BJR
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
British Journal of Radiology (2004) 77, S154-S166
© 2004 British Institute of Radiology
doi: 10.1259/bjr/16652509
This Article
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Jackson, A
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Jackson, A

Full Paper

Analysis of dynamic contrast enhanced MRI

A Jackson, PhD, FRCR, FRCP

Imaging Science and Biomedical Engineering, The Medical School, University of Manchester, Oxford Road, Manchester M13 9PT, UK


   Introduction
Top
 Introduction
Conclusions
 References
 
The advances in MRI image acquisition have led to the increasing use of parametric or calculated images, which are designed to display physiological features of tissues rather than simply anatomical structure. One common type of dynamic imaging data is produced when rapidly repeated sequential imaging sequences are performed during injection of an intravenous contrast media [1–8]. Although many of the applications of dynamic contrast enhanced imaging were developed using MRI in the brain they are now being commonly applied using both MRI and CT. These techniques now have both research and clinical applications throughout the body and in many types of pathological tissue. In this article we will describe the analysis techniques which can be applied to dynamic contrast enhanced data. We will start with the specific application of dynamic susceptibility contrast imaging of perfusion using MRI (DSC-MRI) since a number of technical aspects of the data collection and analysis are unique to this approach. We will then discuss the techniques available for the analysis of dynamic relaxivity contrast enhanced MRI (DRC-MRI), which are also directly applicable to dynamic contrast enhanced CT data.

Dynamic susceptibility contrast enhanced imaging
There are two commonly used methods for measuring cerebral blood flow (CBF) using MRI [1, 9]. Arterial spin labelling magnetically labels the water in the blood entering the sample to provide an endogenous tracer of flow, although it is highly attractive this technique is not yet sufficiently robust for routine clinical use [1, 2]. DSC-MRI uses rapid measurements of MR signal change following the injection of a bolus of a paramagnetic MRI contrast agent [1, 2, 4, 10, 11]. The signal loss resulting from passage of the contrast agent bolus on T2* weighted images can be used to calculate the change in contrast concentration occurring in each pixel. These data can be used to produce calculated estimates of cerebral blood volume (CBV), mean transit time (MTT) and CBF. DSC-MRI is simple to perform in a clinical environment and is currently the MR perfusion technique most commonly used in clinical studies [1, 9].

The use of the first-passage of the contrast bolus to measure cerebral perfusion is fundamentally attractive. The use of contrast injection produces controllable decreases in signal intensity, whilst basing the analysis purely on first pass data imposes a short image acquisition which can be easily incorporated into existing clinical imaging protocols. First-pass bolus kinetics are also well documented and highly generic, so that any successful technique can be used with a range of imaging technologies including both MRI and CT.

Data collection considerations for DSC-MRI
The typical imaging strategy is to collect data using a fast imaging technique such as single or multishot echo planar imaging (EPI) to produce a temporal resolution of approximately 2 s. During this 2 s acquisition window it is usually possible to acquire in the region of 5–15 slices at a resolution of 128 x 128 or greater, depending on the scanner specifications [1, 2, 12–14]. The imaging sequence may be gradient echo which will maximize T2* weighting or alternatively a spin echo approach can be used which will minimize the signal contribution from large vessels. Many early authors preferred the latter approach since it theoretically produces signal changes which predominantly reflect the passage of contrast through the capillary bed [10, 15]. More recently several studies have shown that there is little effective benefit from the use of spin echo images but significant cost in terms of signal to noise ratio [16]. The patient should be comfortably positioned with adequate cushioning to reduce movement and light restraining straps should be used in the same way that they would be used for normal MRI. This level of restraint, combined with the relatively short acquisition times usually results data with no significant movement so that data coregistration is seldom required. A series of at least five pre-contrast images should be collected prior to the passage of the bolus and many centres will collect for up to 1 min to provide a large number of pre-contrast images to improve the estimation of the signal intensity baseline during analysis. The contrast agent is administered by intravenous injection and the injection technique must be carefully standardized. A standard contrast dose (0.1 mmol kg–1) is adequate in most cases although some centres use double this dose in order to improve signal to noise ratio. The rate of contrast administration should be standardized. Many centres use a standard flow rate although this leads to an injection duration, which is proportional to the contrast volume and therefore to the body weight. We prefer to use a fixed injection time administering the contrast over a period of 4 s using an automated pressure injector. The injection should be given into the right arm where possible in order to minimize the risk of significant backflow into the jugular vein during injection. The injection should be followed by a saline flush of at least 25 ml delivered at the same rate in order to ensure that the bolus which enters the central circulation is as coherent as possible. A careful manual injection technique can produce acceptable and reproducible results. For manual techniques the injection should be given through a large cannula preferably introduced into a large antecubital vein, the cannula should be at least 18 G for manual injection to reduce the resistance of the injection system. The injection should be given at an even rate and should be immediately followed by a chaser of the same amount of normal saline given at the same rate.

Analysis of DSC-MRI data
Basic theory
The analysis of DSC-MRI data is based on the assumption that the contrast agent remains within the vascular space throughout the examination acting as a blood pool marker. This assumption is untrue except within the brain where there is no contrast leakage due to the blood–brain barrier (and in the testes where a similar barrier also exists). The application of DSC-MRI was therefore initially limited to studies of normal brain although modifications of the technique have subsequently allowed its use in enhancing tissues (see below) [17, 18]. One of the main aims of DSC-MRI is the production of image based measurements of blood flow [10]. Although this is theoretically a straightforward process a number of major problems exist which has considerably restricted the clinical use of the technique, these are also discussed separately in the next section [1, 12, 19].

