Functionality

The following sections outline the basic image processing steps that are provided in dcemri. There are numerous choices when performing a quantitative analysis for a DCE-MRI study. The methods specified below follow our data analysis pipeline but are not meant to be restrictive. If there is a feature that you believe should be included, do not hesitate to contact us. Suggestions are welcome!

Motion Correction and Co-registration

Basic motion correction within an acquisition, and co-registration between acquired series, is available using template matching. A reference volume must be pre-specified where a mask has been applied to remove all voxels that should not be included in the algorithm. Note, only three-dimensional translations are allowed and no interpolation is used (i.e., only whole-voxel translations) at this time.

T1 Relaxation and Gadolinium Concentration

Estimation of the tissue T1 relaxation rate is the first step in converting signal intensity, obtained in the dynamic acquisition of the DCE-MRI protocol, to contrast agent concentration. The subsequent steps provided here focus on pharmacokinetic modeling and assumes one has converted the dynamic acquisition to contrast agent concentration. Please see Collins and Padhani (2004) for a discussion on this point.

There are a miriad of techniques to quantify T1 using MRI. Currently curve-fitting methods for two popular acquisition schemes are available

Once the tissue T1 relaxation rate has been estimated, the dynamic acquisition is then converted to contrast agent concentration. Note, the B1 field is assumed to be constant (and accurate) when using multiple flip angles to estimate T1. At higher fields strengths (e.g., 3T) the B1 field should be estimated in order to correct the prescribed flip angles.

Arterial Input Function

Whereas quantitative PET studies routinely perform arterial cannulation on the subject in order to characterize the arterial input function (AIF), it has been common to use literature-based AIFs in the DCE-MRI literature. Examples include

There has been progress in measuring the AIF using the dynamic acquisition and fitting a parametric model to the observations. Recent models include

dcemri has incorporated all parametric models given above for the AIF, except Parker et al. (2006), into the kinetic parameter estimation step. While default values for each model are provided, there is also the ability to include user-specified parameters.

Kinetic Parameter Estimation

The standard Kety model, a single-compartment model, or the extended Kety model, the standard Kety model with an extra “vascular” term, form the collection of basic parametric models one can apply using dcemri. Regardless of which parametric model is chosen for the biological system, the contrast agent concentration curve at each voxel in the region of interest (ROI) is approximated using the convolution of an arterial input function (AIF) and the standard/extended Kety model.

Parameter estimation is achieved using one of two options in the current version of this software:

Least-square estimates of the kinetic parameters Ktrans and kep (also vp for the extended Kety model) are provided in dce_nlreg while the posterior median is provided in dce_bayes. When using Bayesian estimation all samples from the joint posterior distribution are also provided, allowing one to interrogate the empirical probability density function (PDF) of the parameter estimates.

Statistical Inference

No specific support is provided for hypothesis testing in dcemri. We recommend one uses builtin facilities in R to perform ANOVA or mixed-effects models based on statistical summaries of the kinetic parameters over the ROI per subject per visit. An alternative to this traditional approach is to analyze an entire study using a Bayesian hierarchical model (BHM) in PILFER.

One may also question the rationale for hypothesis testing in only one kinetic parameter. Preliminary work has been performed in looking at the joint response to treatment of both Ktrans and kep in DCE-MRI by O'Connor et al. (2010).