Study Of Lung Adenocarcinoma With Different Gene Types Based On Computer Tomography Texture Analysis

Objective To evaluate the efficiency of computer tomography texture analysis(CTTA)technology,a retrospective analysis of non-contrast computer tomography(CT)images and contrast-enhanced CT images of lung adenocarcinoma(LAC)was performed.To explore the identification and predictive ability of this technique for the wild-type LAC and epidermal growth factor receptor(EGFR)mutation LAC.Methods Consecutive cases of EGFR testing was selected from 2014.01 to 2016.12.After excluding the invalid cases,the basic information and CT images were collected.Image Segmenter was used to segment the CT Images.21 texture features were derived from the image after segmentation,the features were analyzed and the prediction model was established.Categorical variables were analyzed by chi-square tests.Continuous variables were analyzed by Kolmogorov-Smirnov test first.If the data followed a normal distribution,the t-test,ANOVA or Welch’s ANOVA was performed as appropriate.Otherwise,Kruskal-Wallis test was conducted.To establish a reliable diagnostic model,the presence of multicollinearity among the data was first examined.A prediction tool for EGFR mutation subtype was devised from the principal component analysis.We used the prediction data to draw receiver operating characteristic(ROC)curves,and we calculated the area under the curve(AUC).The Youden index was used to derive the optimum cut-off values.The difference of diagnostic efficacy between the non-contrast group and the contrast-enhanced group was evaluated.Results For the non-contrast group,statistical differences were found in variance between the wild-type group and the exon 18 mutation groups(p<0.001),the AUC was 0.853,the sensitivity was 70.00%and the specificity was 87.05%,Six texture features were significantly different between the wild-type group and the exon 19 mutation group(p<0.001),the predictive model AUC was 0.702,the sensitivity of the test was 58.44%,and the specificity was 74.82%.Between the wild-type group and the exon 20 mutation group,five texture features were different,the AUC was 0.773,the sensitivity and the specificity was 90.91%and 56.83%,respectivly.We found that six features were significantly different between the wild-type group and the exon 21 mutation group(p<0.001),the AUC was 0.743,the sensitivity was 67.40%,and the specificity was 74.82%.For the contrast-enhanced group,no statistically significant differences were found between two subtypes(exon 18 mutation and exon 20 mutation)and the wild-type group.Three features were significantly different between the wild-type group and the exon 19 mutation group,the AUC was 0.708,the sensitivity and the specificity was 70.79%and 66.69%,respectivly.The results were the same for the exon 21 mutation group as in the exon 19 mutation group,AUC was 0.636,the sensitivity was 54.93%,and the specificity was 73.21%.The predictive ability of exon 19 in the two groups was compared,the p value was 0.914,in the exon 21 group the p value was 0.128.Conclusion Our results suggest that the 21 texture features extracted from this study can be used as a complementary tool to detect EGFR mutations with moderate sensitivity and specificity,or for patients who are clinically undetectable and suspect EGFR mutations.Both Non-contrast and contrast-enhanced CTTA may provide information on the identification of EGFR subtype mutations.Compared to contrast-enhanced scans,non-contrast CTTA may provide more information on the identification of exon 18 and exon 20 mutations.Non-contrast and contrast-enhanced CTTA had the same diagnostic efficacy in exon 19 and exon 21 mutations.