Impact of kinetic model design on the analysis of dynamic [15O]O2 studies

Introduction: Various kinetic models have been proposed for calculating oxygen extraction fraction (OEF) and consumption (CMRO2) assuming that, in tissue, oxygen is instantaneously metabolised into water [1,2] or allowing for different washout rates for oxygen and water [3,4]. The purpose of this study was to evaluate performance of different kinetic models for analysing dynamic [15O]O2 studies. Methods: Eleven studies consisting of a 3 min [15O]CO scan and 10 min dynamic [15O]H2O and [15O]O2 (bolus inhalation) scans were performed. Scans were acquired in 3D mode using an HR+ PET scanner. On-line continuous arterial blood sampling was performed to derive an input function. Data were analysed using models specified in Table 1. Type I models used the assumption of instantaneous metabolism of oxygen [1,2]. In its simplest form CBV was fixed to the value obtained from the CO scan, and CBF and Vd to those from the H2O scan. Variations of this model used CBV, CBF and/or Vd as fit parameters. Model II consisted of two parallel single tissue compartments, allowing for different washout rates for oxygen and water [3,4]. Model III contained a two-tissue compartment model for oxygen and a single tissue compartment model for water. Model performance was evaluated by goodness of fit using weighted residual squared errors (WRSE) and Akaike criterion (AIC), and by regression analysis of OEF and CMRO2. Results: and discussion: According to both WRSE and AIC, models II and III provided 'better' fits than type I models. For type I models use of Vd as fit parameter provided better fits than reusing it from the [15O]H2O scan, consistent with preference for models II and III that use separate washout rates for oxygen. Use of Vb as fit parameter provided more accurate fits than use of CBV obtained from [15O]CO scans. Regression analysis showed a fair to good correlation of OEF (coefficient = 0.67 to 0.73) and CMRO2 (coefficient=0.77 to 0.80) for most models compared with model 1A, i.e. the model that has been used routinely in the past. Models allowing for different oxygen washout rates, however, provided on average 5 to 20% lower OEF and CMRO2. Conclusions: Models II and III, which are not based on the assumption of instantaneous metabolism of oxygen into water in tissue, provide more accurate fits of dynamic [15O]O2 studies than conventional (type I) models, which is consistent with experimental findings of Seki et al. [3]. OEF and CMRO2 values may, on average, differ up to 20%, depending on the actual model being used