Dynamic Contrast-Enhanced MRI (DCE-MRI) Analysis

Having set up your input and output images as described in the Introduction, you can now set up the quantification parameters for DCE-MRI.

Quantification Parameters

• Plasma relaxivity (relaxivity of the contrast agent in blood plasma) and interstitium relaxivity (relaxivity of the contrast agent in the extra-vascular extra-cellular space). Many workers assume that these two relaxivities are the same as each other. If this is the case, then the absolute values are not important, and they can be left at their default values of 1.0. However, if these two relaxivities are known to be different, then their correct values should be entered. Enter the relaxivities in L s^-1 mol^-1. If the two relaxivities are different, then entering incorrect values will result in an error in the estimate of v_e (the extracellular space volume fraction).

  Note: the relaxivity is the slope of a plot of R₁ (y-axis) against [Gd] (x-axis).

• Haematocrit. The haematocrit (the fraction of whole blood volume occupied by the red blood cells) in the blood vessels. Needed for correct estimation of v_e. The haematocrit is assumed to be the same in all vessels.

Further down you will also see:

• Pulse sequence type. The DCE-MRI tool assumes that you are using a T₁-weighted pulse sequence, and that the signal intensity increases as the contrast agent washes through the tissue. The change in R₁ is assumed to be proportional to the concentration of contrast agent. Choose the type of sequence used:
  • Saturation-recovery. The signal intensity is assumed to be of the form:

    M=M₀ (1-exp^-TR.R1)

    Enter the recovery time (TR) in milliseconds.
  • Inversion-recovery. The signal intensity is assumed to be of the form:

    M=M₀(1-2exp^-TI.R1)

    Enter the inversion time (TI) in milliseconds.
  • Fast Low-Angle SHot (FLASH). The signal intensity is assumed to be of the form:

    M=M₀sin(α)(1-exp^-TR.R1)/ (1-cos(α)exp^-TR.R1)

    Enter the recovery time (TR) in milliseconds, and the flip angle (α) in degrees.
  • Signal change ∝ R₁. The concentration of contrast agent is assumed to be proportional to the change in signal intensity. If this option is chosen, then the plasma and interstitium relaxivity values are used as the slope of the plot of signal intensity (y-axis) versus concentration (x-axis). Thus, this option can be regarded as semi-quantitative for many sequences. However, if you pre-compute the R₁ values and use input images that are R₁ images rather than signal intensities, then you can also use this option to give quantitative results. Use this option if you use some other method (outside the DCE-MRI tool) to create pre-computed R₁ images.
  If you choose the Saturation-recovery, Inversion-recovery or FLASH option, then you will also need to set the reference image type and the reference image. This image is also used in the conversion from the measured signal intensities to changes in R₁. As a reference image, select:
  • M0-map image if your reference image is a map of equilibrium signal intensities. This is the signal intensity in an image if it were acquired with a very long repetition time and a 90° flip angle.
  • Pre-contrast T1-map image if your reference image is a map of longitudinal relaxation time (in seconds).
  • Pre-contrast R1-map image if your reference image is a map of longitudinal relaxation rate (in units of seconds^-1.).

  You set the reference image by clicking on the icon, by typing in the directory and file names, or by pressing the right mouse button and selecting from the menu of recently-used images.

  

  Whatever type of reference image you use, it must have as many slices as there are physical slice locations in your input image(s).

  For the Saturation-recovery and Inversion-recovery sequences, an M₀ reference image can be acquired by setting a long recovery time (TR or TI). For the FLASH sequence, a reference image can be obtained in one of two ways:

  • Set the flip angle (α) to 90° and acquire an image with a long TR to create an M₀ reference image; or

  • Acquire a series of images with different flip angles (α) and use the Image Fitter to estimate M₀ and R₁ by fitting the expression

    M=M₀sin(α)(1-exp^-TR.R1)/ (1-cos(α)exp^-TR.R1)

    to the resulting series of images. The fitter will produce images of M₀ and R₁, and you can use either of them as your reference image.
  There are, of course, many other ways to acquire and create reference images.

