Automatic definition of image-derived input functions in brain PET images using clustering based on population arterial blood curves Mark Lubberink, J.E.M. Mourik, R. Boellaard, A.A. Lammertsma VU University Medical Centre, Amsterdam, The Netherlands
Introduction: Derivation of input functions directly from dynamic brain PET images would obviate the need for arterial sampling and increase the applicability of quantitative brain PET. The aim of the present study was to evaluate definition of image-derived input functions (IDIF) using clustering based on an arbitrary set of measured arterial blood curves. Methods: PET and blood sampler data from ten dynamic (R)-[11C]verapamil scans, acquired using an ECAT Exact HR + scanner, were used in the present study. IDIF were determined using either the whole brain or the brain below the base of the skull only. A voxel was assigned to blood if it was included in the blood cluster for at least 5 out of 10 single K-means clusterings with arbitrarily measured verapamil brain and blood time – activity curves (TAC) of different subjects as starting points. IDIF were calculated as the mean of the TAC of all blood voxels. Partial volume correction (PVC) was performed per plane with the background determined by a 1.5 cm wide region at a distance of 0.75 cm around the blood voxels. Integrals of sampler curves and IDIF were compared. Whole brain volume of distribution (Vd) was calculated using Logan analysis. Correlation coefficients between Vd values based on sampler curves and IDIF were calculated. Results: Fig. 1 shows correlations between IDIF-based Vd values without PVC (r 2 = 0.79 using whole brain) and after PVC of the peak only (r 2 = 0.86) versus those based on sampler curves for 10 subjects. Integrals of PVC IDIF were 10% lower than those of sampler curves. Conclusion: There is a good correlation between (R)-[11C]verapamil Vd values based on blood sampler input curves and IDIF without PVC or with peak-only PVC, especially when using the whole brain to determine the IDIF. Patient movement is a major limitation for the automatic definition of IDIF resulting in underestimation of the blood vessel size and poor performance of PVC. Further investigations are needed to improve PVC, to correct for patient movement, to assess the applicability of this method for other tracers, and to compare it with other methods for defining blood voxels.