.79;95%CI,0.69-0.90).The alignmentdiagram based on radiomics nomogram has good correction efficiency in both training queue and verification queue.There is no significant difference between the predicted value of the analysis risk rate and the observation probability,and there is no deviation from the perfect fit.Conclusion:Radiomics based on CT enhanced images can provide a method to predict the level of Ki-67 expression in peripheral lung cancer.The combination of radiomics model and clinicopathological model(radiomics nomogram)can improve the predictive performance of the predictive model,and the arterial phaseradiomicsnomogram has the best predictive performance than the venous phase.this may help to provide a new non-invasive way to understand the molecular information of lung cancer cells.