Quantitative Approaches for 18F-FDG PET Imaging in Lymphoma

This thesis focuses on improving the robustness, reproducibility, and clinical relevance of quantitative FDG PET imaging in lymphoma, with particular emphasis on metabolic tumor volume (MTV) assessment in diffuse large B-cell lymphoma (DLBCL) and classical Hodgkin lymphoma (cHL). First, the reproducibility of liver standardized uptake value (SUV), widely used as a reference for visual and quantitative PET interpretation, was systematically investigated. A spherical liver volume of interest of at least 3 cm diameter yielded the most stable SUVmean measurements, supporting and extending current EANM recommendations. However, consistent treatment-related changes in liver uptake were observed, indicating that liver SUV may not be fully stable across treatment phases and should be used with caution in longitudinal response assessment. Second, several semi-automated segmentation methods for baseline MTV quantification were evaluated in both cHL and DLBCL. Although absolute MTV values varied substantially between methods, their prognostic discrimination was largely comparable. SUV4.0 emerged as the most practical and reproducible approach, requiring minimal manual correction while maintaining robust prognostic performance, supporting its suitability for standardized use in multicenter trials. Third, the thesis addressed the specific challenges of MTV delineation at interim and end-of-treatment PET, where reduced lesion uptake and increased heterogeneity impair segmentation reliability. Lesion-specific parameters, particularly SUVpeak and background SUV, were identified as key determinants of segmentation quality. Adaptive, lesion-tailored strategies combining different segmentation methods significantly improved interobserver agreement across treatment phases. A simplified decision-rule approach achieved performance comparable to machine-learning models, enabling feasible clinical implementation. Finally, quantitative PET-based prognostic models were developed to improve outcome prediction at end-of-treatment PET in DLBCL. Models incorporating lesion count and tumor-to-liver uptake ratios substantially outperformed the Deauville score in positive predictive value and markedly reduced false-positive findings, independent of conventional clinical risk factors.