Validation of Image-Derived Input Function for Non-Invasive Kinetic Modeling of [18F]-Fluorodeprenyl-D2 in Neurological Diseases​

Background. [¹⁸F]fluorodeprenyl-D2 ([ ¹⁸ F]F-DED) PET enables in vivo quantification of monoamine oxidase B (MAO-B) expression, a marker of reactive astrogliosis implicated in a range of neurodegenerative diseases. Full quantification of MAO-B expression with PET requires an arterial input function (AIF) derived from continuous blood sampling - an invasive procedure that limits clinical applicability. The aim of this study was to develop and validate a non-invasive, carotid image-derived input function (IDIF) extraction pipeline for [¹⁸F]F-DED PET and to assess its agreement with the gold-standard AIF at the regional and voxel level. Eight subjects underwent 60-minute dynamic [¹⁸F]F-DED PET/CT with concurrent arterial blood sampling and radiometabolite analysis. An IDIF was manually extracted from the internal carotid arteries and compared with the AIF. To improve the agreement of IDIF and AIF, a population-based, time-decaying peak-recovery correction was applied, and both input functions then received the same population-derived plasma-to-whole-blood and parent-fraction correction. Using the AIF, five kinetic models were ranked by region-level: one-tissue (1T2K), irreversible (2T3K) and reversible (2T4K) two-tissue compartment models, and Logan and Patlak graphical analyses. Total distribution volume (VT; 1T2K, 2T4K, Logan) and net influx rate constant (Ki; 2T3K, Patlak) were estimated from both input functions, regionally and voxel-wise, and compared with the AIF. Results. The uncalibrated IDIF underestimated the arterial bolus peak by 78.1%, reduced to 18.2% after the population-based peak-recovery calibration was applied. The reversible 2T4K model best described regional tissue kinetics, not supporting purely irreversible trapping. After plasma- and metabolite-correction, IDIF- and AIF-derived regional VT and Ki agreed significantly for all five kinetic models (R² = 0.53–0.77, all p < 0.05), with Logan analysis yielding the most robust results across conditions (R² = 0.62–0.73). Voxel-wise VT maps preserved the AIF spatial pattern, differing only by a near-uniform positive offset (median + 60%). Conclusions. A carotid IDIF, corrected with population-derived plasma-to-whole-blood and parent-fraction curves and a time-decaying peak-recovery calibration, closely approximated the AIF for [¹⁸F]F-DED PET. After calibration, all kinetic approaches agreed with the AIF reference, with Logan graphical analysis providing the most robust regional VT estimates across conditions.