A nonlinear regression technique for the separation of kinetic compartments in dynamic SPECT imaging

The accuracy of SPECT images are compromised when the tracer distribution changes during image acquisition. Image analysis can be further complicated if there is an overlap between the kinetic compartments. For example, Tc-99m Teboroxime is an imaging agent with rapid washout from the myocardium and high liver uptake. The reconstructed myocardium in Tc-99m Teboroxime SPECT may be compromised with artifacts due to changing activity and contaminated by reconstruction artifacts due to the liver. This study proposes a nonlinear regression technique to separate the kinetic compartments, based on the knowledge of time-activity curves (TAC) coupled with dynamic SPECT scans. Dynamic scans (acquired with a rapid fanning acquisition) are reconstructed independently to generate dynamic short-axis images (representing the tracer distribution at different times), which are then transformed to allow each pixel to be evaluated over time independently. The objective function to be minimized is the deviation of the separated compartmental pixels scaled by the TAC plus a distance penalty. This minimization problem is solved subject to reference and nonnegative constraints. The reference constraint is produced from the reconstruction of the summed projections from the dynamic scans (with minimal artifacts due to changing activity). Patient studies of Tc-99m Teboroxime SPECT were used to evaluate this technique. Qualitatively, the kinetic compartments (the liver and heart) were separated producing images with reduced liver contamination in the myocardium. The quantitative accuracy of the separation depends on the accuracy of the tracer kinetics and can also be degraded by physical factors such as photon attenuation and Compton scatter.