The Study Of Preoperative Prediction Of Visceral Pleura Invasion And Invasiveness Based On Imaging In Clinical Stage IA Lung Adenocarcinoma

Part 1 The value of CT features in predicting visceral pleural invasion in subpleural clinical stage IA lung adenocarcinoma【Objective】To investigate the value of CT features in predicting visceral pleural invasion（VPI）in clinical stage IA peripheral lung adenocarcinoma with pleural contact.【Methods and Materials】According to the type of contact between tumor and pleura,the patients were divided into pleural tags type and pleural attachment type.Pleural tags were defined as one or multiple high-density linear strands connecting the tumor margin and the pleura with the distance between the lesion and the pleura（DLP）ranged from 0mm to 10 mm.Pleural attachment was defined as tumor directly contacting the pleura（DLP= 0mm）.All the patients were divided into VPI-positive group and VPI-negative group according to the VPI status of pathological report.404 patients with clinical stage IA peripheral lung adenocarcinoma under the pleura diagnosed in our hospital were retrospectively collected as an internal dataset,including 166 cases with pleural tags and 238 cases with pleural attachment.The training set and the internal validation set were partitioned in the internal dataset at a ratio,respectively.The external validation set included 81 patients from Zhejiang Provincial People’s Hospital and the First Affiliated Hospital of Soochow University,including 36 cases with pleural tags and 45 cases with pleural attachment.CT feature analysis included:（1）For the assessment of pleural end signs: the pleural indentation sign was evaluated,and its subgroup analysis was performed in the interlobar and non-interlobar pleura groups.The classification of pleural morphology and density changes were qualitatively analyzed in the two groups.（2）For the assessment of tumor signs: the tumor size and the solid component size in the two groups were quantitatively measured,and consolidation-to-tumor ratios（CTR）was calculated.The tumor location,density type（mixed ground glass nodules,solid nodules）,shape（round/round-like,irregular）,interface（clear,unclear）,margin（lobulation,spiculation）,internal structure（vacuoles,cyst/cavity）,adjacent structure（air bronchogram signs,vascular convergence sign）,whether combined with emphysema.（3）For the assessment of tumor-pleural signs:for the tumors with pleural tags,DLP was quantitatively measured,whether combined with bridge tag sign and morphological classification of pleural tags were qualitatively analyzed.For tumors with pleura attachment,the whole tumor contact length and the solid component contact length were quantitatively measured,and the ratio of whole tumor contact length to the maximum diameter of the tumor was calculated.Qualitative analysis of whether solid components contact the pleura,whether combined with pleural tags sign and the classification based on the proportion of whole tumor contact length and pleural morphology was performed.Univariate analysis was used to identify variables associated with VPI in both groups.Variables with P value < 0.1 in univariate logistic regression analysis were included for multivariate logistic regression analysis to determine the independent risk factors of VPI.Variables were selected by Backward stepwise method,and a multivariate logistic regression model was established.The discrimination of the model was evaluated by ROC curve and AUC value in the training set,internal validation set and external validation set.【Results】For tumors with pleural tags or pleural attachment,there was no significant difference in the density change types of pleural indentation area between the two groups（P > 0.05）.There was no significant difference in interlobar pleural indentation sign between the two groups（P > 0.05）.For tumors with pleural tags,there were 89 cases with VPI-positive and 113 cases with VPI-negative among the 202 patients with lung adenocarcinoma.There were 49 cases with VPI and 65 cases without VPI in the training group.Univariate analysis found that the solid component size,CTR,density type,spiculation sign,vascular convergence sign,whether combined with emphysema,pleural tags type and bridge tag sign were significantly different between the two groups（P<0.05）.The solid component size（OR=1.12,95%CI 1.03～1.22,P=0.011）,pleural tags type（OR=2.36,95%CI 1.39～4.45,P=0.010）and vascular convergence sign（OR=9.53,95%CI 2.27～66.21,P=0.006）were independent risk factors for VPI analyzed by multivariate logistic regression analysis.The best combination of predictors included the solid component size,vascular convergence sign,combined with emphysema and pleural tags type,were selected by multivariate logistic regression analysis to construct a model.The accuracy,sensitivity,specificity,and AUC for the logistic model in the training set were 82.46%,83.67%,81.54% and 0.887,respectively,using the optimal cutoff value of 0.435.The accuracy,sensitivity,specificity,and AUC for the model in the internal validation set were 78.84%,72.41%,86.96% and 0.799,respectively.The external validation set obtained an acuraccy,sensitivity,specifcity,and AUC of 88.89%,72.73%,96.00% and 0.862,respectively.For tumors with pleural attachment,there were 137 cases with VPI-positive and 146 cases with VPI-negative among the 283 patients with lung adenocarcinoma.There were85 cases with VPI-positive and 84 cases with VPI-negative in the training group.Univariate analysis found that the size of the solid part,CTR,the whole tumor contact length,and the solid component contact length,the density type,presence or absence of solid component contacting the pleura,pleural indentation,spiculation,air bronchogram sign,vascular convergence sign and combined with emphysema were significantly different between the two groups（P<0.05）.The size of solid component（OR=1.28,95%CI 1.18～1.42,P<0.001）and pleural indentation（OR=2.73,95% CI 1.06～7.43,P=0.041）were independent risk factors for VPI analyzed by multivariate logistic regression analysis.The best combination of predictors included the size of solid component,pleural indentation,combined with pleural tags and presence or absence of solid component contacting the pleura,selected by multivariate logistic regression analysis to construct a model.The best predictors.The accuracy,sensitivity,specificity,and AUC for the logistic model in the training set were 82.25%,75.29%,89.29% and 0.903,respectively,using the optimal cutoff value of 0.643.The accuracy,sensitivity,specificity,and AUC for the logistic model in the internal validation set were 76.81%,90.32%,65.79% and 0.848,respectively.The external validation set obtained an acuraccy,sensitivity,specifcity,and AUC of 80.00%,81.82%,78.26% and 0.842,respectively.【Conclusion】For clinical stage IA lung adenocarcinoma in contact with pleura,it is necessary to pay attention to the solid component size.For tumors with pleural tags,type Ⅲ pleural tags sign（the tumor was connected with the pleura with one or multiple thick strip with pleural indentation,showing a typical bell-shaped change）,vascular convergence sign and whether combined with emphysema are the key factors to predict VPI status.For tumors with pleural attachment,whether combined with pleural tags and solid components in contact with the pleura,and pleural indention sigh are the key factors to predict VPI status.Attention to these CT features can help to determine whether the visceral pleura in contact with the clinical stage IA lung adenocarcinoma is invaded before surgery,and provide guidance for clinical treatment decisions.Part 2 The value of a nomogram based on ¹⁸F-FDG PET/CT in predicting visceral pleural invasion of clinical stage IA subpleural lung adenocarcinoma【Objective】To investigate the value of SUVmax based on ¹⁸F-FDG PET/CT combined with CT signs in predicting visceral pleural invasion（VPI）in clinical stage IA lung adenocarcinoma with pleural contact.【Materials and Methods】Retrospective analysis was performed on patients diagnosed clinical stage IA lung adenocarcinoma who had ¹⁸F-FDG PET/CT examination before surgery and simultaneous HRCT showed that the tumor was in contact with the pleura.A total of 140 patients were included in the analysis after excluding pure ground glass nodules.The patients were divided into training set and internal time verification set according to the examination time sequence and the ratio of 7:3,and were divided into VPI-positive group and VPI-negative group according to the VPI status of pathological report.In the training set,univariate analysis was performed to compare clinical data（gender,age,smoking history,CEA level）,SUVmax value,CT characteristics（tumor size,solid component size,CTR,whole tumor contact length,solid component contact length,DLP,tumor location,density,type of contact with the pleura,whether the solid component contacts the pleura,pleural indentation,shape,margin,internal structure and adjacent structure）between the two groups.Multivariate logistic regression analysis was used to identify the independent risk factors of VPI,and the factors included in the multivariate logistic regression analysis were variables with P value < 0.1 in univariate analysis.Variables were selected by backward stepwise method,and a multivariate logistic regression model was established and a nomogram was developed.The ROC curve and AUC value were used to evaluate the diagnostic performance of the models in the training set and internal validation set.The goodness-of-fit of the model was evaluated by the Hosmer-Lemesow test and calibration curve.The practical application value of this model was analyzed by decision curve analysis（DCA）.