Arterial Input Function (AIF)

• The time at which contrast agent first appears in the feeding artery. If you cannot see a feeding artery and are measuring the AIF in a draining vein (or by some other means) then this should be the time at which the image first changes in intensity as the contrast arrives in the tissue. By selecting from , you can specify this either as:
  • The time at which contrast first appears (the first image is acquired at t=0).
  • The scan number at which contrast first appears (the first scan is numbered 1).
  Enter one of these. You can graphically confirm the arrival time using the Roaming Response dialog.

• You may optionally enter the time at which the series of images is truncated for the purposes of analysis using . Images after this time will not be used as part of dataset for the analysis. Analysis will proceed as if the data after this time point has not been acquired. By selecting from , you can specify this either as:
  • The time at which to truncate the data set, or
  • The scan number at which to truncate the data set. Enter one of these, or leave the field blank to use the whole dataset. you can graphically confirm the time of truncation using the Roaming Response dialog.

• Any lag from the measured arterial input function to the actual arterial input feeding directly into the tissue. This is specified as a number of frames (time points) of lag. Change the AIF lag using the AIF lag spinner.

  

  The text to the right of the spinner explains the meaning of the lag value you set.

• Choose how you want to obtain the AIF. You can:
  • Manually draw round one or more feeding arteries to produce a set of regions of interest (ROIs). In this case you can produce an ROI file that contains one or more ROIs drawn at one particular time point. This same set of ROIs will be used for all time points to obtain the AIF. Alternatively, you can define a set of ROIs at one time point, then copy and paste them to the same slice location for all time points, and manually adjust the position to account for patient movement during the scan. If you do this, the tool will expect to find a set of ROIs at every time point, and will given an error if it does not.

    Select the ROI file used for the AIF by pressing the button, or type the name into the text field:

    

    If you have specified the pulse sequence to be saturation-recovery, inversion-recovery or FLASH, then you must also supply an ROI file that contains an ROI drawn on the feeding artery on the reference image (M₀, T₁ or R₁ map). Space will be provided in the tool for you to enter the name of this ROI file.

  • Use a predefined AIF. You can read the values of the AIF from a file that contains a list of concentration of contrast agent values. The file must contain two columns of numbers:
    • The first column is the time point at which the concentration of contrast agent ([Gd]) is measured.
    • The second column contains the blood plasma concentration values (moles/L).
    Whitespace or a comma (",") should separate the two columns.

    Note: the format required for the AIF file has changed between Version 6 of and Version 7 of Jim. The first column (time values) is now used. The onset of the arterial input function is taken to occur at the first time point where the concentration value first rises above zero. This time point (after taking account of any tissue lag that you input) is taken to coincide with the time that you specify for contrast arrival. The time between measures of the concentration may now be unevenly-spaced and does not have to match that of the input images. The pre-defined AIF is automatically resampled to the required time-base of the input images.

    The DCE-MRI tool will output such a file for every analysis it performs, which will give you an example of what is needed. For semi-quantitative analysis you just need values that are proportional to the concentration of contrast agent in the artery. Select the file used for the predefined AIF by pressing the button, or type the name into the text field:

    

• Check the "Show AIF graph" check-box if you would like a window to pop up showing you the AIF that is being used in the DCE-MRI analysis.

Analysis Type

• iAUC. This computes the initial area under the tissue response curve and compares it to the integrated area under the arterial input function at particular time points after contrast arrival. This is considered to give a semi-quantitative assessment of tissue response.
• Tofts model. Here, the rate of flux of contrast agent from the plasma to the extra-vascular extra-cellular space (EES) is assumed to be proportional to the concentration difference between the plasma and the EES. Within the tissue of interest, the blood plasma is assumed to make a negligible contribution to the overall signal intensity.
• Tofts with vp term. This is a modified Tofts model that includes a contribution to the signal intensity in the tissue of interest from the blood plasma. An extra term (v_p - the blood plasma volume fraction of the whole tissue) is estimated.
• 1Cmpt (one compartment). This is like the Tofts model in that it has a single-exponential residue function. However, the calculated output parameters are geared towards a situation where flow is being estimated (see below).
• Fermi. This model assumes that the residue function is in the form of a Fermi function. This form of residue function has been observed in the case of cardiac perfusion.
• 2CXM (Two-compartment exchange model). This model does not make the assumption of a negligible plasma transit time and is a more general model that is able to distinguish (given sufficiently good sampling temporal resolution) between the delivery of contrast agent to the tissue (flow) and the permeation of the contrast agent across the endothelial wall.

