Metabolic imaging techniques for dynamic PET data

Summary form only given, as follows. A variety of sophisticated physiologic models is available for the quantitative interpretation of dynamic positron emission tomography (PET) data. Two techniques for computing metabolic images in terms of such models are discussed: PIXEL carries out a pixelwise parameter optimization; MIXEL incorporates an additive mixture representation to account for effects induced by instrumental and biological blurring. The implementation of these methods makes use of divisive clustering and recursive backwards elimination techniques. Appropriate cross-validation statistics are developed. The methodology is being implemented operationally for a set of PET protocols studying cancer. Examples of metabolic images defined in terms of axially distributed compartmental models have been obtained for studies using labeled thymidine, glucose, and fluoromizonidisole tracers.<>