Image-derived input function for brain PET quantification

Abstract

Quantification of positron emission tomography (PET) images using compartmental models requires the estimation of the tracer concentration in plasma as a function of time. Estimating this function directly from PET images, by measuring the tracer concentration on the carotid arteries, is an attractive alternative to the invasive gold-standard method of arterial cannulation of the radial artery. Nevertheless, most image-derived input function methods still rely on extracting a small number of blood samples to correct for partial volume effects, metabolites and plasma fraction (blood-based). In this work, we assess two non-invasive image-derived input function techniques (blood-free). The two blood-free methods were first applied to a computational phantom and compared with a well established blood-based method. Using image-derived input functions, parametric maps of the binding potential were obtained for [11C]-Raclopride PET images from ongoing studies. These were in turn compared to maps that had been obtained using a reference-region based quantification approach. Although good quantification estimates were found for some subjects, it was hard to guarantee consistency. The biggest obstacle seems to be an underestimation of the spill-out effects, which can be minimized using a small number of venous blood samples.