| Part ? The value of spectral CT radiomics in the differential diagnosis of lung cancer nodule and inflammatory noduleObjective To explore the application value of spectral CT radiomics quantitative features in differentiating lung cancer nodule from inflammatory nodule.Methods The spectral CT imaging data of 96 lung cancer nodules and 45 inflammatory nodules were analyzed retrospectively.According to a ratio of two to one,patients were randomly assigned to the training group and validation group,including 64 lung cancer nodules and 30 inflammatory nodules in the training group,32 lung cancer nodules and 15 inflammatory nodules in the validation group.MaZda software was used for radiomics analysis in the 70 keV monochromatic images in arterial phase(AP)and venous phase(VP)for the lung cancer nodules and inflammatory nodules in the training group.MaZda extracted features including the gray level histogram(GLH),absolute gradient(GRA),gray-level co-occurrence matrix(GLCM),gray-level run-length matrix(GLRLM),auto-regressive model(ARM)and wavelets transform(WAV),and so on.Fisher coefficients(Fisher),classification error probability combined average correlation coefficients(POE+ACC)and mutual information(MI)were used to select 10 optimal features making up the optimal feature subsets.The optimal feature subsets were analyzed by using linear discriminant analysis(LDA)and nonlinear discriminant analysis(NDA)to calculate the accuracy,sensitivity and specificity in differentiating lung cancer nodule from inflammatory nodule.The prediction model was established using the optimal feature subsets with the highest accuracy in the training group with artificial neural network(ANN).The two sample t test and ROC curve analysis were performed.The established prediction model was used to differentiate lung cancer nodule from inflammatory nodule in the validation group.Results In arterial phase,the optimal feature subset obtained from MI-NDA had the highest accuracy of 87.69%,sensitivity of 87.5%and specificity of 88.0%,in the differential diagnosis of lung cancer nodule and inflammatory nodule in the training group.Area under the curve of the best quantitative parameter S(1,-1)Contrast is 0.663.The threshold of 14.03 had the sensitivity and specificity of 87.5%and 60%,respectively,in differentiating lung cancer nodule from inflammatory nodule.There were no significant differences in sensitivity(P=1.00)but significant differences in specificity(P<0.001)between the optimal feature subsets and the best quantitative parameter S(1,-1)Contrast for differentiating lung cancer nodule from inflammatory nodule.In venous phase,the optimal feature subset obtained from(POE+ACC)-NDA had the highest accuracy of 86.15%,sensitivity of 92.5%and specificity of 76.0%,in the differential diagnosis of lung cancer nodule and inflammatory nodule in the training group.Area under the curve of the best quantitative parameters S(0,1)DifVarnc is 0.688.The threshold of 2.52 had the sensitivity and specificity of 72%and 57.5%,respectively,in differentiating lung cancer nodule from inflammatory nodule.There were significant differences in sensitivity and specificity between the optimal feature subsets and the best quantitative parameter S(0,1)DifVarnc for differentiating lung cancer nodule from inflammatory nodule(P<0.05).The prediction model was established using the optimal feature subset in AP obtained from MI-NDA with ANN.The prediction model had the accuracy,sensitivity and specificity of 80.85%,81.25%and 80.0%,respectively,in the differential diagnosis of lung cancer nodule and inflammatory nodule in the validation group.Concluision Spectral CT radiomics quantitative features can be used to distinguish lung cancer nodule from inflammatory nodule,which has a high diagnostic value.Part ? The value of spectral CT radiomics in the prediction of lymph node metastasis in lung cancerObjective To explore the application value of spectral CT radiomics quantitative features in differentiating metastatic lymph nodes from non-metastatic lymph nodes.Methods The spectral CT imaging data of 102 patients including 52 lung cancers with metastatic lymph nodes and 50 lung cancers with non-metastatic lymph nodes were analyzed retrospectively.A total of 84 metastatic lymph nodes and 60 non-metastatic lymph nodes were included in the study by comparing CT images and pathological results.According to a ratio of two to one,patients were randomly assigned to the training group and validation group,including 56 metastatic lymph nodes and 40 non-metastatic lymph nodes in the training group,28 metastatic lymph nodes and 20 non-metastatic lymph nodes in the validation group.MaZda software was used for radiomics analysis in the 70 keV monochromatic images in arterial phase(AP)and venous phase(VP)for the metastatic and non-metastatic lymph nodes in the training group.MaZda extracted features including the gray level histogram(GLH),absolute gradient(GRA),gray-level co-occurrence matrix(GLCM),gray-level run-length matrix(GLRLM),auto-regressive model(ARM)and wavelets transform(WAV),and so on.Fisher coefficients(Fisher),classification error probability combined average correlation coefficients(POE+ACC)and mutual information(MI)were used to select 10 optimal features making up the optimal feature subsets.The optimal feature subsets were analyzed by using linear discriminant analysis(LDA)and nonlinear discriminant analysis(NDA)to calculate the accuracy,sensitivity and specificity in differentiating metastatic lymph nodes from non-metastatic lymph nodes in lung cancers.The prediction model was established using the optimal feature subsets with the highest accuracy in the training group with artificial neural network(ANN).The two sample t test and ROC curve analysis were performed.The established prediction model was used to differentiate lung cancer metastatic lymph nodes from