The conventional approach to calculating CBF uses the area under the contrast concentration curve as an estimate of blood volume within the pixel (CBV) and the width of the contrast bolus as an estimate of the MTT [4, 11, 20] (Figure 1Go). The blood flow can then be calculated using the central volume theoremGo


{77S154E001}

The initial calculation of local contrast concentration from the observed signal change a straightforward and contrast concentration is linearly related to the T2 rate changes ({Delta}R2), which can be calculated for using the relationshipGo


{77S154E002}

where S(0) is the base line signal intensity, S(t) is the pixel intensity at time t and TE is the echo time. This allows transformation of signal intensity time course data to contrast concentration time course data.



View larger version (17K):
[in this window]
[in a new window]
  		Figure 1. Graph showing the change in contrast concentration against time in a single voxel of normal grey matter of a normal brain. Crosses represent the original measurement points and a straight line shows the optimal curve fit result. Notice that the curve fitting process effectively removes the noise present in the original data. Since the curve fitting procedure uses only data during the early part of the first passage of the contrast bolus it also removes the effect of re-circulation of contrast which is responsible for the elevation of measurements in the later part of the curve.

 
The CBV will be proportional to the area under the concentration time course curve shown in and is calculated analytically as:Go


{77S154E003}

where t0 is the time of first arrival of contrast and te is the time at which {Delta}R2 returns to baseline values.

The MTT is then estimated as some form of standardized measurement of the width of the curve such as the width at half the maximum height (full width at half maximum; FWHM) and these values can be used to calculate parametric maps of CBV, MTT and CBF (CBV/MTT) (Figure 2Go).



View larger version (80K):
[in this window]
[in a new window]
  		Figure 2. Calculated parametric images from a normal brain. Images represent (a) cerebral blood volume (CBV); (b) cerebral blood flow (CBF); (c) mean transit time (MTT); (d) time of arrival (T0); (e) time to peak concentration (TTP) and (f) standard fitting error (SFE). Note that maps of CBV show high levels within the blood vessels, much lower levels in grey matter and the lowest measurements of all in white matter. Maps of CBF show a similar distribution of values. MTT is relatively uniform across the entire brain except for a slight prolongation is in the anterior and posterior cerebral watershed areas, particularly on the right (left of the image). Both T0 and TTP maps show contrast delay in the central white matter as with early arrival in peripheral cortex. The map of SFE shows very low values (red) in areas of high signal to noise ratio corresponding to blood vessels. High values are seen in cortex and the highest of all in the white matter reflecting measurement uncertainty due to be creasing temporal signal to noise ratio in these tissues.

 
In addition to these flow related parameters it is possible to produce maps based on the time the contrast takes to arrive in the voxel using the time of arrival (T0) or, more commonly the time to peak concentration (TTP). These parameters are unique to contrast based bolus tracking techniques (Figures 2 and 3GoGo).



View larger version (101K):
[in this window]
[in a new window]
  		Figure 3. A series of calculated images of time to peak concentration (TTP) in a patient with a severe right-sided carotid stenosis. There is severe prolongation of contrast arrival throughout the right hemisphere indicated by a loss of the very early arterial arrival (red) seen on the left and increased areas of delayed contrast arrival seen in blue.

 
Problems with DSC-MRI data
Unfortunately this deceptively simple approach to the measurement of CBF is subject to several major errors which have led to a number of major modifications of the analysis approach in an attempt to produce more accurate quantitative estimates of blood flow. We will discuss three main problems with the technique:

Contrast recirculation: Analysis of the contrast bolus passage assumes that the bolus passes through the voxel and that the concentration of contrast then returns to zero. In fact the contrast re-circulates through the body and a second re-circulation peak is seen following the first. As the contrast continues to circulate the bolus disperses and widens so that the second peak is lower and broader than the first and by the time of the third re-circulation the intravascular contrast has mixed evenly throughout the blood volume causing a small constant baseline elevation in the contrast concentration. Measurement of CBV is therefore subject to errors due to the presence of both first pass and re-circulating contrast in the vessels during the later part of the bolus passage [4, 11, 17, 20]. In addition the identification of the end of the bolus passage is complicated by the presence of the second re-circulation of the bolus (Figure 1Go).

One way in which this can be approached is by using a curve fitting technique [1, 4, 5]. This relies on the fact that the shape of the contrast concentration curve during the passage of the first bolus can be shown theoretically to always conform to a specific shape known as a gamma variate ({Gamma} variate) [21]. A number of standard mathematical approaches exist which allowed the identification of the {Gamma} variate curve which best fits the data in individual pixel. Estimation of the optimal curve produces a mathematical description of the contrast concentration changes during the first passage of the contrast bolus and allows direct calculation of the area under the curve, its width and the timing parameters T0 and TTP (Figure 1Go). The use of curve fitting also smoothes the data, effectively reducing noise and eliminates the contamination of the first pass bolus due to contrast agent re-circulation. [1, 2, 12, 14, 20, 22, 23] (Figure 4Go).



View larger version (13K):
[in this window]
[in a new window]
  		Figure 4. A graph showing the change in contrast concentration against time in a single voxel of normal grey matter of a normal brain. The data collection has been performed with a deliberately degraded acquisition sequence in order to reduce contrast to noise ratio in the data. A direct comparison with Figure 1Go shows significantly lower signal to noise ratio in these data. Despite this the curve fitting programme has derived an optimal fit which will be used the calculation of vascular parameters. However, if fitting error is calculated as described in the text then this curve fit result will be associated with very broad confidence intervals.

