Room: Exhibit Hall | Forum 1
Purpose: Radiotherapy is a common treatment modality for non-small cell lung cancer (NSCLC) patients. However, radiotherapy may induce radiation pneumonitis (RP). Published studies have demonstrated dose-volume parameters associated with RP, but report large standard deviations of the tallied dosimetric variables with no clear distinction between RP and non-RP groups. This study investigates CT textures as potential predictors of RP incidences.
Methods: 76 consecutive NSCLC patients who underwent radiotherapy in our institution were included. Treatment planning CT with dose for each patient was analyzed. For quantitative imaging purposes planning target volume, segmented by a radiation oncologist, was subtracted from the total (both) lung volume. From this structure 92 geometric, first-, second-, and third-order texture features were extracted. In addition, an in-house software modified the derived features by dose-weighting on voxel-by-voxel basis. It resulted in extraction of 65 features. RP grade was dichotomized (grade 0 vs 1-3) in the statistical analyses. A univariate logistic regression model was used to estimate odds ratio (OR) as well as the corresponding 95% confidence interval and p-value for each imaging feature â€“ with and without dose weighting.
Results: In the conventional texture feature analyses only 2 histogram-based (out of 92) features showed significant associations with the dichotomized RP grade (two-tailed p-values < 0.05). Those two features however exhibited substantial effect, with OR > 1.8. In the dose-weighted analyses 8 (out of 65) features associated with RP (p-values ranging from 0.018 to 0.045), with effects ranging from 0.27 to 2.1.
Conclusion: The presented results suggest that pre-treatment CT textures may help predict the development of RP for patients with advanced NSCLC undergoing radiotherapy. Future directions include harmonizing CT data, expansion of the database, and validation through multivariate analyses.