Anatomy-assisted direct parametric PET imaging for myocardial blood flow abnormality detection

Dynamic myocardial perfusion (MP) PET imaging followed by tracer kinetic modeling provides quantitative measurement of myocardial blood flow (MBF). The purpose of this study is to incorporate anatomical information in the 4D direct parametric image reconstruction and to evaluate the performance in detecting regional MBF abnormality. The one-tissue compartment model was formulated in the maximum likelihood (ML) problem to relate the dynamic projection datasets directly to the kinetic parameters. A maximum a posteriori (MAP) algorithm that incorporates the joint entropy (JE) between the anatomic and parametric images in the reconstruction was developed. The preconditioned steepest ascent (PSA) algorithm was used to solve the ML and the JE-MAP estimation problems. Using the XCAT phantom and the patient-based organ time activity curves, we simulated two sets of dynamic MP Rb-82 PET data, one carrying normal MBF and the other with reduced MBF on a region of interest, each with 20 noise realizations. Corresponding MR images were simulated with the 3D T1-weighted sequence as specified in a clinical PET/MRI protocol. The reconstructed parametric images from the ML and the JE-MAP algorithms were compared using the tradeoff between noise and bias and the signal to noise ratio (SNR), which reflects the separability between the normal and abnormal K1 parameters. The proposed JE-MAP algorithm resulted in improved noise versus bias tradeoff compared to the ML algorithm and also demonstrated better performance in the regional abnormal MBF detection task.