【Results】Among 140 cases of lung adenocarcinoma,57 cases were VPI-positivee and 83 cases were VPI-negative.Among the 98 lung adenocarcinomas in the training set,40 cases were VPI-positive.In the training set,univariate analysis found that SUVmax,solid component size,CTR,solid component contact length,density type,spiculation sign,vascular convergence sign and pleural indentation sigh were statistically different between VPI-positive and VPI-negative lung adenocarcinoma patients（P < 0.05）.Multivariate logistic regression analysis was used to select the best predictors combination,including SUVmax,solid component contact length,pleural indentation sign and vascular convergence sign.SUVmax（OR = 1.753,95%CI 1.232 2.496,P =0.002）,solid component contact length（OR = 1.101,95%CI 1.007 1.204,P = 0.034）,pleural indentation sign（OR = 5.075,95%CI 1.065 24.172,P = 0.041）and vascular convergence sign（OR = 13.324,95%CI 1.379 128.691,P = 0.025）were independent risk factors for VPI in patients with lung adenocarcinoma.A nomogram was constructed based on the multivariate logistic regression model,with 0.35 as the cut-off value,the accuracy,sensitivity and specificity of the model for predicting VPI in the training set were 80.61%,82.50% and 79.31%,respectively,the AUC value was 0.892（95%CI0.813⁰.946,P < 0.001）.In the internal validation set,the accuracy,sensitivity and specificity of the model was 85.71%,100.00%,76.00%,and the AUC value was 0.885（95%CI 0.748⁰.962,P < 0.001）.The calibration curve demonstrated that its predicted probabilities were in acceptable agreement with the actual probability.Decision curve analysis（DCA）illustrated that the current nomogram would add more net benefit.【Conclusion】SUVmax,solid component contact length,pleural indentation sign and vascular convergence sign are independent risk factors for predicting VPI in patients with clinical stage IA subpleural peripheral lung adenocarcinoma.The nomogram integrating SUVmax and CT features,could noninvasively predict VPI status before surgery in subpleural clinical stage IA lung adenocarcinoma patients.Part 3 The value of a nomogram based on intratumoral and peritumoral radiomics and CT features for predicting visceral pleural invasion of clinical stage IA lung adenocarcinoma【Objective】To evaluate the value of a nomogram based on intratumoral and peritumoral radiomics and CT features for predicting visceral pleural invasion（VPI）of clinical stage IA lung adenocarcinoma【Methods and Materials】A total of 404 cases of clinical stage IA lung adenocarcinoma diagnosed in our hospital were retrospectively collected as an internal data set,which were randomly divided into training set（n= 283,134 cases with VPI-positive and 149 cases with VPI-negative）and internal validation set（n= 121,60 cases with VPI-positive and 61 cases with VPI-negative）at a ratio of 7:3.Patients with clinical stage IA lung adenocarcinoma diagnosed in Zhejiang Provincial People’s Hospital and the First Affiliated Hospital of Soochow University were used as the external validation set（n=81,33 cases with VPI-positive and 48 cases with VPI-negative）.Univariate analysis was used to compare the differences in CT features between the two groups,and multivariate logistic regression analysis was performed to select the best combination of CT features related to VPI to construct a clinical model.A total of 1218 radiomics features were extracted from the VOI of total tumor volume（GTV）and tumor extension of 5 mm,10 mm and 15 mm from non-enhanced CT images to construct radiomics models,named GTV,GPTV₅,GPTV₁₀ and GPTV₁₅ radiomics models.The Radscore of the optimal radiomics model and related CT features were used to construct a combined model and a nomogram was developed based on multivariate logistic regression.The Delong test was used to assess the difference in AUC values between the different models.The goodness-of-fit of the model was evaluated by the Hosmer-Lemesow test and calibration curve.The practical application value of this model was analyzed by decision curve analysis（DCA）.