 
Contrast leakage and tissue enhancement: Leakage of contrast into the interstitial space will cause signal changes, principally by relaxivity mechanisms. Susceptibility based imaging methods offer the opportunity to separate these relaxivity and susceptibility based effects and to produce images in which the effect of contrast leakage is eliminated or minimized. The use of techniques with reduced T1 sensitivity, such as low flip angle gradient echo based sequences, effectively removes relaxivity effects although some workers have observed residual effects in rapidly enhancing tumours [18, 24, 25]. The major problem with this method is the loss of signal to noise ratio produced by the reduction in flip angle although this can be partially compensated by increased contrast doses [18]. Another approach to reducing T1 sensitivity is to use a dual echo technique in which the T1 weighted first echo is used to correct the predominantly T2 weighted second echo [26]. Unfortunately, the requirement for two echoes places considerable demands on the sampling time and inevitably restricts the number of samples, and therefore slices which can be obtained. An alternative method is to use pre-enhancement with an additional dose of contrast agent. The change in signal intensity resulting from T1 shortening is bi-exponential (Figure 5Go) so that for any given sequence, there is a plateaux phase during which signal intensity remains relatively constant. The effect of this response curve is that pre-enhancement of tumours will reduce the relaxivity based signal intensity responses to subsequent contrast doses (Figure 6Go). None of these approaches offers a perfect solution and the choice of method must be based on individual analysis task to be undertaken. For a fuller discussion see Kassner et al [17].



View larger version (11K):
[in this window]
[in a new window]
  		Figure 5. Curves showing changes in signal intensity on T2* weighted gradient echo images at different concentrations of standard paramagnetic contrast material. Note that the use of a low flip angle (bottom graph) results in almost no sensitivity to contrast media so that this strategy can be used in areas of marked contrast leakage. However, note that overall signal is far lower than is obtained with higher flip angles. At higher flip angles there is a significant contrast effect resulting in elevation of signal intensity as contrast concentration rises. However, this rise reaches a plateau after which the response to contrast concentration remains linear and eventually drops. This acts as the basis for a second strategy to remove the effect of contrast leakage by pre-enhancing tissues with a preliminary dose of contrast in order to reach a plateau phase. If this is achieved then the second injection of contrast will not be affected by further contrast leakage.

 


View larger version (12K):
[in this window]
[in a new window]
  		Figure 6. Signal intensity data from a single voxel in an enhancing tumour imaged using a gradient echo sequence with a flip angle of 35°. (a) The data in the graph show an initial signal drop followed by rapid signal rise due to enhancement. (b) The data in the graph was obtained from the same region of interest in the same tumour using the same sequence. However, these data were collected following pre-enhancement of the tumour and shows no sign of relaxivity related enhancement.

 
Bolus dispersion and the measurement of absolute CBF: The use of the central volume theorum to calculate values for CBF assumes that the technique can produce quantitative measurements of CBV and MTT. In fact the use of the area under the curve to estimate CBV results in relative measure which allows comparison of CBV between tissues rather than producing an absolute measurement. These relative measures can be calibrated by measurement of CBV in major vessels, which appears to be a relatively reliable and reproducible approach [14, 19].

The measurement of CBF also requires accurate estimation of MTT which is extracted from the width of the contrast bolus in each voxel [4, 5, 11, 20]. The width of the contrast bolus is actually affected by a combination of three factors. These are: 1) the width of the bolus entering the voxel (the arterial input function or AIF); 2) changes in bolus width due to regional alterations in flow; and 3) physical bolus broadening due to dispersive effects which are unrelated to flow (Figure 7Go).



View larger version (18K):
[in this window]
[in a new window]
  		Figure 7. Scattergrams showing the change in bolus width within a single slice of brain tissue plotted against contrast arrival time (TTM, time to mean contrast concentration). (a) The graph shows values from pixels with blood volume values greater than 5% (i.e. vessels) whereas (b) shows values from all pixels within the image. There is a clear linear relationship between contrast to age and duration and resulting from bolus dispersion of the contrast agent passes through the brain.

 
In practice the width of the bolus is strongly affected by individual variations in injection technique, contrast dose and cardiovascular function so that direct comparison of derived CBF measurements between individuals required assumptions that these sources of variation had been minimized or removed [11, 27]. One approach to this problem is to deconvolve data from each voxel with an input response function that removes such variability to produce an estimate of the contrast concentration time course in each voxel which approximates the changes which would be seen if the input were a bolus of infinitely short duration. This has been used as the basis of quantitative techniques for the absolute measurement of CBF [2, 3, 10–12, 23]. The use of a deconvolution approach assumes that the input function to each voxel can be accurately measured which is in actually impossible. These techniques therefore measure a surrogate AIF from one of the major arteries in the basal cisterns, most commonly the middle cerebral artery. The use of a surrogate measure requires an assumption that no additional broadening of the contrast bolus occurs between the AIF measurement point and the voxel [2, 10]. In fact it is clear that these effects do occur even in normal subjects [1, 10, 12, 28]. Recent simulation studies have suggested that this error introduces significant underestimation of CBF and overestimation of MTT. Simulation studies suggest that additional broadening of the bolus by over 2.5 s overestimates MTT by 200% and underestimates CBF by 50% and this underestimation increases further as bolus broadening increases [12].

In conclusion the use of bolus tracking experiments appears to be very attractive and allows rapid collection of high quality data concerning blood flow. Some of these data, specifically the contrast arrival time parameters, are unique to this technique and provide useful clinical information in a range of vascular disorders including stroke. The measurement of CBV appears to have clinical value in a number of areas, particularly in the study of neoplasm microvascular structure [1, 18, 25, 29]. The quantitative measurement of CBF is far more complex than was originally expected. Attempts to produce quantitative measures must account for individual variations in bolus delivery and bolus width between patients and the statistical variability in estimates of flow related parameters between voxels must be addressed. However, even if these steps are taken the effects of bolus dispersion will produce very significant errors in any quantified estimation of blood flow and such measurements must be treated with extreme caution in clinical applications [1, 12].

