Nonlinear function estimation and other numerical methods for imaging myocardial perfusion by N-13 ammonia PET

We study three numerical methods for parametric imaging of myocardial perfusion by N-13 ammonia positron emission tomography (PET): weighted nonlinear regression (WNLR), nonlinear function estimation (NFE), and Patlak analysis. NFE is a set of empirical tools that learn nonlinear input-output mappings. Using Bayesian models of PET data, we trained a sigmoidal network, a tool for NFE, to produce MAP estimates of resting perfusion (0.1-1.5 ml/min/g). We compared the three methods with canine data (n=4) and simulation (m=2000). The simulation data show that NFE is the most accurate method The canine data show that NFE produces parametric images similar to WNLR but two orders of magnitude faster. NFE requires 5 seconds to produce a 128×128 image of perfusion, whereas WNLR requires 2 hours. We conclude that NFE is a fast method for parametric imaging of myocardial pefusion