【Results】The solid component size,pleural indentation sign,whether solid component contact with pleura and vascular convergence sign in the clinical model were independent predictors of VPI in patients with clinical stage IA lung adenocarcinoma（all P < 0.05）.Based on GTV,GPTV₅,GPTV₁₀ and GPTV₁₅,2,5,7 and 3 optimal radiomics features were selected to construct GTV,GPTV₅,GPTV₁₀ and GPTV₁₅ radiomics models,respectively.The AUC values of GPTV₁₀ radiomics model in the training set,internal validation set and external validation set were 0.855,0.842 and0.842,respectively,which was the best radiomics model.The AUC values of the combined model based on GPTV₁₀-Radscore and CT features in the training set,internal validation set and external validation set were 0.894,0.828 and 0.876,respectively.The prediction performance of the combined model was better than that of the GPTV₁₀ radiomics model in the training set（P < 0.05）,and was better than that of the clinical model in the external validation set（P < 0.05）.No significant difference of the AUC values was found between the three models in the internal validation set（P > 0.05）.The Hosmer-Lemesow test and calibration curve showed that the combined model fitted well in the three cohorts（all P > 0.05）.The DCA curve showed that the combined model achieved more net benefit in predicting VPI status than the clinical model and the GPTV₁₀ radiomics model.【Conclusion】The nomogram based on GPTV₁₀ radiomics features and CT features could effectively predict the VPI status of patients with subpleural clinical stage IA lung adenocarcinoma before surgery.Part 4 The value of radiomics combined with CT features for predicting invasiveness of lung adenocarcinoma presenting as subpleural ground-glass nodule with consolidation-to-tumor ratio ≤ 50%【Objective】To evaluate the value of a nomogram based on radiomics combined with CT features for predicting invasiveness of lung adenocarcinoma presenting as subpleural ground-glass nodule with consolidation-to-tumor ratio（CTR）≤50%.【Methods and Materials】A total of 247 patients diagnosed with clinical stage IA lung adenocarcinoma,ground glass nodules with pleural contact and CTR≤50% in Changzheng Hospital of Naval Medical University were retrospectively collected as an internal data set.According to the pathological types,they were divided into minimally invasive lung adenocarcinoma（MIA）and invasive lung adenocarcinoma（IAC）,they were randomly divided into training set（n= 173,72 cases with MIA and 101 cases with IA）and internal validation set（n=74,36 cases with MIA and 38 cases with IAC）at a ratio of 7:3.47 patients from Zhejiang Provincial People’s Hospital,the Affiliated Hospital of Weifang Medical College and the First Affiliated Hospital of Soochow University were collected as an external validation set（n= 47,14 cases with MIA and 33 cases with IAC）.Univariate analysis and multivariate logistic regression analysis were used to select the best combination of predictive variables in CT features related to invasiveness to construct a clinical model.A total of 1218 radiomics features were extracted from the VOI of total tumor volume（GTV）and tumor extension of 5 mm,10 mm and 15 mm from non-enhanced CT images to construct radiomics models,named GTV,GPTV₅,GPTV₁₀ and GPTV₁₅ radiomics models.The Radscore of the optimal radiomics model and related CT features were used to construct a combined model and a nomogram was developed based on multivariate logistic regression.The Delong test was used to assess the difference in AUC values between the different models.The goodness-of-fit of the model was evaluated by the Hosmer-Lemesow test and calibration curve.The practical application value of this model was analyzed by decision curve analysis（DCA）.【Results】The tomor size,CTR in the clinical model were independent predictors of VPI in patients with clinical stage IA lung adenocarcinoma（all P < 0.05）.Based on GTV,GPTV₅,GPTV₁₀ and GPTV₁₅,12,11,8 and 10 optimal radiomics features were selected to construct GTV,GPTV₅,GPTV₁₀ and GPTV₁₅ radiomics models,respectively.The AUC values of GPTV₁₀ radiomics model in the training set,internal validation set and external validation set were 0.910,0.870 and 0.887,respectively,which was the best radiomics model.The AUC values of the combined model based on GPTV₁₀-Radscore and CT features in the training set,internal validation set and external validation set were 0.912,0.874 and 0.887,respectively.In the training set,the predictive efficacy of the combined model was better than that of the clinical model（P < 0.05）,and the prediction performance of the combined model and GPTV₁₀ radiomics model was better than that of the clinical model in the external validation set（P < 0.05）.No significant difference of the AUC values was found between the three models in the internal validation set（P > 0.05）.The Hosmer-Lemesow test and calibration curve showed that the combined model fitted well in the three cohorts（all P > 0.05）.The DCA curve showed that the combined model achieved more net benefit than the clinical model and the GPTV₁₀ radiomics model in predicting invasiveness of lung adenocarcinoma presenting as subpleural ground-glass nodule with CTR≤50%.【Conclusion】The nomogram based on GPTV₁₀ radiomics features,tumor size and CTR could effectively predict invasiveness of lung adenocarcinoma presenting as subpleural ground-glass nodule with consolidation-to-tumor ratio ≤50% before surgery.