LAUSR

Purpose An open issue of radiomic is its robustness in respect to image acquisition and post-processing parameters changes. The aim of this study is to evaluate the stability of PET radiomic features when decreasing injected activity in pediatric PET/MRI for their potential use for lesion biologic characterization in low-activity examinations. Moreover, different SUV resampling and contouring strategies were investigated. Methods Fifty-six lesions detected in 21 whole-body 18F-FDG-PET/MRI pediatric examinations (3 MBq/kg, 5 min/bed) performed on a clinical PET/MR scanner (Siemens Biograph mMR) were considered. Activity reduction was simulated by truncating the original Full Tracer Activity (FTA) list-mode data correspondingly to tracer activity of 2.4, 1.8, 1.5, 1.2 and 0.6 MBq/kg (Figure). VOIs were defined by two methods: 1) applying a threshold equal to 40% of SUVmax in the FTA image and coping the identified VOIs to low-activity images (Fixed VOIs); 2) thresholding in each image (Variable VOIs). SUV were discretized in two ways: 1) resampling to 64 levels (R64); 2) rounding to 0.25 (R025). Ninety-one features (36 first order and 55 higher-order statistics) were extracted using IBEX (v.1.0 β ) . To assess features stability for each metric, relative changes from FTA images were computed and paired t-tests were performed. Results Differences in features changes at lower activity levels were found between VOI contouring (Fixed VOIs vs Variable VOIs) and SUV resampling (R64 vs R025) methods. Considering Fixed VOIs and R64 resampling, 52/91 features showed non-statistically significant differences for all activity levels. Focusing on metrics as robust as SUVmax and SUVpeak to changes in voxel size and FWHM of the gaussian filter [1] , the effects of decreasing tracer activity are quantitatively variable (Table). Conclusions Selected PET radiomic features can provide reliable information of tumor heterogeneity for low-activity pediatric protocols depending on SUV resampling and lesion delineation strategies.