**4. Perfusion magnetic resonance imaging**

Clinical perfusion MRI measurements are currently recognized as one of the most powerful tools to assess tumor vascularisation and responses to treatment through the evaluation of hemodynamic parameters. MRI approaches to measure perfusion, can be divided in two broad classes: those monitoring tissue signal changes using an exogenous relaxation contrast agent and those that use endogenous contrast. MR perfusion weighted imaging (PWI) allows for the noninvasive estimation of tumor grade preoperatively. This can help by directing the surgeon to biopsy the most aggressive portions of the neoplasm. In addition, perfusion MR provides an exquisite delineation of tumor margins and it is useful in the follow-up of brain tumor patients after treatment by allowing differentiation between radiation effects and recurrent tumor zones. Looking into the future, perfusion changes in brain tissues hold promise as surrogate markers of response to therapy in clinical trials of new antiangiogenic drugs.

#### **4.1 Physiological principles**

Histopathology as gold standard of tumor grading suffers of several limitations. Mainly, since just a few samples of tissue are obtained after biopsy, the most malignant regions may be missed. Neovascularization is one of the most important criteria of malignancy for glioma grading. Malignant astrocytomas recruit existing vessels in the microenvironment and lead to form new vascular networks to supply the nutrients and oxygen required for tumor growth and proliferation. Hence, it is widely accepted nowadays the importance of neoangiogenic pathways in the treatment of brain tumors, since tumor growth of a few millimeters, involves the development of sufficiently competent networks of vascular supplies (Folkman 1971). The net result is a complex system of abnormal vessels in the peritumoral space, characterized by abnormal blood flow and increased permeability to macromolecules, both factors being potentially used as biomarkers for the evaluation of tumor growth. Concomitantly, tumor vessels are more tortuous than the healthy brain, increasing the distance that blood must travel as it moves through the tumor. Perfusion abnormalities are therefore detected not only because the increased number of vessels in gliomas but also for the aberrant reaction of those vessels to the environment. Under these circumstances, disruption of angiogenesis plays an important role in established therapies of gliomas and in the development of new ones.

#### **4.2 Techniques and methods**

236 Advances in the Biology, Imaging and Therapies for Glioblastoma

through the early detection of induced changes in their morphology (Moffat, Chenevert et al. 2006; Hamstra, Galban et al. 2008). A patient with a brain tumor that shows significant changes in its ADC values after treatment is supposed to be responsive to therapy. There is the hypothesis that the water liberated by cell necrosis as induced by therapy is the major mechanism for the induced ADC increase. So, the magnitude of change in tumor water diffusion as detected by MRI is related to the quantity of cells killed by therapy and hence, to its efficacy. These changes precede the volume changes in the tumor by several weeks, providing the main justification for DWI as an early predictor of treatment response in individual patients. Even more, no changes in diffusion coefficient indicate a therapeutically unresponsive tumor suggesting the prescription of an alternative

Clinical perfusion MRI measurements are currently recognized as one of the most powerful tools to assess tumor vascularisation and responses to treatment through the evaluation of hemodynamic parameters. MRI approaches to measure perfusion, can be divided in two broad classes: those monitoring tissue signal changes using an exogenous relaxation contrast agent and those that use endogenous contrast. MR perfusion weighted imaging (PWI) allows for the noninvasive estimation of tumor grade preoperatively. This can help by directing the surgeon to biopsy the most aggressive portions of the neoplasm. In addition, perfusion MR provides an exquisite delineation of tumor margins and it is useful in the follow-up of brain tumor patients after treatment by allowing differentiation between radiation effects and recurrent tumor zones. Looking into the future, perfusion changes in brain tissues hold promise as surrogate markers of response to therapy in clinical trials of