non-metastatic lymph nodes in the validation group.Results In arterial phase,the optimal feature subset obtained from(POE+ACC)-NDA had the highest accuracy of 97.92%,sensitivity of 96.43%and specificity of 100%,in the differential diagnosis of lung cancer metastatic lymph nodes and non-metastatic lymph nodes in the training group.Area under the curve of the best quantitative parameter Horzl_GLevNonU is 0.980.The threshold of 10.29 had the sensitivity and specificity of 96.4%and 95%,respectively,in differentiating lung cancer metastatic lymph nodes from non-metastatic lymph nodes.There were no significant differences in sensitivity and specificity between the optimal feature subsets and the best quantitative parameter Horzl_GLevNonU for differentiating metastatic lymph nodes from non-metastatic lymph nodes(P>0.05).In venous phase,the optimal feature subset obtained from Fisher-NDA and(POE+ACC)-NDA had the highest accuracy of 93.75%,sensitivity of 92.86%and 89.28%and specificity of 95%and 100%,respectively,in the differential diagnosis of lung cancer metastatic lymph nodes and non-metastatic lymph nodes in the training group.Area under the curve of the best quantitative parameters S(2,2)InvDfMom is 0.941.The threshold of 0.26 had the sensitivity and specificity of 92.9%and 85%,respectively,in differentiating lung cancer metastatic lymph nodes from non-metastatic lymph nodes.There were no significant differences in sensitivity(P=1.00)but significant differences in specificity(P=0.03)between the optimal feature subsets and the best quantitative parameter S(2,2)InvDfMom for differentiating metastatic lymph nodes from non-metastatic lymph nodes.The prediction model was established using the optimal feature subset in AP obtained from(POE+ACC)-NDA with ANN.The prediction model had the accuracy,sensitivity and specificity of 85.42%,85.71%and 85%,respectively,in the differential diagnosis of lung cancer metastatic lymph nodes and non-metastatic lymph nodes in the validation group.Conclusion Spectral CT radiomics quantitative features have high value in the prediction of lymph node metastasis in lung cancer and can be used to distinguish lung cancer metastatic lymph nodes from non-metastatic lymph nodes.Part ? The value of CT radiomics in the prediction of EGFR mutation and EML4-ALK fusion gene expression in lung cancerObjective To explore the value of CT radiomics quantitative features in the prediction of epidermal growth factor receptor(EGFR)mutation and echinoderm microtubule-associated protein-like 4-anaplastic lymphoma kinase(EML4-ALK)fusion gene expression in lung cancer.Methods The data of 144 patients with EGFR gene test results and 162 patients with ALK gene test results in lung cancers were retrospectively analyzed,including 81 patients with EGFR mutations,63 patients with EGFR wild types,72 patients with positive EML4-ALKs and 90 patients with negative EML4-ALKs,According to a ratio of two to one,patients were randomly assigned to the training group and validation group,including 54 EGFR mutant types,42 wild types,48 positive EML4-ALKs and 60 negative EML4-ALKs in the training group;27 EGFR mutant types,21 wild types,24 positive EML4-ALKs and 30 negative EML4-ALKs in the validation group.MaZda software was used for radiomics analysis in the CT images for the patients with EGFR mutant types,EGFR wild types,positive EML4-ALKs and negative EML4-ALKs in the training group.MaZda extracted features including the gray level histogram(GLH),absolute gradient(GRA),gray-level co-occurrence matrix(GLCM),gray-level run-length matrix(GLRLM),auto-regressive model(ARM)and wavelets transform(WAV),and so on.Fisher coefficients(Fisher),classification error probability combined average correlation coefficients(POE+ACC)and mutual information(MI)were used to select 10 optimal features making up the optimal feature subsets.The optimal feature subsets were analyzed by using linear discriminant analysis(LDA)and nonlinear discriminant analysis(NDA)to calculate the accuracy,sensitivity and specificity in the differential diagnosis of EGFR mutant types and EGFR wild types,EML4-ALKs positive group and EML4-ALKs negative group in lung cancers.The prediction model was established using the optimal feature subsets with the highest accuracy in the training group with artificial neural network(ANN).The established prediction model was used to differentiate EGFR mutant types from EGFR wild types,EML4-ALKs positive group from EML4-ALKs negative group in the validation group.Results MaZda software extracted about 600 quantitative features in the CT images for the patients with EGFR mutant types and EGFR wild types,positive EML4-ALKs and negative EML4-ALKs in the training group.The optimal feature subsets obtained from Fisher-NDA and(POE+ACC)-NDA had the highest accuracy of 93.88%,sensitivity of 92.59%and specificity of 95.24%,in the differential diagnosis of the EGFR mutant types and EGFR wild types of lung cancer in the training group.The optimal feature subset prediction model had the accuracy,sensitivity and specificity of 83.33%,85.18%and 80.95%,respectively,in the differential diagnosis of the EGFR mutant types and EGFR wild types of lung cancer in the validation group.The optimal feature subset obtained from MI-NDA had the highest accuracy of 92.59%,sensitivity of 83.33%and specificity of 100%,in the differential diagnosis of the EML4-ALKs positive group and EML4-ALKs negative group of lung cancer in the training group.The optimal feature subset prediction model had the accuracy,sensitivity and specificity of 85.19%,87.5%and 83.33%,respectively,in the differential diagnosis of the EML4-ALKs positive group and EML4-ALKs negative group of lung cancer in the validation group.Conclusion CT radiomics quantitative features have high value in the prediction of EGFR mutation and EML4-ALK fusion gene expression in lung cancer. |
PDF Full Text Request