Dynamic relaxivity contrast enhanced imaging
The description of the DSC-MRI analysis given above illustrates a number of peculiarities associated with the technique. Contrast agent is treated as an intravascular marker and leakage into the interstitial space is ignored or where possible eliminated. In practice the pharmacokinetics of contrast agent distribution are more complex and considerable additional data can be obtained from explicit modelling of the contrast leakage (enhancement) process (Figure 8Go) [6, 7, 30]. In the presence of leaky capillary endothelial membranes intravascular contrast agents will pass into the extravascular extracellular space (EES) causing enhancement. The speed with which this leakage occurs will be governed by the surface area of leaky endothelium within the voxel, the permeability of the endothelium and the concentration gradient of the contrast agent across the vessel wall. It has become apparent that quantification of contrast leakage can provide powerful indicators of the state of neovascular angiogenesis in pathologies such as tumours and inflammatory tissue. This has particular relevance in cancer where inhibition of angiogenesis presents new therapeutic opportunities by targeting of the newly formed vessels or by inhibition of the angiogenic process itself [31–33]. Since promotion of endothelial permeability is a prime effect of the cytokines which stimulate angiogenesis such as vascular endothelial growth factor (VEGF), capillary leakage of contrast agents on MRI or CT provides an attractive possible approach for monitoring the effects of these drugs [31, 34–36].



View larger version (30K):
[in this window]
[in a new window]
  		Figure 8. Figure illustrating the basic compartments involved in pharmacokinetic modelling of intravenous contrast distribution. Movements of contrast between compartments will be governed by the ratio of the contrast concentrations between those compartments and by the local blood flow and the surface area and permeability of the capillary endothelium to contrast agent (represented by the thickness of the connecting arrows). The composite transfer coefficient for contrast between blood and the tissue of interest is represented as ktrans.

 
Quantification of the contrast enhancement effect can be performed using a variety of techniques which range from simple measures of the rate of enhancement to complex algorithmic analyses that apply pharmacokinetic models to the imaging data with the intention of measuring the transfer constant (ktrans) of contrast between the blood stream and the extracellular space [7]. In tissues where flow is adequate to deliver contrast to the tissues ktrans represents the product of the endothelial permeability and endothelial surface area.

Simple analysis techniques
The most straightforward approach to the quantification of enhancement is to directly compare the signal intensity curves from regions of interest. There are many measurements which have been suggested to allow this type of analysis. The simplest of these is a measurement of the time taken for the tumour tissue to attain 90% of its subsequent maximal enhancement (T90) [37]. Another parameter measures the maximum rate of change of enhancement (maximal intensity change per time interval ratio (MITR) [38]). Each of these is designed to minimize the variation which will occur between patients as a result of variations in contrast dose, injection and scanning techniques and scanner type. A slightly more independent description of curve shape is provided by the approach of Brix et al [39] which calculates a standardized slope of the enhancement curve, commonly referred to as k21 which should relate directly to endothelial permeability. Unfortunately most of these signal intensity based methods when applied to MRI data remain sensitive to variations between acquisition systems one reason for this is a non-linear relationship between contrast concentration and signal intensity. The non-linear to this relationship combined with water transfer effects means that areas of very high contrast concentration will tend to be represented by an appropriate the low signal intensities. This has a profound effect on the accuracy of measurements of the AIF were transient high concentrations of contrast occur. Since variations in the AIF must be compensated in quantitative analyses unpredictable inaccuracies can be expected when signal intensity data are used. Despite these limitations, qualitative and signal based analysis approaches to analysis have demonstrated clear clinical utility in a number of areas and may provide the majority of information required in clinical use [40–42].

Pharmacokinetic analysis techniques
The poor reproducibility of techniques which describe the shape of the enhancement curve has led many groups to attempt to develop a quantitative technique which will be independent of scanner type, scanning technique or individual patient variations [7]. This could be achieved if the data were used to calculate the endothelial permeability and the endothelial surface area which are biological features, independent of imaging approach. The product of the endothelial permeability and endothelial surface area represents the transfer coefficient ktrans which governs the leakage of contrast from the vascular to the extravascular compartment (Figure 9Go). The leakage of contrast will have the form:Go


{77S154E004}

where ve is the proportion of the voxel into which contrast can leak (contrast distribution space), Cl is the concentration of contrast in the space and Cp is the concentration of contrast in the blood. From this simple equation we can calculate ktrans if we have accurate estimates of the change in concentration of contrast in the blood stream and in the tissue over time. The calculation of contrast concentration also requires knowledge of the initial T1 value of the tissues before contrast arrival (R10) which must therefore be measured before the dynamic imaging is performed [30].



View larger version (27K):
[in this window]
[in a new window]
  		Figure 9. Graphical representation of the contrast distribution occurring within an individual voxel of tissue. Contrast passes from the blood into the interstitial tissues and the figure shows the standard mathematical abbreviations used to describe each of the individual tissue spaces.

 
The measurement of ktrans therefore has six principle steps:
  1. Acquire images to allow the calculation of R10
  2. Acquire T1 weighted dynamic images of contrast enhancement which include the enhancing tissue and its major vessels
  3. Correct images for motion using co-registration techniques
  4. Calculate contrast agent concentration maps
  5. Identify and measure the AIF
  6. Calculate ktrans for each voxel

The measurement of R10 can be performed in a number of ways but the use of gradient echo images with variable flip angles is quick and accurate [43] (Figure 10Go). The dynamic series of images must be acquired at a temporal resolution adequate to allow accurate characterization of the AIF which will in turn depend on the injection technique. Many workers prefer the use of a bolus injection similar to the technique used for DSC-MRI which imposes a need for high temporal resolution (≤5 s) and restricts the coverage and spatial resolution which can be achieved. The dynamic imaging sequence must include pre-contrast images and must collect data for a sufficient period to allow accurate estimation of the ve and of the contribution of renal excretion. In practice this requires data collection over a period of 5–10 s using a simple three-dimensional (3D) gradient echo sequence with a spatial resolution of 128 x 128 x 25, TR=4.3, TE 1.1 min. The data collection technique can use any T1 weighted approach and we routinely and a flip angle of 35° which gives a temporal resolution of approximately 5 s. Measurement of the AIF is problematic and it is important to ensure that no inflow effects contribute to the signal change which requires some form of pre-saturation of inflowing spins. This can be achieved using a pre-saturation slab with the inevitable time penalty or by using the slice select gradients of the imaging sequence to saturate inflowing spins where this is possible. Once the images have been co-registered they can be used to calculate contrast agent concentration maps for use in the final pharmacokinetic model (Figure 10Go). Maps of R10, C(t) and the AIF are used to calculate ktrans and contrast leakage space (vl) on a pixel by pixel basis using the tri-exponential model described by Tofts and Kermode [30].



View larger version (114K):
[in this window]
[in a new window]
  		Figure 10. A series of dynamic MR images (top) showing contrast enhancement and passage of the contrast agent into the interstitial tissues using a 3D T1 weighted gradient echo acquisition. Illustrated images are spaced approximately 10 s apart. The lower row of images shows the calculated concentration of contrast agent which is derived from the images in the top row and which can be used as the basis for pharmacokinetic analysis of enhancement patterns.

 
Problems with measurement of ktrans
The measurement of ktrans using the approaches described above has two major disadvantages which are addressed below. The first of these is the assumption that the signal change in each voxel of the target tissue can be attributed to contrast leakage. The second is that the measurement takes a considerable period of time requiring imaging over at least 5 min and the third is that measurements of ktrans will be markedly affected by flow to the voxel as well as the permeability and surface area of the vascular endothelium.

Partial volume averaging effects: The analysis technique we have described assumes that samples taken from voxels in blood vessels will represent blood concentration changes whilst voxels within the target tissue will represent extravascular contrast leakage, In practice this assumption is incorrect and voxels within the target tissue are actually likely to represent a mixture some of which will have significant intravascular contrast content [44] (Figure 11Go). This will result in overestimation of ktrans in these voxels which can be seen as areas of apparently high permeability in areas of normal brain. The original approach to exclude these erroneous measurements was to exclude any voxel which produced values over a certain threshold (1.2 min–1) as being vascular in origin [30, 45]. Unfortunately this approach does not completely exclude contributions from intravascular contrast agent and can lead to exclusion of up to 50% of voxels from the image in enhancing tumours or other very vascular enhancing tissues. This problem has been approached by the development of more complex pharmacokinetic models (see below) which explicitly model the contribution of intravascular contrast to the signal change. These techniques result in greater reproducibility and do not require the loss of any pixels containing significant intravascular contrast [8, 46].



View larger version (16K):
[in this window]
[in a new window]
  		Figure 11. An example of contrast concentration changes occurring in various tissues during contrast enhancement. The concentration in the capillary plasma can be seen to reach an early peak as the bolus of contrast passes through the tissue vasculature. A second re-circulation peak can also be seen and subsequently there is elevation of the contrast concentration within the plasma which gradually decreases due to renal elimination. The change of concentration of contrast material in the interstitial space can be seen to represent a gradual elevation throughout the time course illustrated here. The amplitude and gradient of that rise will reflect capillary endothelial permeability surface area product and blood flow. The third curve shows the concentration of contrast observed in the tissue as a whole. In this case a small initial peak can be seen due to contrast within blood vessels in the voxel as well as a gradual elevation due to contrast leakage.

 
Breathold imaging approaches: The requirement to collect data for 5 min or longer is a major problem with permeability imaging. In the head this is easily possible and results in little or no misregistration of data and can be easily corrected by data co-registration. In other areas of the body problems with movement are far greater. This is particularly true in areas affected by respiratory motion where image acquisition during normal respiration leads to significant misregistration whereas respiratory gating techniques markedly limit the image acquisition strategy and the temporal sampling rate which can be achieved. One approach to this is to modify the pharmacokinetic model to describe only be first passage of the contrast bolus [46–48]. In these circumstances it is possible to assume that the concentration of contrast agent in the interstitium is negligible and this provides sufficient freedom in the model to estimate ktrans from data obtained during a breathold acquisition. This technique also eliminates the problems with partial volume averaging described above and produces highly reproducible parametric maps of both ktrans and CBV [48] (Figure 12Go).



View larger version (98K):
[in this window]
[in a new window]
  		Figure 12. Transverse contrast enhanced image from (A) dynamic series and (B) maps of T0, (C) kfp and (D) blood volume (BV) in a patient with metastatic colonic carcinoma. Metastatic deposits are seen in the right and left lobes. The T0 map shows early contrast arrival compared with normal liver in both metastases. Maps of kfp (C) and BV (D) show a peripheral rim of high kfp and BV in both metastases with low values in the tumour centre. This tumour rim shows K values that appear lower than those of normal liver parenchyma and BV values that appear lower.

 
Dependency of ktrans on regional blood flow: A third major problem in the model described above is the assumption that for any given combination of intravascular and extravascular contrast concentrations the rate of contrast leakage will be proportional to the permeability and surface area of the vascular endothelium. In fact this is true only where the supply of contrast to the vascular space is sufficiently high that contrast leakage will not affect the intravascular concentration. In practice plasma contrast concentration will decrease as leakage occurs in areas where the blood flow is inadequate to replenish contrast at adequate rate. In these circumstances the measurements of ktrans will reflect local blood flow and not the endothelial permeability surface area product. It is therefore incorrect to use measurements of ktrans as an indicator of endothelial permeability. Once again this significant disadvantage can be addressed by the use of more explicit pharmacokinetic models in the analysis (see below) [6, 7].

