Early Lung Adenocarcinoma: The Correlation Between Quantitative Measurements And Radiomics Features Of Low-Dose Computed Tomography(LDCT)and Pathology

Part one Early Lung Adenocarcinoma: the Correlation betweenQuantitative Measurements of LDCT and PathologyObjective: To investigate the correlation between the quantitative measurements of LDCT of subsolid nodules(SSNs)and pathology of early lung adenocarcinoma.Methods: Retrospectively analysis the images of LDCT of the patients with SSNs in our hospital from January 2015 to December 2017.A total of242 patients with 254 SSNs were included.All patients had 1.0mm axial images of lung window.The maximum diameter,vertical diameter,volume,mean CT value and peak CT value of SSNs were obtained by using a semi-automatic measurement software.Calculate mean diameter and mass of the nodules.According to the pathology and the mode of operation,the SSNs were divided into two groups.One included Adenocarcinoma In Situ(AIS)and Minimally Invasive Adenocarcinoma(MIA),another included invasive pulmonary adenocarcinoma(IPA).Compare these indicators between the two groups.All statistical analyses were performed using SPSS 21.0.The measurement data were analyzed by using two independent samples T test if satisfy the normal distribution,otherwise by Mann-Whitney U test.Count data were compared using chi square test.Receiver operating characteristic(ROC)curve analysis was performed to evaluate diagnostic test.P<0.05 was considered statistically significant differences.The Spearman correlation coefficient(rs)was used to evaluate the consistency between the two observers,and the correlation was significant when P < 0.01.Results: The maximum diameter,mean diameter,mean CT value,peak CT value,volume and mass of SSNs of group B were higher than those ofgroup A(P<0.001).The mean CT value(AUC,0.785)was superior to the peak CT value(AUC,0.748)in differential diagnosis of group A and group B.Combined with all indicators,the best index to distinguish between group A and group B was mass(AUC,0.860),which was better than volume(AUC,0.816)and mean CT value(AUC,0.785).Spearman analysis showed that the rs of mean CT value(0.922)was higher than that of peak CT value(0.915),while in all indexes,the correlation of mass(rs=0.967)was best,which was higher than volume(rs=0.953)and mean CT value(rs=0.922).Part two Radiomics Classification of Early Lung Adenocarcinoma ofSSNs on LDCTObjective: To develop and validate a useful radiomics model for preoperative classification of early lung adenocarcinoma with SSNs on LDCT scan.Methods: A total of 196 patients(203 nodules)with pathology-proven early lung adenocarcinoma and complete clinical data were obtained from January 2015 to December 2017 in our hospital,which each of the adenocarcinomas was showing SSNs on chest thin-section(1.0mm axial image)CT images.The 203 SSNs were classified into two pathology groups,including 111 nodules with adenocarcinoma in situ(AIS)or minimally invasive adenocarcinoma(MIA)in Group A and 92 nodules with invasive adenocarcinoma in Group B.Contour of each SSN was drawn manually by a radiologist on all nodule-containing axial slices in lung window.The high throughput features from the region of interests(ROIs)within the radiologist-drawn contour were extracted for classification analysis by use of a radiomics software.In primary cohort,T-test was used to reduce the dimension of features of the first 141 SSNs.The classification model was established by KNN algorithm and 3 times 10-fold cross validation.The accuracy,sensitivity,specificity,positive predictive value(PPV),and negative predictive value(NPV)in classification of pathology Group A and Group B were calculated.The permutation test was applied to determine whether theaccuracy was significantly higher than values expected by chance.In validation cohort,the remaining 62 SSNs were brought into the model for verification.Blind diagnoses that classify the 62 SSNs in validation cohort into two pathology groups on thin-section LDCT images,was carried out by two radiologists with 5 and 10 years of experience.Diagnostic performance in group classification by the radiomics model and radiologists were evaluated by use of Receiver operating characteristic(ROC)analysis as well as accuracy,sensitivity,specificity,PPV and NPV.The diagnostic performance by the radiologist was compared with the radiomics classification model.Results: Classification results of the model with 3 times 10-fold cross validation showed the AUC of 0.87,accuracy of 79.9%,sensitivity of 82.5%,specificity of 78.3%,PPV of 81.1%,NPV of 79.2%,respectively.Independent verification results showed AUC of 0.92,accuracy of 83.9%,sensitivity of94.1%,specificity of 71.4%,PPV of 80.0%,NPV of 90.9%,respectively.Two blinded radiologists had AUC of 0.89 and 0.92(similar to the model)and accuracy of 74.2% and 77.4%(both slightly lower than the model).Conclusion:1.The quantitative measurement indexs may help to distinguish non-/minimally invasive adenocarcinoma from invasive adenocarcinoma,among them,the mean CT value was better than the peak CT value.However,mass was better than the mean CT value and volume,therefore,the best quantitative index to distinguish non-/minimally invasive adenocarcinoma and invasive adenocarcinoma was mass.2.Radiomics model was full of importance in classification of early lung adenocarcinoma presented as SSNs on LDCT images and has the potential to improve patient management in their treatment plan.