Maximum-Likelihood Estimation for Indicator Dilution Analysis

Abstract

Indicator-dilution methods are widely used by many medical imaging techniques and by dye-, lithium-, and thermodilution measurements. The measured indicator dilution curves are typically fitted by a mathematical model to estimate the hemodynamic parameters of interest. This paper presents a new maximum-likelihood algorithm for parameter estimation, where indicator dilution curves are considered as the histogram of underlying transit-time distribution. Apart from a general description of the algorithm, semianalytical solutions are provided for three well-known indicator dilution models. An adaptation of the algorithm is also introduced to cope with indicator recirculation. In simulations as well as in experimental data obtained by dynamic contrast-enhanced ultrasound imaging, the proposed algorithm shows a superior parameter estimation accuracy over nonlinear least-squares regression. The feasibility of the algorithm for use in vivo is evaluated using dynamic contrast-enhanced ultrasound recordings obtained with the purpose of prostate cancer detection. The proposed algorithm shows an improved ability (increase in receiver-operating characteristic curve area of up to 0.13) with respect to existing methods to differentiate between healthy tissue and cancer.