Rank-shaping regularization of exponential spectral analysis for application to functional parametric mapping.

摘要: Compartmental models are widely used for the mathematical modelling of dynamic studies acquired with positron emission tomography (PET). The numerical problem involves estimation a sum decaying real exponentials convolved an input function. In exponential spectral analysis (SA), nonlinear functions is replaced by linear coefficients predefined set basis functions. This set-up guarantees fast and attainment global optimum. SA, however, hampered high sensitivity to noise and, because positivity constraints implemented in algorithm, cannot be extended reference region modelling. this paper, SA limitations addressed new rank-shaping (RS) estimator that defines appropriate regularization over unconstrained least-squares solution obtained through singular value decomposition base. Shrinkage parameters conditioned on expected signal-to-noise ratio. Through application simulated datasets, it shown RS ameliorates extends properties case production functional parametric maps from PET studies.