Assessment of Perfusion by Dynamic Contrast-Enhanced Imaging Using a Deconvolution Approach Based on Regression and Singular Value Decomposition

Abstract

The assessment of tissue perfusion by dynamic contrast-enhanced (DCE) imaging involves a deconvolution process. For analysis of DCE imaging data, we implemented a regression approach to select appropriate regularization parameters for deconvolution using the standard and generalized singular value decomposition methods. Monte Carlo simulation experiments were carried out to study the performance and to compare with other existing methods used for deconvolution analysis of DCE imaging data. The present approach is found to be robust and reliable at the levels of noise commonly encountered in DCE imaging, and for different models of the underlying tissue vasculature. The advantages of the present method, as compared with previous methods, include its efficiency of computation, ability to achieve adequate regularization to reproduce less noisy solutions, and that it does not require prior knowledge of the noise condition. The proposed method is applied on actual patient study cases with brain tumors and ischemic stroke, to illustrate its applicability as a clinical tool for diagnosis and assessment of treatment response.

Fundamentals of tracer kinetics for dynamic contrast‐enhanced MRI

“…Robust implementations of numerical deconvolution usually attempt to include some form of regularization to achieve stability and to arrive at reasonable solutions. For example, in the singular value decomposition methods, regularization can be achieved by specifying an appropriate truncation parameter (21,30,31), and in Fourier transform methods regularization can be imposed in the form of filters that attenuate certain undesirable frequencies. Another purpose of regularization is to incorporate prior knowledge about the characteristics of F p R(t) in the numerical deconvolution process.…”

“…Another purpose of regularization is to incorporate prior knowledge about the characteristics of F p R(t) in the numerical deconvolution process. For example, we know that R(t) is a positive-definite, nonincreasing function and such requirements can be imposed on the deconvolution solution in the form of regularization constraints (30)(31)(32).…”

Effects of respiratory cycle and body position on quantitative pulmonary perfusion by MRI

“…Mean pulmonary perfusion was taken as the average of the left and right lung perfusion in anterior, mid, and posterior lung fields. The model-independent deconvolution program was based on the assumption that the pulmonary circulation is a linear system and the pulmonary signal intensity and blood pool signal intensity can be used to deduct the pulmonary perfusion parameters (14). The arterial input function (C a ) is the signal intensity of contrast agent in blood pool and the output function is the pulmonary tissue contrast signal C pul .…”

“…TSVD algorithm is a numerical deconvolution regularization technique using less than standard matrix components (14). Through a process of removing singular values less than a user defined threshold, a noise filtering effect is achieved as the corresponding linear equation rank decreases.…”

Mapping myocardial perfusion with an intravascular MR contrast agent, a robust estimation by a spatially constrained approach

“…Existing approaches consider time-domain regularization [3], [4], [5]. Our approach differs in the sense that we perform regularization of perfusion parameters in the spatial domain, which allows to smooth the parameter estimates in homogeneous regions while preserving discontinuities between different tissues or different perfuse myocardial territories.…”