Use of improved pharmacokinetic models for analysis of DRC-MRI data
The specific problems associated with partial volume effects from vascular contrast and the flow dependency of ktrans measurements can be addressed by the use of improved pharmacokinetic models in the analysis technique. Extended pharmacokinetic models which explicitly model the effects of blood flow and partial volume averaging of vascular structures have been described. The main problem with the use of these complex models is that they introduce an increased number of free parameters into the analysis.

Unfortunately, these analyses rely on the use of curve fitting routines which perform less reliably as the number of free fitting parameters increases. The effect of this is to produce increasing uncertainty in the derived measurements and a considerable increase in the demands made on the data acquisition protocol in terms of temporal resolution and signal to noise ratio. As a result the full adiabatic model which compensates for both flow and intravascular contrast is seldom used even in research applications [6] (Figure 13Go). However a number of approaches have been described to specifically model the effect of intravascular contrast without a significant effect on the accuracy or reproducibility of the algorithm (Figure 14Go). It must be appreciated that ktrans will represent different biological effects depending on the analysis model that has been used. If the full adiabatic model is used then ktrans truly represents the permeability surface area product of the vascular endothelium. If the model used compensates for intravascular contrast effects then ktrans will be affected by both permeability surface area product and local blood flow. If the model does not compensates the intravascular contrast effects then ktrans will be affected by both permeability surface area product and local blood flow and will be significantly elevated by so-called "pseudopermeability" effects where the algorithm has failed to distinguish between intravascular and extravascular contrast agent.



View larger version (15K):
[in this window]
[in a new window]
  		Figure 13. Curve fit of dynamic contrast enhanced data from (a) a low-grade and (b) a high-grade glioma showing curve fit performed with an extended pharmacokinetic model. Illustrated values show separate calculated estimates of flow (F) and permeability surface area product (PS) as well as the proportional blood volume (vb) and the volume of the extravascular extracellular space (ve). The original data (represented as circles) shows some significant spread around the curve fit due to inherent signal to noise characteristics despite the fact that this data was taken from large regions of interest rather than single voxels.

 


View larger version (73K):
[in this window]
[in a new window]
  		Figure 14. Parametric calculated images of (a) ktrans and (b) cerebral blood volume (CBV) from dynamic T1 weighted contrast enhanced data. Parametric images were calculated using a first-pass analysis algorithm which decomposes its intravascular and extravascular contrast concentrations. ktrans will therefore be affected by both permeability surface area product and flow.

 
Statistical stability
The production of parametric images derived from mathematical calculations performed on each voxel raises additional problems since the accuracy of the estimates and therefore the potential errors will vary with the model used, the signal to noise ratio of the data in an individual case and from voxel to voxel within an individual set of images. This is a particular problem with the use of curve-fitting techniques which may obscure variations in the quality of the original data leading to generation of parametric images derived by fitting to data in which noise is the dominant, or only cause of variation (Figure 4Go). In DSC-MRI studies for instance the accuracy of the calculated cerebral blood volume is decreased if the signal to noise ratio in the temporal data is poor, or if the signal drop induced by bolus passage is small [13, 49, 50].

In clinical practice it is clearly impossible to assess the accuracy with which parametric variables reflect the true underlying changes in concentration of contrast agent since these are, by definition unknown. It is possible however to measure the signal to noise ratio and the magnitude of the signal drop and to derive a simple parameter that reflects the quantified uncertainty in the estimated perfusion parameters. If the changes in contrast concentration truly conform to the function which is being fitted, then the root mean fitting error between the calculated curve and the time course data provides an estimate of the uncertainty reflecting the signal to noise ratio in the temporal domain. For DSC-MRI the accuracy with which the {Gamma} variate function will reflect the underlying changes in gadolinium concentration will also be affected by the magnitude of the signal drop induced by bolus passage of contrast agent. Thus a very large signal drop will obviate for the effect of high background noise in the images. Given the scale similarity of {Gamma} variate curves, it is appropriate to scale the measurement of root mean fitting error with the area under the fitted curve to allow for variations in maximal signal difference. This scaled fitting error measurement (SFE) scales approximately linearly with the errors in the estimation of MTT and CBV and provides an important index of the reliability of individual parametric maps [14] (Figure 15Go). Since the SFE can be used to generate a parametric map of error (Figure 2Go) this can be used to control the inclusion of erroneous measurements of MTT or CBV into regions for analysis. A similar approach can be taken with pharmacokinetic analyses of DRC-MRI or CT.



View larger version (19K):
[in this window]
[in a new window]
  		Figure 15. Scattergrams of the results from a Monte Carlo simulation showing the relationship between calculated standard fitting error and the expected variation in mean cerebral blood volume (CBV) from images acquired with variable levels of image noise.

 

   Conclusions
Top
 Introduction
 Conclusions
References
 
Dynamic contrast enhanced MRI and CT imaging techniques are becoming more widely available in total practice with the increased availability of fast multislice CT machines and high specification MR systems. The analysis of interpretation of this dynamic data is complex and is complicated by the availability of multiple analysis algorithms and the fact that some calculated parameters may represent different biological effects depending on the algorithm applied. Despite this the clinical use of these techniques is growing and there are clear clinical benefit both in diagnosis and patient monitoring in a wide range of pathologies. For the clinician embarking on the use of these techniques it is essential to develop a basic understanding of the analysis algorithms being used and their relative advantages and potential pitfalls. A combination of appropriate imaging protocol and analysis algorithms can be confidently expected to give highly reproducible and precise physiological information out of form the basis for high-quality clinical implementations excellent and research applications.


