Maximum Entropy Technique and Regularization Functional for Determining the Pharmacokinetic Parameters in DCE-MRI.

Area of Science:

• Medical Imaging
• Biophysics
• Computational Science

Background:

• Accurate determination of the arterial input function (AIF) is crucial for pharmacokinetic analysis in dynamic contrast-enhanced MRI (DCE-MRI).
• The AIF estimation is a challenging linear ill-posed inverse problem, often leading to ill-conditioned systems of equations after discretization.
• Existing methods may lack robustness or require complex optimization techniques.

Purpose of the Study:

• To develop and validate a novel algorithm for robust AIF determination in DCE-MRI.
• To improve the estimation of pharmacokinetic parameters by enhancing AIF accuracy.
• To address the ill-posed nature of AIF estimation using maximum entropy and Bayesian inference.

Main Methods:

• Utilized the maximum entropy technique (MET) combined with regularization functionals.
• Employed a Bayesian estimation approach to infer pharmacokinetic parameters and quantify solution uncertainties.
• Developed a new algorithm to estimate probability distribution functions for the AIF, specifically exploring Gamma and Erlang distributions.

Main Results:

• The proposed algorithm demonstrated improved convergence behavior and more consistent results compared to previous methods (e.g., exponential or Weibull distributions).
• Analysis of simulated and real breast tumor datasets showed enhanced morphological and functional statistics.
• The Gamma and Erlang distributions, estimated by the new algorithm, proved to be more appropriate and robust AIFs.

Conclusions:

• The novel algorithm effectively solves the AIF determination problem in DCE-MRI, offering a more robust and accurate solution.
• Combining Bayesian inference with maximum entropy regularization enhances the reliability of pharmacokinetic parameter estimation.
• This approach provides a significant advancement for quantitative analysis in DCE-MRI studies, particularly for tumor characterization.