The Application Value Of CT Image Texture Analysis In The Differentiation Of Peripheral Small Cell Lung Cancer And Peripheral Non-small Cell Lung Cancer

Objective: To explore the feasibility of MaZda CT image texture analysis in the diagnosis and differentiation of peripheral small cell lung cancer and peripheral non-small cell lung cancer and analyze its clinical application value in the diagnosis of both.Methods:1?A retrospective study was conducted on 89 cases of peripheral lung cancer confirmed by surgery or biopsy,including 58 cases of peripheral small cell lung cancer(P-SCLC)and 31 cases of peripheral non-small cell cancer(P-NSCLC)(16 cases of adenocarcinoma;15 cases of adenosquamous carcinoma);2?The reign of interest(ROI)was manually delineated at the maximum tumor cross section using MaZda 4.6 software,and texture features were extracted;3?Fisher coefficient,MI(mutual information),POE+ACC(classification error probability combined with average correlation coefficients)and FPM [Fisher+MI+(POE+ACC)]were used to reduce the dimensionality of the texture features extracted from the two groups of disease images.Then,the texture features extracted by each method were classified and discriminant analyzed by using the B11 tool;4?Statistical analysis:The measurement data of this study with the(?±s)said.Independent sample t test or mann-whitney U test were used to compare the differences of texture features between the two groups.Then Logistic regression analysis was used to carry out multivariate analysis on texture parameters with significant differences,and to screen the main predictors for the differential diagnosis of P-SCLC and P-NSCLC.Statistical software MedCalc19.1 was used to establish the receiver operating characteristic curve(ROC)for each texture feature with statistically significant difference and calculate the area under the curve(AUC).The differential diagnosis efficacy of each predictor and its combination on P-SCLC and P-NSCLC was analyzed.The threshold value was determined according to the maximum correct index,and the specificity and sensitivity of texture parameters and their different combinations were evaluated to differentiate the P-SCLC from the P-NSCLC,and the differences of AUC of the dominant feature parameters and their different combinations were compared and analyzed.All the statistical results of this experiment were statistically significant with p< 0.05.Results: 1?General results: The P-SCLC group: 50 males and 8 females,aged 63.21±10.48 years,maximum lesion diameter was 3.64±1.75 cm.P-NSCLC: 16 males and 15 females,aged 64.23±8.62 years,and the maximum lesion diameter was 2.89±1.41 cm.There was no statistically significant difference in age and maximum tumor diameter between P-SCLC and P-NSCLC groups(p > 0.05).The ratio of male to female in this study was 66:23,among which 75.8%(50/66)were males with P-SCLC.Combined with chi-square test results(p< 0.05),it suggested that men had a higher risk of P-SCLC.2?Different dimension reduction method processing results: All texture feature parameters in this study were processed by four dimensionality reduction methods,and a total of 32 texture features were screened out.Among them,the S(4,4)Correlat?[S(4,0),S(3,0)?S(2,-2),S(1,-1)]SumVarnc ?[S(3,0),S(0,2),S(2,0),S(1,-1)?S(0,1)?S(1,0)]SumAverg of P-SCLC group were all lower than that of the P-NSCLC group,while the S(4,4)DifVarnc and S(4,4)Contrast of P-SCLC were higher than that of P-NSCLC,with statistical significance(p < 0.05).The remaining 17 texture parameters had no statistical significance.3?Results of texture feature classification: the error rate of NDA/ANN classification discriminant analysis was the lowest in each group.The error rate of FPM combined with NDA/ANN is the lowest among all the dimension reduction and classification methods.4?Evaluation of diagnostic efficacy of discriminating texture features: there are 13 texture parameters with discriminating diagnostic significance,and they all belong to the feature of gray-level co-occurrence matrix.The diagnostic efficiency of each single texture feature is general,and the AUC is between 0.6 and 0.7.Comparatively speaking,the diagnostic efficiency of S(4,4)DifVarnc is higher(AUC: 0.696;95% CI: 0.589~0.789),the sensitivity and specificity were 67.24% and 70.97%.5?Logistic regression analysis: S(4,4)DifVarnc,S(2,-2)SumVarnc and S(1,-1)SumAverg were independent predictors of differential diagnosis between P-SCLC and P-NSCLC(p < 0.05).In the differential diagnosis,S(4,4)DifVarnc was positively correlated with the diagnosis of P-SCLC;S(2,-2)SumVarnc and S(1,-1)SumAverg were negatively correlated with the diagnosis of P-SCLC.6?Evaluation of diagnostic efficacy of dominant texture features and their different combinations:S(4,4)DifVarnc,S(2,-2)SumVarnc and S(1,-1)SumAverg were the main predictors for differentiating P-SCLC from P-NSCLC.(1)the single diagnostic efficiency of the three is general(AUC ? 0.6-0.7).Among the three dominant texture parameters,the AUC of S(4,4)DifVarnc is relatively the highest,and its sensitivity and specificity are the best among the three.(2)after different combinations of the three dominant texture parameters,the diagnostic efficiency was improved(AUC>0.7),among which the combination of the three had the largest AUC and higher specificity(AUC=0.767,the sensitivity was 56.90% and the specificity was 87.10%).7?Comparison of AUC between dominant texture features and their different combinations: When comparing the AUC of S(1,-1)SumAverg and the three combinations,z=2.063,p < 0.05,but their 95%CI had overlapping areas,so,it cannot be said that the difference in AUC between the union of three dominant textures and S(1,-1)SumAverg was statistically significant.However,the comparison of AUC between the remaining dominant texture features and the different combinations were not statistically significant(p > 0.05).Conclusion:1?Based on the texture analysis of CT plain scan image,multiple texture parameters from gray-level co-occurrence matrix have statistical significance for the identification of P-SCLC and P-NSCLC,and they have certain diagnostic efficacy for the identification of both,which can provide reliable quantitative information for clinical diagnosis.2 ? S(4,4)DifVarnc,S(2,-2)SumVarnc,and S(1,-1)SumAverg are the main predictors in differentiating P-SCLC from P-NSCLC,and have potential clinical application value.