The impact of optimal respiratory gating and image noise on evaluation of intra-tumor heterogeneity in 18F-FDG positron emission tomography imaging of lung cancer.

Abstract

Assessment of measurement accuracy of intra-tumor heterogeneity using texture features in positron emission tomography (PET) images is essential to characterize cancer lesions with high precision. In this study, we investigated the influence of respiratory motion and varying noise levels on quantification of texture features in patients with lung cancer.Respiratory gating was performed on list-mode data of 60 lung cancer patients, who underwent (18)F-fluorodeoxyglucose (FDG) PET, using an optimal respiratory gating algorithm (ORG). The ORG images were reconstructed using a duty cycle (percentage of the total acquired PET data) of 35%. In addition to ORG images, non-gated images with varying statistical quality (using 35% and 100% of PET data) were reconstructed to investigate the effects of image noise. Several global image-derived indices and texture parameters (entropy, high intensity emphasis (HIE), zone percentage (ZP), and dissimilarity) that have been associated with patient outcome, were calculated. Clinical impact of ORG and image noise on assessment of intra-tumor heterogeneity was evaluated using Cox regression models with overall survival (OS) as outcome measure for non-small cell lung cancer patients. Threshold for statistical significance was adjusted for multiple comparisons using Bonferroni.Respiratory motion significantly affected intra-tumor heterogeneity quantification for lesions in the lower lung lobes (p<0.007), with the exception of entropy (p>0.007). The mean increase of entropy, dissimilarity, ZP, and HIE, for lower lobe lesions was 1.3±1.5% (P = 0.02), 11.6±11.8% (P = 0.006) 2.3±2.2% (P = 0.002), and 16.8%±17.2% (P = 0.006), respectively. No significant differences were observed for lesions located in the upper lung lobes (p>0.007). Differences in the statistical quality of the PET images affected the texture parameters to a lesser extent than respiratory motion, with no statistically significant differences between the images. The median follow-up time of this patient cohort was 35 months (range 7 - 39 months). In multivariate analysis for OS, total lesion glycolysis (TLG) and HIE were the two most relevant image-derived indices considered to be independent significant co-variates for the model, regardless of the image type considered.The results of this study suggest that the tested textural features are robust in the presence of respiratory motion artefacts and varying levels of image noise.