Now choose whether you want to perform pixel-by-pixel analysis, or region of interest (ROI) analysis. You can either:

• Estimate the DCE-MRI parameters for every image pixel, producing output images of these parameters, or
• Estimate the DCE-MRI parameters for a whole ROI, producing single values of these parameters. If you choose this option, you will choose an ROI file which can contain either:
  • A single ROI drawn on one slice. This same ROI will be used for all time points.
  • An ROI at every time point. Each ROI should be the same shape and size (which can be achieved by copying and pasting an ROI to all the desired slices) for all time points, but you can manually adjust the position to account for movement of the tissue of interest. If you choose this option, you will need to supply the name of the ROI file, and also the base name of the AIF output file. This output file will be written with the AIF used in the analysis.
  If you have specified the pulse sequence to be either saturation-recovery, inversion-recovery or FLASH, then you must also supply an ROI file that contains an ROI of the same shape and size as the region to be analysed, positioned on the reference image (M₀, T₁ or R₁image. Space will be provided in the tool for you to enter the name of this ROI file.

DCE-MRI Output Parametric Images

• For the iAUC model:
  • iAUC30, iAUC60, iAUC90, iAUC120, iAUC150, iAUC180. These are the initial areas under the tissue concentration curves, integrated from the time of first contrast appearance to the specified times (30, 60, 90, 120, 150 and 180 seconds). If the DCE-MRI time-series does not extend out to all these times, the iAUC images beyond the end of the time-series will be blank.
  • iAUC30bn, iAUC60bn, iAUC90bn, iAUC120bn, iAUC150bn, iAUC180bn. These are the blood-normalised initial areas under the enhancement curve, integrated from the time of first contrast appearance to the specified times (30, 60, 90, 120, 150 and 180 seconds). If the DCE-MRI time-series does not extend out to all these times, the iAUC images beyond the end of the time-series will be blank. Blood normalisation is performed by dividing the iAUC values by the corresponding iAUC values for the plasma concentration of contrast agent.

• For the Tofts model:
  • Ktrans. The volume transfer constant, with units of ml / ml / minute.
  • ve. The extra-vascular extra-cellular space volume fraction (as a fraction of the whole tissue volume). Values in this output image will range between 0 and 100%.

• For the extended Tofts model, the same as for the basic Tofts model, plus
  • vp. The blood plasma volume fraction (as a fraction of the whole tissue volume). Values in this output image will range between 0 and 100%.

• For the 1Cmpt model:
  • F. The tissue whole blood flow (perfusion) in units of ml / ml / minute.
  • vd. The distribution volume for the contrast agent as a percentage of the tissue volume. Values in this output image will range between 0 and 100%.
  • MTT
  . The mean transit time through the tissue, in seconds.

• For the Fermi model:
  • Fp. The plasma flow (perfusion) in units of ml / ml / minute.

• For the 2CX model:
  • Fp. The flow into the tissue, with units of ml / ml / minute.
  • PS. The permeability-surface area product, for exchange between the intravascular space and the extra-vascular, extra-cellular space (EES). Units are ml / ml / minute.
  • ve. The extra-vascular extra-cellular space volume fraction (as a fraction of the whole tissue volume). Values in this output image will range between 0 and 100%.
  • vp. The blood plasma volume fraction (as a fraction of the whole tissue volume). Values in this output image will range between 0 and 100%.
  • MTT. The mean transit time for passage of contrast agent through the tissue, with units of seconds.

All models will also produce an RMSE (root-mean-square error) image, which is a measure of the goodness of fit. This is the root-mean-square of the residuals between the fitted curve and the time-series data. These output images are in floating-point format, and can be viewed using Jim. The output images will be of the same image type as the first input image.

What Happens When Fitting Fails?

• The data are too noisy to perform an accurate fit.
• The data are corrupted by artefacts.
• The model is a poor match to the data.
• Fitting the model results in some fitted parameters which are not physically plausible. Examples include Fp, Ktrans, PS, vp and ve values that are negative.

If you are performing ROI analysis, a fitting failure will result in an pop-up error message indicating the nature of the failure. If you are performing pixel-by-pixel analysis, any pixel where fitting fails will have a value in all output images which is either zero, or NaN (not-a-number). Which of these two is written to the images is controlled by a setting in the user preferences.