Program Information

CT Texture Analysis of Non-Small Cell Lung Carcinoma Is Not Predictive of SNP Status

J Yang*, T Xu, S Tucker, R Williamson, Z Liao, L Court, The University of Texas MD Anderson Cancer Center, Houston, TX

SU-E-T-280 Sunday 3:00PM - 6:00PM Room: Exhibit Hall

Purpose: Several single-nucleotide polymorphisms (SNPs) have been shown to be associated with overall survival in patients with non-small cell lung cancer. Also, CT texture analysis of the tumor has shown that tumor heterogeneity is associated with overall survival. Here we investigate whether there is any association between texture features and SNP status.

Methods: 139 NSCLC patients were identified who had been treated for chemo-radiation therapy, and for whom both SNP data and 4DCT simulation images with consistent imaging parameters were available. 14 geometric and texture features, selected from a previous reproducibility/redundancy study, were extracted from the physician-drawn target on the exhale-phase CT images. The data were randomly divided into a training set consisting of 2/3 of the patients (n=93), and a test set consisting of the remaining 50 cases. For each SNP and texture feature, the Kruskal-Wallis test was used to investigate whether there were differences in the texture features between patients of different genotypes. To get an estimate of how many false positives would be expected per SNP, permutation tests were performed, where the SNP genotypes were randomly permuted among the patients and the Kruskal-Wallis tests were repeated. This process was repeated 500 times.

Results: 6 comparisons met the nominal P<0.05 in the training set, but none of these were confirmed in the test data, indicating that there are no association between texture features and SNPs. When the analysis was repeated with the entire dataset, 9 comparisons were significant at the P<0.05 level, but permutation analysis indicated that 8.1 comparisons would have P<0.05 by chance.

Conclusion: There is no strong evidence for an association between any of the 12 SNPs and 14 texture features investigated here. Our next step is to investigate whether combining SNPs and CT textures has a better predictive capability than using SNPs only.

Contact Email: