Advances in Diffusion and Perfusion MRI for Quantitative Cancer Imaging

Abstract

Purpose of review:

This article is to review recent technical developments and their clinical applications in cancer imaging quantitative measurement of cellular and vascular properties of the tumors.

Recent findings:

Rapid development of fast Magnetic Resonance Imaging (MRI) technologies over last decade brought new opportunities in quantitative MRI methods to measure both cellular and vascular properties of tumors simultaneously.

Summary:

Diffusion MRI (dMRI) and dynamic contrast enhanced (DCE)-MRI have become widely used to assess the tissue structural and vascular properties, respectively. However, the ultimate potential of these advanced imaging modalities has not been fully exploited. The dependency of dMRI on the diffusion weighting gradient strength and diffusion time can be utilized to measure tumor perfusion, cellular structure, and cellular membrane permeability. Similarly, DCE-MRI can be used to measure vascular and cellular membrane permeability along with cellular compartment volume fractions. To facilitate the understanding of these potentially important methods for quantitative cancer imaging, we discuss the basic concepts and recent developments, as well as future directions for further development.

INTRODUCTION

MRI has been widely used for cancer diagnosis and assessment of treatment response, particularly with a Gadolinium-based contrast agent (GBCA). Contrast enhancement is a common feature of chaotic angiogenic vessels, a hallmark of cancer, which has been a crucial part of MRI-based detection of malignant lesions. It has also been used in assessment of tumor response to therapy for the development of new therapeutic strategies and for patient management. Volumetric assessment of tumors, including the Response Evaluation Criteria in Solid Tumors (RECIST) [1], has wide acceptance, but also severe limitations; volume change may not be appreciable even when therapy successfully halted tumor growth, or measurable volume change may not guarantee favorable response throughout the entire tumor. MRI has been used to map the internal treatment response of a tumor [2-4]. However, the ultimate potential of MRI, as a unique imaging modality that can probe the microstructural and functional properties of soft tissue noninvasively, has not been fully utilized for cancer imaging.

Among many MRI methods, diffusion MRI (dMRI) and dynamic contrast enhanced (DCE)-MRI using a GBCA have been two most commonly used advanced MRI methods. dMRI measures the diffusivity of endogenous water molecules in a tissue which reflects the mean size of the tissue microstructure that restricts and/or hinders Brownian random motion of water molecules. Previous studies have shown that this can be a powerful tool to detect densely populated cancer cells and their changes induced by therapy [2-4]. In contrast, DCE-MRI has been the choice of modality to assess the tumor vascular properties [5]. DCE-MRI typically uses a fast MRI method to continuously capture the signal intensity change during and after contrast injection into the circulation system, which contains rich information about the tumor vasculature and how well blood is delivered to the tumor. These two MRI methods, separately or combined, have been widely used for cancer diagnosis and monitoring treatment response.

While dMRI and DCE-MRI are promising imaging methods for cancer studies, however, it still remains challenging to use them as quantitative and specific imaging biomarkers in practice. The advantages and disadvantages of conventional dMRI and DCE-MRI methods are summarized in Table 1. The diffusivity measured by dMRI is a sensitive measure for tissue structural characteristics and their changes. But it is important to recognize that the measured diffusivity can be influenced by multiple measurement conditions, such as diffusion gradient strength and diffusion time. The biophysical meaning of what is actually measured as diffusivity can be quite different depending on the selection of those dMRI scan parameters. Similarly, the contrast kinetic parameters measured by DCE-MRI, such as transfer constant K^trans, can also varies depending on the acquisition method as well as a particular choice of contrast kinetic model. Furthermore, dMRI and DCE-MRI measurements are often affected by both cellular structure and perfusion properties, which should be considered in the experimental design when using these methods even for just one aspect of them. Alternatively, one of the methods can be used to measure both cellular structural and vascular perfusion properties simultaneously such that the scan time can be reduced if such potential can be fully exploited.

Table 1.

Advantages and disadvantages of conventional dMRI and DCE-MRI

Technique	Information Obtained	Advantages	Disadvantages
dMRI	Diffusion weighted images Tissue diffusivity (D)	Easy to interpret raw images with restricted diffusion in tumor Easy and automated post-processing Short acquisition time Widely available on most commerical scanners	Reduced spatial resolution Geometric distortion Limited specificity of D in conventional dMRI
DCE-MRI	Contrast enhanced images Transfer constant (Ktrans ) Plasma volume fraction (vp) Interstitial volume fraction (ve)	Easy to interpret raw images with contrast enhancement in tumor Quantitative assessment of tumor angiogenesis Low Geometric artefact High signal to noise ratio High spatial resolution	Requires Gd-based Contrast Long acquisition time Complexity in image acquisition Complexity of pharmacokinetic model post-processing Lack of widely available and easy-to-use post-processing software

In this work, we review the theoretical concepts as well as the recent advance in both dMRI and DCE-MRI methods toward making them as quantitative tools for robust and reproducible measurement of tumor microstructural and vascular properties. The review will also include the potential of using each modality to replace macroscopic tumor size criteria with more specific measure for cellular-level markers of cellular structure, perfusion, and cellular metabolism.

DIFFUSION MRI

The diffusional movement of water molecules in a tissue has a rich information about the structural properties of the tissue, which can be probed by dMRI. A diffusion weighted image is typically acquired by using a pair of gradients in opposite polarities applied with a time interval to allow water molecule diffusion [6] as shown in Figure 1a. The MRI signal is attenuated by the signal phase dispersion due to diffusion of water molecules in a tissue (D). The amount of attenuation is also related to the degree of diffusion weighting (b-value) which is determined by the diffusion weighting strength (q ~ δG) and diffusion time (t ~ Δ); b = q²t. In the absence of any obstacles, D characterizes free water diffusion due to random microscopic Brownian motion. In a biological tissue, such as tumor, D reflects various tissue microstructural properties that can restrict or hinder displacement of water molecules as well as exchange between compartments (Fig. 1b). A careful selection of these parameters can reveal different properties of the tissue, including vascularity, microstructure, and cellularity, as briefly summarized in this section [7].

Figure 1.

(a) Basic concept of the pulsed gradient spin echo (PGSE) dMRI experiment. A pair of magnetic field gradient pulses with amplitude G and duration δ are used to measure the random displacements of water molecules, particularly by self-diffusion. The first pulse is used to encode the phase offsets of water proton spins depending on their positions along the gradient pulse direction. It is followed by a delay of Δ during which the water molecules are left to diffuse freely in the tissue. Then the second pulse of gradient is applied in the opposite direction to realign the spin phases. Water molecule diffusion along the gradient field direction results in an imperfect re-alignment of the spin phases, leading to an attenuation of the signal magnitude. (b) Water diffusion in a tissue can be a sensitive tool to probe the tissue microstructure. The dMRI signal may reflects restricted diffusion with a confined space, such as cancer cells, or hindered diffusion in the extracellular space in between cells. It could also be affected by water exchange between the restricting structures, such as cell membranes.

In the case of simple Gaussian diffusion (e.g. in free water), dMRI measurement is characterized by a single parameter, the b-value, such that the dMRI signal decays as S=S₀exp(− bD). In contrast, tissue complexity gives rise to non-Gaussian diffusion [8], which makes the signal S depend on q and t separately, and is characterized by (i) the presence of the higher-order terms in the cumulant expansion of the signal [9], ln S/S₀ = −bD + (K/6)·(bD)² + … (such as the kurtosis term K), and (ii) the time-dependence of all the cumulants including D(t) and K(t) [10]. Hence, tissue complexity can be probed in two complementary directions [11]: (i) to quantify higher-order cumulants at a given diffusion time, by increasing q (or the b-value at fixed t), and (ii) to probe the time dependence of the cumulants by varying the diffusion time t as cumulants are the signal derivatives at b→0. Hence, depending on the microstructural property of interest, a proper combination of q and t needs to be selected in order to make the dMRI experiment most sensitive and specific to the target tissue property (Fig. 2).

Figure 2.

Two-dimensional parameter space of dMRI. Conventional dMRI methods, such as IVIM, DTI, and DKI, look at the change of D as a function of diffusion weighting gradient strength q at a fixed diffusion time t. Potentially, dMRI as a function of both gradient strength and diffusion time can provide opportunities to measure parameters more specific to tissue structural and vascular properties as discussed in the text.

