Program Information
Volume-Dependence of Quantitative Imaging Features From CT and CE-CT Images of NSCLC
X Fave1,2*, D Fried1,2, L Zhang1, J Yang1, P Balter1, D Followill1, D Gomez1, A Jones1, F Stingo1, L Court1, (1) UT MD Anderson Cancer Center, Houston, TX, (2) UT Health Science Center Graduate School of Biomedical Sciences, Houston, TX
Presentations
SU-E-J-242 (Sunday, July 12, 2015) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose:To determine whether tumor volume plays a significant role in the values obtained for texture features when they are extracted from computed tomography (CT) images of non-small cell lung cancer (NSCLC). We also sought to identify whether features can be reliably measured at all volumes or if a minimum volume threshold should be recommended.
Methods: Eleven features were measured on 40 CT and 32 contrast-enhanced CT (CE-CT) patient images for this study. Features were selected for their prognostic/diagnostic value in previous publications. Direct correlations between these textures and volume were evaluated using the Spearman correlation coefficient. Any texture that demonstrated a strong correlation had its equation normalized to account for volume, values recalculated, and a new Spearman correlation coefficient computed. To determine if a volume threshold should be recommended, the Wilcoxon rank-sum test was used to compare the variation above and below a volume cutoff. Four different volume thresholds (5, 10, 15, and 20 cm³) were tested.
Results:Four textures were found to be significantly correlated with volume in both the CT and CE-CT images. These were busyness, coarseness, gray-level nonuniformity, and run-length nonuniformity with correlation coefficients of 0.92, -0.96, 0.94, and 0.98 for the CT images and 0.95, -0.97, 0.98, and 0.98 for the CE-CT images. After volume normalization, the correlation coefficients decreased substantially. For the data obtained from the CT images, the results of the Wilcoxon rank-sum test were significant when volume thresholds of 5-15 cm3 were used. No volume threshold was shown to be significant for the CE-CT data.
Conclusion:Equations for four features that have been used in several published studies were found to be volume-dependent. Future studies should consider implementing normalization factors or removing these features entirely to prevent this potential source of redundancy or bias.
Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by National Cancer Institute grant R03CA178495-01. Xenia Fave is a recipient of the American Association of Physicists in Medicine Graduate Fellowship.
Contact Email: