Effects of Image Characteristics on Performance of Tumor Delineation Methods: A Test–Retest Assessment

Abstract

PET can be used to monitor response during chemotherapy and assess biologic target volumes for radiotherapy. Previous simulation studies have shown that the performance of various automatic or semiautomatic tumor delineation methods depends on image characteristics. The purpose of this study was to assess test-retest variability of tumor delineation methods, with emphasis on the effects of several image characteristics (e.g., resolution and contrast). Methods: Baseline test-retest data from 19 non-small cell lung cancer patients were obtained using 18 F-FDG (n 5 10) and 39-deoxy-39-18 F-fluorothymidine ( 18 F-FLT) (n 5 9). Images were reconstructed with varying spatial resolution and contrast. Six different types of tumor delineation methods, based on various thresholds or on a gradient, were applied to all datasets. Test-retest variability of metabolic volume and standardized uptake value (SUV) was determined. Results: For both tracers, size of metabolic volume and testretest variability of both metabolic volume and SUV were affected by the image characteristics and tumor delineation method used. The median volume test-retest variability ranged from 8.3% to 23% and from 7.4% to 29% for 18 F-FDG and 18 F-FLT, respectively. For all image characteristics studied, larger differences (#10-fold higher) were seen in test-retest variability of metabolic volume than in SUV. Conclusion: Test-retest variability of both metabolic volume and SUV varied with tumor delineation method, radiotracer, and image characteristics. The results indicate that a careful optimization of imaging and delineation method parameters is needed when metabolic volume is used, for example, as a response assessment parameter.

On the Reliability of Automatic Volume Delineation in Low-Contrast [18F]FMISO-PET Imaging

“…The chosen threshold and resulting final delineation depends also on the individual investigator and applied display window settings (Nestle et al 2005). A similar effect of image characteristics to delineations was observed for automatic segmentation algorithms (Cheebsumon and van Velden 2011;Cheebsumon and Yaqub 2011). Furthermore, small changes on the threshold may in one case result in contour changes which are hard to see but in other cases may lead to dramatic volume differences.…”

Repeatability of 18F-FDG Uptake Measurements in Tumors: A Metaanalysis

“…These inclusion criteria were mandatory because variation in these parameters can result in parameter differences of up to 50%, representing only noise and not a true biologic change (27,28). Although interstudy differences in image contrast and resolution remained, their effects on repeatability were negligible (29). Whether this is also true for newer systems with better spatial resolution cannot be predicted on the basis of the current data.…”

18F-FCho PET and MRI for the prediction of response in glioblastoma patients according to the RANO criteria

“…Also, therapy response is assessed at different time-points in different studies, which may also (partly) explain the different threshold values. In addition, cut-off values are method specific because they are affected by acquisition parameters, the choice of reconstruction algorithm, and region of interest definition [31][32][33]. All these factors may explain the often large differences between thresholds and underline the importance of a validation of the proposed thresholds, ideally by histology.…”