Application Valur Of CT In Predicting Pathological Features Of Non-small Cell Lung Cancer

PART 1 ANALYSIS OF CLINICOPATHOLOGY AND CT FEATURES IN DIFFERENT HISTOLOGY SUBTYPES OF INVASIVE LUNG ADENOCARCINOMAPurpose To investigate the differences of clinicopathological characteristics and CT features in different histology subtypes of invasive lung adenocarcinoma,and to find a noninvasive imaging method that can help to distinguish them.Methods A total of 422 patients confirmed with solitary invasive lung adenocarcinoma by surgical resection from July 2013 to July 2019 were included.All patients were divided into group A(those with a lepidic/acinar/papillary predominant pattern)and group B(those with a solid/micropapillary predominant pattern)according to the main pathological growth pattern.Moreover,group A was divided into group A1(those with a lepidic predominant pattern)and group A2(those with an acinar/papillary predominant pattern).The clinicopathology and CT features between group A and group B as well as those between group A1 and group A2 were compared with independent sample t-test,Mann Whitney U-test and X~2 test,respectively.Results(1)Comparison of clinicopathological characteristics between group A and B: gender and smoking history of patients,histologic differentiation,lymph node metastasis(LNM)and pathologic TNM stage of tumors differed significantly between the two groups(all P < 0.05).In the present study,patients with male and smoking history and poor histology differentiation of the tumor,LNM,and pathologic TNM stage of III-IV were more frequently seen in group B,while female,non-smoking,well differentiation,no LNM and TNM stage I-II were more common in group A.However,no significant difference was observed in the distant metastasis between the two groups(P > 0.05).(2)Comparison of CT features between group A and B: lesion size,density,vascular convergence sign,air bronchogram,calcification,necrosis and pleural effusion were associated with the histological subtypes(all P <0.05).Tumors in group B(with an average diameter of 34.50 mm)were significantly larger than those in group A.In addition,tumors with solid density,calcification,necrosis and pleural effusion were more frequently observed in group B,whereas those with vascular convergence sign and air bronchogram were more common in group A.However,no significant associations were found between other CT features and histological subtypes,including the distribution of tumors,air space,speculation,lobulation,and pleural retraction(all P > 0.05).(3)Comparison of clinicopathological characteristics and CT features between group A1 and A2: tumors with subsolid density were more common in group A1,while those with solid density were more common in group A2(P < 0.001).However,no significant differences were detected in other clinicopathological characteristics and CT features between A1 and A2 groups(all P > 0.05).Conclusion Different growth patterns of invasive adenocarcinoma have different clinical,pathological and CT features.Correct understanding of these signs can better identify the pathological subtypes of invasive lung adenocarcinoma,and provide more valuable clues for clinical treatment and prognosis evaluation.PART 2 PREDICTION OF EPIDERMAL GROWTH FACTOR RECEPTOR GENE MUTATION IN NON-SMALL CELL LUNG CANCER BY SPECTRAL CT COMBINED WITH CLINICAL AND PATHOLOGICAL FEATURESPurpose To explore the value of clinical-pathological and conventional CT features combined with spectral CT quantitative parameters of non-small cell lung cancer(NSCLC)in evaluating the epidermal growth factor receptor(EGFR)mutation status.Methods A total of 91 NSCLC patients confirmed pathologically underwent EGFR gene detection and spectral CT before treatment were retrospectively analyzed and divided into EGFR mutation-positive group(47 cases,51.65%)and negative group(44 cases,48.35%)based on the gene test results.The clinical-pathological characteristics,conventional CT features and spectral CT quantitative parameters were compared between the two groups.Multivariate Logistic regression model 1 was constructed based on clinical-pathological characteristics and conventional CT features.Additionally,model 2 was constructed with the combination of clinical-pathological characteristics,conventional CT features and spectral CT quantitative parameters.Receiver operating characteristic curve was used to compare the diagnostic efficacy of the two models for the prediction of EGFR mutation.Results(1)Comparison of clinical and pathological features between the two groups: there were significant differences in gender,smoking history and pathological type between the two groups(all P < 0.05).Among them,female,non-smoking and lung adenocarcinoma patients were more frequently observed in EGFR gene mutation positive group.However,there was no significant difference in age and pathological TNM stage between the two groups(all P > 0.05).