Histopathology as gold standard of tumor grading suffers of several limitations. Mainly, since just a few samples of tissue are obtained after biopsy, the most malignant regions may be missed. Neovascularization is one of the most important criteria of malignancy for glioma grading. Malignant astrocytomas recruit existing vessels in the microenvironment and lead to form new vascular networks to supply the nutrients and oxygen required for tumor growth and proliferation. Hence, it is widely accepted nowadays the importance of neoangiogenic pathways in the treatment of brain tumors, since tumor growth of a few millimeters, involves the development of sufficiently competent networks of vascular supplies (Folkman 1971). The net result is a complex system of abnormal vessels in the peritumoral space, characterized by abnormal blood flow and increased permeability to macromolecules, both factors being potentially used as biomarkers for the evaluation of tumor growth. Concomitantly, tumor vessels are more tortuous than the healthy brain, increasing the distance that blood must travel as it moves through the tumor. Perfusion abnormalities are therefore detected not only because the increased number of vessels in gliomas but also for the aberrant reaction of those vessels to the environment. Under these circumstances, disruption of angiogenesis plays an important role in established therapies of

therapeutic option.

new antiangiogenic drugs.

**4.1 Physiological principles** 

gliomas and in the development of new ones.

**4. Perfusion magnetic resonance imaging** 

Currently, there are three principal methods to measure hemodynamic perfusion parameters by using MRI: T2\*-weighted DSC, ASL, and T1-weighted DCE acquisitions, respectively. Any of them involves the serial acquisition of images through the brain as magnetically labelled blood passes through it. The labelling procedure can be either based on an exogenous contrast material, typically a paramagnetic complex of gadolinimum (Gd), or on an endogenous magnetic label. Among them, T2\*W dynamic susceptibility imaging is probably the most consolidated method to carry out MR perfusion in clinical environments not only in brain oncology but also in many other pathologies, mainly neurodegenerative diseases.

#### **4.2.1 Dynamic Susceptibility Contrast**

Nowadays, PWI employing DSC techniques is one of the most relevant procedures of functional MRI with an increasing number of clinical applications depending on software and hardware availability. DSC enhanced images allow for the evaluation of regional cerebral hemodynamics by analyzing changes in the intensity of the MR signal during the first pass (and often second) of a narrow and concentrated bolus of an exogenous CA, through the brain microvascular bed. This technique is often called *bolus tracking* and performs perfusion measurements employing very rapid imaging to capture the first pass of a rapidly injected intravenous contrast material (Rosen, Belliveau et al. 1990), exploiting the T2\* susceptibility effects induced by the large local concentrations of the agent rather than the T1 shortening effects routinely used in contrast enhancement by conventional MRI. For accurate assessment of the circulation it is necessary to measure signal drop during the passage of the bolus with sequences endowed with high sensitivity and temporal resolution to the magnetic susceptibility changes occurring in the tissue during the pass of the agent. One approach is to use EPI sequences enabling acquisition times of less than one second because the whole raw data set of an image is acquired after a single spin excitation. EPI is merely a read-out technique and can be combined with any technique of spin excitation (such as spin-echo or gradient echo). Due to the special readout scheme of EPI, and because major parts of k-space are affected by T2\*, each EPI sequence is inherently T2\* sensitive.

In the brain, the first-pass extraction of the agent is zero when the blood-brain barrier (BBB) is intact, and the complete intravascular compartmentalization of the contrast creates strong, microscopic magnetic susceptibility gradients. These gradients cause dephasing of spins as the spins diffuse through the vessel. In general, pulse sequences without complete refocusing of static field inhomogeneities, as such gradient-echo (GE) will suffer a general signal loss due to the presence of microscopic field disturbances in the vessels. In terms of relaxivity, the shortening of T1 values affects only those protons near the unpaired electron clouds of the agent. This is a short-range effect. A distinct feature of the signal loss on T2\*W images is that all protons near the CA are affected because the magnetic susceptibility effect spreads far beyond the immediate vicinity of the agent. This is considered a long-range effect. However, direct damage of the BBB in a tumor may cause a disruption in it and a leakage of contrast material, a circumstance indicating that the factors responsible of the observed MR signal may be more complex.

Experimentally, the contrast medium at high concentrations (narrow bolus), induces T2\* shortening yielding firstly a signal loss, followed by a recovery as the agent is diluted in

The higher vascularity of gliomas is usually assessed with PWI methods by measuring CBV in the tumor. A potential problem with the use of these techniques arises in regions of severe BBB disruptions as it happens in high-grade gliomas. The leakage of the contrast medium into the extravascular space during its first passage across the tumor capillary bed, increases signals above baseline due to the T1 shortening effect of the paramagnetic label used. Since the algorithm used to calculate CBV assumes a constant baseline, this signal increasing relies in an area above the concentration curve of CA versus time (left panel in figure 8) interpreted as a negative blood volume. This may yield a significant underestimation of CBV in gliomas (Aronen and Perkio 2002). Several improvements and multiple methods have been proposed to address this problem. One of the most extensively used in clinical settings is the presaturation of the extravascular extracellular space in the lesion with the preinjection of a low dose of CA (0.05 mmol/Kg) before starting perfusion MR acquisition with a normal dose of Gd (typically 0.2 mmol/Kg) intravenously injected at a high rate. Another approach consists in using a dysprosium (Dy) based contrast medium as a paramagnetic metal with stronger T2\* effects than Gd but only slight capability in T1 shortening (Vander Elst, Roch et al. 2002). A third option to reduce T1 effects goes through increasing the repetition time (TR) which reduces T1 effects but increases total acquisition

Fig. 9. Correction of CBV underestimation in a high-grade glioma. A: T1W image (Gd 0.2 mmol/Kg) shows an enhanced mass. B: Uncorrected CBV map shows low CBV within the lesion (black arrow). C: K2 first-pass permeability map shows increased permeability within the lesion, consistent with severe BBB breakdown. D: Corrected CBV map shows increased CBV within the lesion, as expected for a high-grade neoplasm (Covarrubias, Rosen et al.

In addition, different computational approaches have been reported to avoid the CBV underestimation. In these methods, the images are mathematically corrected to account for extravasation of contrast material in regions of BBB breakdown under the premise that the observed Gd concentration can be divided in intra- and extravascular components (Aronen and Perkio 2002; Provenzale, Wang et al. 2002). With these methods, based in a twocompartmental model, permeability can be estimated in regions of BBB disruption by generation of permeability (K2) maps that can be used to correct CBV as shown in figure 9. Although DSC has the potential to allow quantitative imaging of perfusion parameters at a high signal to noise ratio in a short scan time, quantification of MTT and CBF depends critically on accurate and precise measurement of an arterial input function (AIF). The AIF relies on the measurement of the signal response in the artery or arteries that feed the tissue

time, thus reducing temporal resolution.

2004) Reproduced with permission.

tissues (i.e. healthy brain and tumor). So, the bolus can be tracked by the acquisition of a multislice series of T2\*-sensitive images from which a graph of relative signal intensity against time can be composed showing the pixel by pixel effects of the CA transit in the image. There is an approximate linear relationship between tissue contrast agent concentration and change in T2\* relaxation rate, and from the signal-time course, the concentration-time course of the contrast medium can be calculated with the expression:

$$\text{Ct (t)} = \text{-1/TE [In (St-S\_0)]} \tag{2}$$

where Ct(t) is the concentration of the agent, St is the signal intensity at any time point, S0 is the basal signal intensity before the CA injection and TE the echo time. The concentration of contrast medium is linearly proportional to changes in rate of relaxation T2\* (R2\*). MR data can then be converted to R2\* versus time evolution and these values can be fitted to a gamma variate function to correct for tracer recirculation. Assuming uniform arterial concentration profiles in all arterial inputs, relative cerebral blood volume (CBV) measurements are determined simply by integrating the area under the concentration time curve (figure 8). The peak integral (that is, the peak area under the curve) is taken as the relative CBV because this represents the volume of the effect. Other parameters of interest that can be measure in the curve are the time to peak (TTP) and the mean transit time (MTT), although the true MTT requires knowledge of the input function of the bolus. Figure 8 depicts a simulation of the process.

