Evaluation of PET texture features with heterogeneous phantoms: complementarity and effect of motion and segmentation method

Abstract

A major source of error in quantitative PET/CT scans of lung cancer tumors is respiratory motion. Regarding the variability of PET texture features (TF), the impact of respiratory motion has not been properly studied with experimental phantoms. The primary aim of this work was to evaluate the current use of PET texture analysis for heterogeneity characterization in lesions affected by respiratory motion. Twenty-eight heterogeneous lesions were simulated by a mixture of alginate and 18F-fluoro-2-deoxy-D-glucose (FDG). Sixteen respiratory patterns were applied. Firstly, the TF response for different heterogeneous phantoms and its robustness with respect to the segmentation method were calculated. Secondly, the variability for TF derived from PET image with (gated, G-) and without (ungated, U-) motion compensation was analyzed. Finally, TF complementarity was assessed. In the comparison of TF derived from the ideal contour with respect to TF derived from 40%-threshold and adaptive-threshold PET contours, 7/8 TF showed strong linear correlation (LC) (p < 0.001, r > 0.75), despite a significant volume underestimation. Independence of lesion movement (LC in 100% of the combined pairs of movements, p < 0.05) was obtained for 1/8 TF with U-image (width of the volume-activity histogram, WH) and 4/8 TF with G-image (WH and energy (ENG), local-homogeneity (LH) and entropy (ENT), derived from the co-ocurrence matrix). Their variability in terms of the coefficient of variance () resulted in (WH) = 0.18 on the U-image and (WH) = 0.24, (ENG) = 0.15, (LH) = 0.07 and (ENT) = 0.06 on the G-image. Apart from WH (r > 0.9, p < 0.001), not one of these TF has shown LC with Cmax. Complementarity was observed for the TF pairs: ENG–LH, CONT (contrast)–ENT and LH–ENT. In conclusion, the effect of respiratory motion should be taken into account when the heterogeneity of lung cancer is quantified on PET/CT images. Despite inaccurate volume delineation, TF derived from 40% and COA contours could be reliable for their prognostic use. The TF that exhibited simultaneous added value and independence of lesion movement were ENG and ENT computed from the G-image. Their use is therefore recommended for heterogeneity quantification of lesions affected by respiratory motion.