Abstract
428
Objectives To assess the impact of reconstruction methods on the repeatability of various advanced quantitative tracer uptake imaging features (i.e. radiomics).
Methods Ten lung cancer patients received two baseline whole-body [18F]FDG scans that were acquired on a time-of-flight PET/CT scanner (Philips Healthcare, Cleveland, OH). Test and retest scans were performed within 3 days of each other. Each scan was reconstructed twice using vendor-recommended reconstruction settings: one that used the point-spread function (PSF) and another that complied with the EANM guidelines for tumor PET imaging. Volumes of interest (VOI, n=18) were defined on lesions larger than 10 mL using an isocontour of 50% of the 3D peak SUV corrected for local background. For each VOI, 101 radiomics features were determined that can be divided into first order gray-level statistics (n=22), geometric (n=9) or textural (n=70) features. The latter were based on fractals, gray-level co-occurrence matrices or gray-level run-length matrices.
Results Several radiomics features (56%) showed good test-retest variability (TRT; <15%) and intraclass correlation coefficients (ICC; >0.9) using EANM-compliant or PSF-based images. These features provided a small but significantly better TRT and ICC when extracted from EANM-compliant images than those extracted from PSF-based images (TRT: 6.8% and 7.5%, respectively; ICC: 0.96 and 0.94, respectively; p<0.05). Eighteen features showed more than 3% difference in TRT and/or ICC using EANM-compliant versus PSF-based reconstructions and performed, in general, better when extracted from EANM-compliant images (up to 5% improvement in TRT and up to 14% improvement in ICC).
Conclusions Despite its improved lesion detectability, PSF-based reconstructions do not necessarily result in improved repeatability of quantitative radiomic features when compared to EANM-compliant reconstructions.
Research Support Philips Healthcare