Fig. 8. Schematic behaviour of MR signal intensity as CA passes across the imagin plain in brain tissue

In this way, post-processing of DSC enhanced MR images allows mapping CBV, cerebral blood flow (CBF) and MTT by using singular value decomposition and deconvolution (Ostergaard, Weisskoff et al. 1996). Regional CBV (rCBV) is defined as the total volume of blood traversing a given brain region, measured in milliliters of blood per gram of brain tissue (mL/g). Regional CBF (rCBF) is defined as the net volume of blood traversing a given brain region per unit time, measured in milliliters of blood per gram of brain tissue per minute (mL/g/min). The definition of regional MTT is more complex, but it can be described as the average amount of time it takes any water molecule or particle of CA to pass through the capillary bed, that is, to transverse between arterial inflow and venous outflow measured in seconds (s) (Covarrubias, Rosen et al. 2004). rMTT is equal to the ratio rCBV/rCBF and although CBV may be measured with other methods, MTT measurements necessarily rely on techniques sensitive to motion.

tissues (i.e. healthy brain and tumor). So, the bolus can be tracked by the acquisition of a multislice series of T2\*-sensitive images from which a graph of relative signal intensity against time can be composed showing the pixel by pixel effects of the CA transit in the image. There is an approximate linear relationship between tissue contrast agent concentration and change in T2\* relaxation rate, and from the signal-time course, the concentration-time course of the contrast medium can be calculated with the expression:

where Ct(t) is the concentration of the agent, St is the signal intensity at any time point, S0 is the basal signal intensity before the CA injection and TE the echo time. The concentration of contrast medium is linearly proportional to changes in rate of relaxation T2\* (R2\*). MR data can then be converted to R2\* versus time evolution and these values can be fitted to a gamma variate function to correct for tracer recirculation. Assuming uniform arterial concentration profiles in all arterial inputs, relative cerebral blood volume (CBV) measurements are determined simply by integrating the area under the concentration time curve (figure 8). The peak integral (that is, the peak area under the curve) is taken as the relative CBV because this represents the volume of the effect. Other parameters of interest that can be measure in the curve are the time to peak (TTP) and the mean transit time (MTT), although the true MTT requires knowledge of the input function of the bolus. Figure

Fig. 8. Schematic behaviour of MR signal intensity as CA passes across the imagin plain in

In this way, post-processing of DSC enhanced MR images allows mapping CBV, cerebral blood flow (CBF) and MTT by using singular value decomposition and deconvolution (Ostergaard, Weisskoff et al. 1996). Regional CBV (rCBV) is defined as the total volume of blood traversing a given brain region, measured in milliliters of blood per gram of brain tissue (mL/g). Regional CBF (rCBF) is defined as the net volume of blood traversing a given brain region per unit time, measured in milliliters of blood per gram of brain tissue per minute (mL/g/min). The definition of regional MTT is more complex, but it can be described as the average amount of time it takes any water molecule or particle of CA to pass through the capillary bed, that is, to transverse between arterial inflow and venous outflow measured in seconds (s) (Covarrubias, Rosen et al. 2004). rMTT is equal to the ratio rCBV/rCBF and although CBV may be measured with other methods, MTT measurements

8 depicts a simulation of the process.

necessarily rely on techniques sensitive to motion.

brain tissue

Ct (t) = -1/TE [ln (St-S0)] (2)

The higher vascularity of gliomas is usually assessed with PWI methods by measuring CBV in the tumor. A potential problem with the use of these techniques arises in regions of severe BBB disruptions as it happens in high-grade gliomas. The leakage of the contrast medium into the extravascular space during its first passage across the tumor capillary bed, increases signals above baseline due to the T1 shortening effect of the paramagnetic label used. Since the algorithm used to calculate CBV assumes a constant baseline, this signal increasing relies in an area above the concentration curve of CA versus time (left panel in figure 8) interpreted as a negative blood volume. This may yield a significant underestimation of CBV in gliomas (Aronen and Perkio 2002). Several improvements and multiple methods have been proposed to address this problem. One of the most extensively used in clinical settings is the presaturation of the extravascular extracellular space in the lesion with the preinjection of a low dose of CA (0.05 mmol/Kg) before starting perfusion MR acquisition with a normal dose of Gd (typically 0.2 mmol/Kg) intravenously injected at a high rate. Another approach consists in using a dysprosium (Dy) based contrast medium as a paramagnetic metal with stronger T2\* effects than Gd but only slight capability in T1 shortening (Vander Elst, Roch et al. 2002). A third option to reduce T1 effects goes through increasing the repetition time (TR) which reduces T1 effects but increases total acquisition time, thus reducing temporal resolution.