Conventional dMRI

The most prevalent way to use dMRI for cancer imaging applications is to use a simple Gaussian mono-exponential model for diffusion weighted signal decay regardless of the structural complexity in which the measured D is often referred to as apparent diffusion coefficient (ADC) [7]. In this method, a D map is calculated by acquiring images with two different b-values in the range of approximately 200 to 1500 s/mm² [12] with a fixed diffusion time. Several studies have shown that the D is inversely correlated with the cell density or cellularity [13, 14]. For instance, the D value of pancreatic cancer has been shown to be lower than that of the normal pancreas [15-17], as a result of rapid tumor cell growth leading to high cellularity, or the decrease of extracellular space from dense cellularity and extracellular fibrosis that restricts water diffusion [17]. D was also successfully used for the differentiation between solid and cystic musculoskeletal tumors [18]. In breast imaging, dMRI is performed along with conventional contrast-enhanced breast MRI to reduce the false positives of lesion detection. Several studies showed that D values are significantly lower in malignant breast lesions than in benign lesions [19]. dMRI has also been used to assess the therapeutic response. Several studies reported early changes in tumor D after the first round of chemotherapy is significantly different between responders and nonresponders [20].

Non-Gaussian Diffusion

Non-monoexponential diffusion models offer multiple parameters, in addition to D, to account for tissue complexity, [21] that can be used in tumor detection and characterization. The non-Gaussian behavior is most evident at high b-values (e.g. >2000 s/mm²) (Fig.2). Among a number of non-Gaussian models introduced to date, Diffusional Kurtosis Imaging (DKI) has been used with promising results in cancer imaging. DKI uses diffusional kurtosis K, a dimensionless higher order term in the cumulant expansion of the diffusion weighted signal, to quantify the deviation of water molecular displacement from the Gaussian behavior [7, 22].

It has been demonstrated that K can be measured in the metastatic nodes of Head and neck squamous cell carcinoma (HNSCC) patients (1.53 ± 0.42) [23], prostate cancer (0.96 ± 0.24) [24] and hepatocellular carcinomas (1.18±0.50) [25]. Furthermore, Goshima et al. [26] showed that K has greater sensitivity and specificity than D (85.7% vs 79.6%, and 98.0% vs 68.3%, respectively) for the assessment of hepatocellular carcinoma viability after treatment. These studies suggest that K is a promising biomarker for evaluation of cancer treatment response, providing potentially complementary information to D. Moreover, it has been demonstrated that K is a more specific measure of tissue structure, such as cellular compartments and membranes, than D [22], and can be used as a biomarker for cell viability in addition to D.

Intra-Voxel Incoherent Motion

Cancer cell proliferation is accompanied by the abnormal vascular proliferation that supports cell growth. Movement of water molecules in blood plasma through randomly oriented capillaries can be considered as pseudo-diffusion and has been terms as intravoxel incoherent motion (IVIM) [27]. Such pseudo-diffusivity is typically higher than self-diffusivity of water molecules in biological tissues, and its effect can be observed with a faster signal decay in low b-values (approximately < 200 s/mm²) (Fig.2). The IVIM model consists of two exponential terms, D and pseudo-diffusion coefficient (D*), with perfusion volume fraction (f) for the vascular contribution in a voxel [27]. To capture both perfusion and diffusion contributions, the IVIM model requires more than three b-value, typically between 6 and 10 b-values, between 0 and 1000 s/mm² [7]. The IVIM model can be an efficient tool to capture both tissue and vascular properties.

Probing Tumor Microstructure using dMRI

While D is typically interpreted as an approximate measure of the cell density and extracellular water fraction, D and other commonly used dMRI metrics remain non-specific markers [28], because a diffusion measurement at a fixed diffusion time is affected by multiple factors, such as cell size, cell density, composition of the extracellular matrix, and compartmental diffusivities. Since cancer treatment can induce changes in these factors, it remains challenging to understand how these biophysical changes in the tumor tissue affect dMRI metrics. In order to address this ambiguity, the diffusion-time dependency of D, D(t), has been utilized to measure specific microstructural properties of cancer tissue with an assumption that the effect of water exchange between tissue compartments is negligible with the diffusion times used in the study.