(2)Comparison of conventional CT signs between the two groups: air space was more common in the positive group,whilst calcification and necrosis were more frequently in the negative group(all P <0.05).There was no significant difference in other CT features between the two groups(P > 0.05).(3)Comparison of quantitative parameters of spectral CT between the two groups: iodine(water)concentration and water(iodine)concentration in arterial and venous phase in mutation positive group were higher than those in mutation negative group(P < 0.05).There was no significant difference in other spectral CT parameters between the two groups(all P >0.05).(4)Multivariate logistic regression model: the prediction efficacy of model 2 was statistically better than that of model 1(AUC: 0.788 vs 0.686,Z=2.606,P = 0.019).Conclusion Combination of clinical-pathological characteristics,conventional CT features and spectral CT quantitative parameters of NSCLC can effectively improve the prediction efficacy of EGFR gene mutations.PART 3 DEVELOPMENT AND VALIDATION OF PREOPERATIVE PREDICTION MODEL FOR LYMPH NODE METASTASIS OF PERIPHERAL LUNG AENOCARCINOMA BASED ON CTPurpose Based on the hypothesis of “seed and soil”,we aimed to develop and validate a combined model of machine learning for predicting lymph node metastasis(LNM)in patients with peripheral lung adenocarcinoma(PLADC).Methods Radiomics models were developed in a primary cohort that consisted of 390 PLADC patients confirmed pathologically from January 2016 to August 2018,and 166 consecutive patients from September 2018 to November 2019 were collected as internal validation cohort to evaluate and validate the combined model.According to the postoperative pathological results,all patients were divided into LNM(-)group and LNM(+)group,and all patients were further divided into N0,N1,N2 and N3 according to the 8th Edition TNM staging of lung cancer.First,the clinical and CT features of all patients were analyzed and compared to screen out the risk factors of LNM.Secondly,the DICOM format plain CT images of all patients were obtained from the image archiving and communication system,and the lesions were segmented manually.The imaging features of primary tumor(R1)and adjacent pleura(R2)were extracted,and the consistency of all features was analyzed by intra-correlation coefficient.Thereafter,the best subset method and the minimum Akike’s information criterion were combined to select all the imaging features.Finally,the machine learning model was constructed with multivariate logistic regression analysis based on the characteristics of R1,R2,CT and clinical risk factors,and the performance of the combined model was evaluated by the receiver operating characteristic curve.Results(1)Comparison of clinical and CT features between the two groups:gender,smoking history,lesion size,lesion density,air bronchogram,spiculation,lobulation,necrosis,pleural effusion,and pleural involvement were significantly different between the LNM(-)group and the LNM(+)group(all P < 0.05).(2)Radiomics features selection and machine learning model construction: a total of 1300 features were extracted from R1 and R2,of which 31 features were significantly correlated with LNM(P < 0.05).The R1 model was built with 13 features,including original first-order variance,wavelet transform,gray histogram features,gradient,and lbp.3D.k glszm small-area emphasis.AUC of training and validation cohorts were 0.847 and 0.859,respectively.The R2 model was built with 19 features,including wavelet,square root,logarithm,and gradient,with AUC of 0.837 and 0.815 for the prediction of LNM in the training cohort and validation cohorts,respectively.Meanwhile,AUC of R1+R2 model were 0.878 and 0.870 in the training and validation cohorts,respectively.(3)Construction and validation of the combined model: in the combined model,R1,R2,tumor size and spiculation were independent predictors of LNM in patients with PLADC,with AUC of 0.897 and 0.883 in the training and validation cohorts,respectively.Radiomic signatures also showed a good performance in identifying the lymph node stage of N0,N1,N2,and N3,with AUC of 0.691-0.927 and 0.700-0.951 in the training and validation cohorts,respectively.Conclusion The "seed soil" hypothesis is also applicable to the prediction of lymph node metastasis in patients with peripheral lung adenocarcinoma.Comprehensive analysis of primary tumor and adjacent pleural information will provide more information for clinical decision-making.