Fig. 9. Correction of CBV underestimation in a high-grade glioma. A: T1W image (Gd 0.2 mmol/Kg) shows an enhanced mass. B: Uncorrected CBV map shows low CBV within the lesion (black arrow). C: K2 first-pass permeability map shows increased permeability within the lesion, consistent with severe BBB breakdown. D: Corrected CBV map shows increased CBV within the lesion, as expected for a high-grade neoplasm (Covarrubias, Rosen et al. 2004) Reproduced with permission.

In addition, different computational approaches have been reported to avoid the CBV underestimation. In these methods, the images are mathematically corrected to account for extravasation of contrast material in regions of BBB breakdown under the premise that the observed Gd concentration can be divided in intra- and extravascular components (Aronen and Perkio 2002; Provenzale, Wang et al. 2002). With these methods, based in a twocompartmental model, permeability can be estimated in regions of BBB disruption by generation of permeability (K2) maps that can be used to correct CBV as shown in figure 9. Although DSC has the potential to allow quantitative imaging of perfusion parameters at a high signal to noise ratio in a short scan time, quantification of MTT and CBF depends critically on accurate and precise measurement of an arterial input function (AIF). The AIF relies on the measurement of the signal response in the artery or arteries that feed the tissue

Fig. 10. DCE images of a patient with histologycally proved astrocytoma grade III. A: Post Gd-injection T1W image. B: T2W image. D: T1W image (without CA administration). C:

ASL perfusion imaging in the brain adds a new functional dimension to non invasive evaluation of the brain functionality with MRI. It is the only method that yields quantitative monitoring of cerebral perfusion over a period of several hours. Its non invasive nature allows an unlimited number of repeated measurements to be made with a high temporal resolution. ASL uses spatially selective inversion of the inflowing arterial blood magnetization as a method to label water molecules in blood. So, in a typical experiment the arterial blood is tagged and after a given delay, this tagged blood arrives at the imaging plane and an image is acquired. The MRI signal from inverted magnetization is made negative relative to uninverted one. When the labelled blood reaches the tissue, it attenuates the pre-existing signal from the image of that tissue. Therefore, a control measurement is required without tagging the arterial blood. The difference in signal intensity between both images is proportional to rCBF, through a calibration factor that depends primarily on global properties of blood. A schematic diagram of the process is shown in figure 11. This difference signal measures how much of the original arterial magnetization created by the inversion pulse has been delivered to the voxel and survives to the time of measurement. The magnetization difference (control minus taged) carried out into the voxel can be conceptually similar to a contrast material delivered to it by CBF. Under this assumption the same kinetic modelling arguments used in bolus tracking or DSC studies can be applied to ASL (Buxton, Frank et al. 1998). Nevertheless, several premises remain to be taken into account for an accurate

CBV and E: CBF maps generated

**4.2.3 Arterial Spin Labelling** 

of interest. This is one factor that limits the ability to quantitate hemodynamic parameters from DSC studies. To measure AIF the imaging planes through the brain have to be chosen to contain a major intracranial artery feeding the tissue, typically one paraclinoid internal carotid artery or the middle cerebral artery (Covarrubias, Rosen et al. 2004). Then, a deconvolution algorithm is used to post-process the data and generate CBF and MTT maps. However, the absolute determination of hemodynamic parameters in not a crucial factor in grading human gliomas, the resulting ratios being equally valid for this purposes since they incorporate the measurement errors from both values.