   References
Top
 Introduction
 Conclusions
 References
 

  1. Barbier EL, Lamalle L, Decorps M. Methodology of brain perfusion imaging. J Magn Reson Imaging 2001;13:496–520.[CrossRef][Medline]
  2. Calamante F, et al. Measuring cerebral blood flow using magnetic resonance imaging techniques. J Cereb Blood Flow Metab 1999;19:701–35.[CrossRef][Medline]
  3. Ostergaard L, et al. Modelling cerebral blood flow and flow heterogeneity from magnetic resonance residue data. J Cereb Blood Flow Metab 1999;19:690–9.[CrossRef][Medline]
  4. Rosen BR, et al. Contrast agents and cerebral hemodynamics. Magn Reson Med 1991;19:285–92.[Medline]
  5. Sorensen A, Reimer P. Cerebral MR perfusion imaging: Principles and current applications. Stuttgart: Thieme, 2000:152.
  6. St Lawrence KS, Lee TY. An adiabatic approximation to the tissue homogeneity model for water exchange in the brain: I. Theoretical derivation. J Cerebral Blood Flow Metab 1998;18:1365–77.[CrossRef][Medline]
  7. Tofts PS, et al. Estimating kinetic parameters from dynamic contrast-enhanced T(1)-weighted MRI of a diffusable tracer: standardized quantities and symbols. J Magn Reson Imaging 1999;10:223–32.[CrossRef][Medline]
  8. Li KL, et al. Improved 3D quantitative mapping of blood volume and endothelial permeability in brain tumors. J Magn Reson Imaging 2000;12:347–57.[CrossRef][Medline]
  9. Keston P, Murray AD, Jackson A. Cerebral perfusion imaging using contrast-enhanced MRI. Clin Radiol 2003;58:505–13.[CrossRef][Medline]
  10. Ostergaard L, et al. Cerebral blood flow measurements by magnetic resonance imaging bolus tracking: comparison with [O15] H2O positron emmission tomography in humans. J Cereb Blood Flow Metab 1998;18:935–40.[CrossRef][Medline]
  11. Rempp KA, et al. Quantification of regional cerebral blood flow and volume with dynamic susceptibility contrast-enhanced MR imaging. Radiology 1994;193:637–41.[Abstract/Free Full Text]
  12. Calamante F, Gadian DG, Connelly A. Delay and dispersion effects in dynamic susceptibility contrast MRI: simulations using singular value decomposition. Magn Reson Med 2000;44:466–73.[CrossRef][Medline]
  13. Boxerman JL, Rosen BR, Weisskoff RM. Signal-to-noise analysis of cerebral blood volume maps from dynamic NMR imaging studies. J Magn Reson Imaging 1997;7:528–37.[Medline]
  14. Li K, Zhu X, Jackson A. Scaled error mapping in MR perfusion imaging. In: Proceedings of the second conference on medical image understanding and analysis. Leeds. 1998.
  15. Boxerman JL, et al. MR contrast due to intravascular magnetic susceptibility perturbations. Magn Reson Med 1995;34:555–66.[Medline]
  16. Sugahara T, et al. Perfusion-sensitive MR imaging of gliomas: comparison between gradient-echo and spin-echo echo-planar imaging techniques. AJNR Am J Neuroradiol 2001;22:1306–15.[Abstract/Free Full Text]
  17. Kassner A, et al. Abnormalities of the contrast re-circulation phase in cerebral tumors demonstrated using dynamic susceptibility contrast-enhanced imaging: a possible marker of vascular tortuosity. J Magn Reson Imaging 2000;11:103–13.[CrossRef][Medline]
  18. Aronen HJ, et al. Cerebral blood volume maps of gliomas: comparison with tumor grade and histologic findings. Radiology 1994;191:41–51.[Abstract/Free Full Text]
  19. Thacker NA, Scott ML, Jackson A. Can dynamic susceptibility contrast magnetic resonance imaging perfusion data be analyzed using a model based on directional flow? J Magn Reson Imaging 2003;17:241–55.[CrossRef][Medline]
  20. Rosen BR, et al. Perfusion imaging with NMR contrast agents. Magn Reson Med 1990;14:249–65.[Medline]
  21. Davenport R. The derivation of gamma variat relationship for tracer dilution curves. J Nucl Med 1983;24:945–8.[Abstract/Free Full Text]
  22. Benner T, et al. Accuracy of gamma-variate fits to concentration-time curves from dynamic susceptibility-contrast enhanced MRI: influence of time resolution, maximal signal drop and signal-to-noise. Magn Reson Imaging 1997;15:307–17.[CrossRef][Medline]
  23. Belliveau JW, et al. Functional imaging by gamma analysis of contrast enhanced NMR. In: 8th Annual meeting of the Society of Magnetic Resonance in Medicine. 1989.
  24. Maeda M, Maley JE, Crosby DL. Application of contrast agents in the evaluation of stroke: conventional MR and echo-planar MR imaging. J Magn Reson Med 1997;7:723–8.
  25. Kuhl CK, et al. Breast neoplasms: T2* susceptibility-contrast, first-pass perfusion MR imaging [see comments]. Radiology 1997;202:87–95.[Abstract/Free Full Text]
  26. Miyati T, et al. Dual dynamic contrast-enhanced MR imaging. J Magn Reson Imaging 1997;7:230–5.[Medline]
  27. Weisskoff RM, et al. Pitfalls in MR measurement of tissue blood flow with intravascular tracers: which mean transit time? Magn Reson Med 1993;29:553–8.[Medline]
  28. Lythgoe DJ, et al. Quantitative perfusion imaging in carotid artery stenosis using dynamic susceptibility contrast-enhanced magnetic resonance imaging. Magn Reson Imaging 2000;18:1–11.[CrossRef][Medline]
  29. Zhu XP, et al. Quantification of endothelial permeability, leakage space, and blood volume in brain tumors using combined T1 and T2* contrast-enhanced dynamic MR imaging. J Magn Reson Imaging 2000;11:575–85.[CrossRef][Medline]
  30. Tofts PS, Kermode AG. Measurement of the blood brain barrier permeability and leakage space using dynamic MR imaging: fundemental concepts. Magn Reson Med 1991;17:357–67.[Medline]
  31. Jayson GC, et al. Molecular imaging and biological evaluation of HuMV833 anti-VEGF antibody: implications for trial design of antiangiogenic antibodies. J Natl Cancer Inst 2002;94:1484–93.[Abstract/Free Full Text]
  32. Amoroso A, et al. Vascular endothelial growth factor: a key mediator of neoangiogenesis. A review. Eur Rev Med Pharmacolog Sci 1997;1:17–25.
  33. Folkman J. Role of angiogenesis in tumor growth and metastasis. Semin Oncol 2002;29(6 Suppl 16):15–8.
  34. Brasch R, et al. Assessing tumor angiogenesis using macromolecular MR imaging contrast media. J Magn Reson Imaging 1997;7:68–74.[Medline]
  35. Dvorak HF, et al. Vascular permeability factor/vascular endothelial growth factor, microvascular hyperpermeability, and angiogenesis. Am J Pathol 1995;146:1029–39.[Abstract]
  36. Jensen RL. Growth factor-mediated angiogenesis in the malignant progression of glial tumors: a review. Surg Neurol 1998;49:189–95; discussion 196.[CrossRef][Medline]
  37. Stack J, Redmond O, Codd M, et al. Breast disease: tissue characterization with Gd-DTPA enhancement profiles. Radiology 1990;174:491–4.[Abstract/Free Full Text]
  38. Flickinger F, et al. Differentiation of benign from malignant breast masses by time-intensity evaluation of contrast enhanced MRI. Magn Reson Imaging 1993;11:617–20.[CrossRef][Medline]
  39. Brix G, et al. Pharmacokinetic parameters in CNS Gd-DTPA enhanced MR imaging. J Comp Assist Tomog 1991;15:621–8.[Medline]
  40. Knopp MV, et al. Pathophysiologic basis of contrast enhancement in breast tumors. J Magn Reson Imaging 1999;10:260–6.[CrossRef][Medline]
  41. Knopp EA, et al. Glial neoplasms: dynamic contrast-enhanced T2*-weighted MR imaging. Radiology 1999;211:791–8.[Abstract/Free Full Text]
  42. Hawighorst H, et al. Uterine cervical carcinoma: comparison of standard and pharmacokinetic analysis of time-intensity curve for assessment of tumor angiogenesis and patient survival. Cancer Res 1998;58:3598–602.[Abstract/Free Full Text]
  43. Parker GJ, et al. Improving image quality and T(1) measurements using saturation recovery turboFLASH with an approximate K-space normalisation filter. Magn Reson Imaging 2000;18:157–67.[CrossRef][Medline]
  44. Haroon HA, et al. A comparison of Ktrans measurements obtained with conventional and first pass pharmacokinetic models in human gliomas. J Magn Reson Imaging 2004;19:527–36.[CrossRef][Medline]
  45. Tofts PS. Modeling tracer kinetics in dynamic Gd-DTPA MR imaging. J Magn Reson Imaging 1997;7:91–101.[Medline]
  46. Jackson A, et al. Breath-hold perfusion and permeability mapping of hepatic malignancies using magnetic resonance imaging and a first-pass leakage profile model. NMR Biomed 2002;15:164–73.[CrossRef][Medline]
  47. Li KL, Jackson A. New hybrid technique for accurate and reproducible quantitation of dynamic contrast-enhanced MRI data. Magn Reson Med 2003;50:1286–95.[CrossRef][Medline]
  48. Li KL, et al. Simultaneous mapping of blood volume and endothelial permeability surface area product in gliomas using iterative analysis of first-pass dynamic contrast enhanced MRI data. Br J Radiol 2003;76:39–50.[Abstract/Free Full Text]
  49. Boxerman JL, et al. Signal to noise and tissue blood volume maps from dynamic NMR imaging studies. In: 11th annual meeting of the Society of Magnetic Resonance in Medicine. 1992.
  50. Robson MD, Zhong J, Gore JC. Errors associated with fitting dynamic susceptibility contrast images for cerebral perfusion calculations. In: 13th annual conference of the Society of Magnetic Resonance in Medicine. 1994.