The vascular, extracellular, and restricted diffusion for cytometry in tumors (VERDICT) model [29] was introduced to measure both cell size and tumor vasculature using diffusion times between 10 and 40 ms [30]. VERDICT uses a three-compartment tissue model to account for interstitial water network as well as water trapped in cells or in the vascular, in order to capture the main histological features that affect the DWI signals in tumors. There are typically six parameters to be estimated: intracellular volume fraction, extracellular volume fraction, cell size, pseudo-diffusivity, and two angles characterizing the directionality of vascular compartment. The free diffusivities in the intra- and extra-cellular spaces are fixed. The successful application of VERDICT was shown in distinguishing two human colorectal carcinoma cell lines (LS174T and SW1222) based on their vascular fraction [29] and in differentiating tumors from benign prostatic tissues in-vivo [31].

The imaging microstructural parameters using limited spectrally edited diffusion (IMPULSED) method was introduced to estimate cell sizes and intracellular volume fractions without using fixed diffusivities for intra- and extra-cellular spaces [32, 33]. IMPULSED employs short diffusion times using multiple oscillating gradient spin echo (OGSE) measurements in addition to relatively long diffusion times (~50 ms) using a conventional pulsed gradient spin echo (PGSE) method. This model assumes that the diffusivity varies linearly with the oscillating gradient frequency. This technique is based on a two-compartment model to include intra- and extracellular spaces in the estimation. The IMPULSED method was successfully used for estimating cancer cell size in vitro in the range of 5 to 10 μm for human leukemia cells [32] and for three types of human colon cancers in vivo. The in vivo result showed a stronger correlation of histology-derived cellularities with the IMPULSED-derived apparent cellularities than the conventional D [33].

Another promising approach is the pulsed and oscillating gradient MRI for assessment of cell size and extracellular space (POMACE) [34] framework that does not use a linear approximation of the diffusion in the short time regime. Instead, POMACE utilizes the universal behavior of diffusion when the diffusion length does not exceed the restriction scale, i.e., Mitra regime [35], in order to estimate the mean free diffusivity and surface-to-volume ratio (Fig.2). POMACE models the tumor microstructure in a time/frequency regime by impermeable spheres (intracellular diffusivity D_ics, radius R_cell) surrounded by extracellular space (ECS) which is approximated by constant apparent diffusivity D_ecs in volume fraction ECS. The POMACE framework was used for in vivo assessment of treatment response in GL261 gliomas and 4T1 mammary carcinomas and a significant ECS decrease was observed after the treatment [36].

Water Exchange in Diffusion MRI

The microstructural dMRI methods discussed above assume that their diffusion times are short enough to ignore the cellular-interstitial water exchange for simplicity and by carefully selecting a range of diffusion times to minimize the influence of the water exchange. However, the water exchange itself can be a useful parameter in cancer studies as it is closely related to interesting cellular properties, such as membrane permeability and size of the cells. Water molecules can cross the plasma membrane through specialized water-selective channels, known as aquaporins (AQP), as well as by simple diffusion through the plasma membrane although substantially slow [37]. In addition, water exchange can be facilitated by cell membrane ion-pump activity such that transcytolemmal water exchange rate may be a sensitive indicator of cellular energy turnover [38-40]. Measurement of cancer metabolism non-invasively during treatment may provide unique insight into how cancer responds to the treatment.

There have been several studies to use dMRI to measure water exchange between tissue compartments. Pfeuffer et al [41] used a strong diffusion weighting gradient with a long diffusion time that could suppress the signal from extracellular space and enabled to monitor the change of intracellular signal depending on the diffusion time. This method is often referred to as a constant gradient (CG) method and has been demonstrated to measure water exchange rate in rat brains [42]. A variant of double diffusion encoding method, also known as a filtered-exchange imaging (FEXI) method (Fig.3), has also been introduced to measure water exchange rate [43]. The FEXI method has been successfully used for breast cancer [44] and intracranial brain tumors [45]. A recent study used both CG and FEXI in perfused cells showed that both methods are sensitive to the changes in cell membrane permeability while FEXI appears to overestimate the exchange rate compared to the CG method [46].

Figure 3.

Filter exchange imaging (FEXI) is one of the dMRI techniques to measure water exchange between intra- and extra-cellular spaces. Shown are the representative images of a typical meningioma (top row), atypical meningioma (middle row), and a glioma (bottom row). The left column shows T1-weighted images, and the right column apparent exchange rate (AXR) parameter maps. The unit of the gray scale bar is s⁻¹. Adapted from [101] with permission.