This article has been cited by other articles:


Home page
 RadiologyHome page
C. van der Leij, M. G. H. van de Sande, C. Lavini, P. P. Tak, and M. Maas
Rheumatoid Synovial Inflammation: Pixel-by-Pixel Dynamic Contrast-enhanced MR Imaging Time-Intensity Curve Shape Analysis--A Feasibility Study
Radiology, October 1, 2009; 253(1): 234 - 240.
[Abstract] [Full Text] [PDF]

Home page
 RadiologyHome page
E. S. Paulson and K. M. Schmainda
Comparison of Dynamic Susceptibility-weighted Contrast-enhanced MR Methods: Recommendations for Measuring Relative Cerebral Blood Volume in Brain Tumors
Radiology, November 1, 2008; 249(2): 601 - 613.
[Abstract] [Full Text] [PDF]

Home page
 Clin. Cancer Res.Home page
A. Jackson, J. P.B. O'Connor, G. J.M. Parker, and G. C. Jayson
Imaging Tumor Vascular Heterogeneity and Angiogenesis using Dynamic Contrast-Enhanced Magnetic Resonance Imaging
Clin. Cancer Res., June 15, 2007; 13(12): 3449 - 3459.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Jackson, A
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Jackson, A

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
BJR DMFR IMAGING  ALL BIR JOURNALS