Another clinically feasible method to measure cellular water exchange is to use time-dependent diffusional kurtosis imaging to measure K(t). The water exchange between intra- and extra-cellular comparts becomes independent of the position of water molecules within each compartment, i.e., the Kärger model, when the diffusion time is sufficiently long (Fig.2). In that Kärger model regime, an analytical relationship between K(t) and exchange time can be derived [47]. It has recently demonstrated that K(t) with t = 150-800 ms could be used to measure the exchange time of mouse glioma model which was between 110 and 380 ms [48].

While these studies showed the feasibility of using the diffusion time dependency of dMRI data to estimate specific cancer tissue properties including cell size and cellular water exchange time, their use in clinical application is largely limited by scan time and the need to use strong gradient depending on individual sites. With the rapid advance of MRI technology in recent years, it would soon become possible to utilize these types of advanced dMRI methods to measure specific microstructural features of cancer cells in basic science research as well as in clinical applications.

DYNAMIC CONTRAST ENHANCED MRI

Perfusion is physiologically defined as the steady-state delivery of blood at the capillary level to tissue [49] and it is related to the supply of oxygen and other nutrients to the tissue [50]. Perfusion MRI techniques include T₂/T₂*-weighted dynamic susceptibility contrast (DSC)-MRI and T₁-weighted DCE-MRI, which are acquired with exogenous GBCAs, and the arterial spin-labeling which is acquired without an exogenous contrast agent. Among them, DCE-MRI has been most commonly used to assess the vascular perfusion properties of cancer, particularly in clinical imaging studies for lesions outside the brain [5]. DCE-MRI essentially measures the change of tissue longitudinal relaxation rate (R₁=1/T₁) in a series of images acquired before, during and after the injection of GBCA (Fig. 4a). The dynamic time course of contrast enhancement in a voxel or region of interest data can be analyzed with semi-quantitative (model-free) or quantitative (model-based) approaches to estimate physiologically relevant parameters as shown in Figure 4b [51].

Figure 4.

(a) DCE-MRI experiment is conducted with a serial acquisition of 3D images while contrast agent is injection, such that the change of tissue longitudinal relaxation rate can be measured before, during and after the injection of the contrast agent. Post-contrast images typically show contrast enhancement in a lesion with increased vasculature and/or increased vascular leakiness. The red box shows an invasive ductal carcinoma in a 70 year old woman. The average time intensity curve of the enhancing lesion is shown in the plot. (b) An example schematic diagram of tissue microenvironment for DCE-MRI data analysis with physiologically relevant parameters including plasma flow (F_p), plasma volume fraction (v_p), vascular permeability surface area product (PS), interstitial volume fraction (v_e). The exchange of Gadolinium-based contrast agents (GBCA) in the vascular and interstitial spaces leads to changes in the longitudinal relaxation rates (R₁) in both compartments, as well as in the intracellular compartment through the water exchange that is often characterized by the intracellular water residence time constant (τ_i).

Semi-quantitative DCE-MRI

Conventional DCE-MRI approaches in most clinical exams use semi-quantitative model-free parameters to describe overall kinetics of signal enhancement curve [52]. The perfusion related quantities include the onset (lag or arrival) time (T_o), peak signal intensity (S_m), wash-in and wash-out slope, maximum intensity time ratio (MITR) and time-to-peak (T_p) signal enhancement ratio [53].

This approach is straightforward and simple. Its empirical parameters often change with diseases suggesting their correlation with the physiology of the organ [53]. However, it cannot estimate the specific physiological quantities of tumor vasculatures such as blood flow and vascular permeability [53]. Moreover, the signal intensity-based analysis is dependent on the MR acquisition protocols and hardware, such as scanner type and RF coil. Hence, it is difficult to compare the results acquired from different sites and/or different times unless the exactly same measurement condition is used [54].

Quantitative DCE-MRI

Time-intensity curves of DCE-MRI contain rich information about the tissue microcirculation environment. But it is nontrivial to extract quantitative tissue information from the dynamic curves. Quantitative DCE-MRI requires careful consideration for fast data acquisition, accurate signal-to-concentration conversion, and selection of a proper contrast kinetic model for the given tissue of interest. It is often a challenging task to include all these considerations in implementation of a quantitative DCE-MRI study, particularly in clinical exams with limited scan times. Thanks to the recent development of fast acquisition techniques, however, it has become more feasible to address these challenges. Some of advances in the related areas are discussed in the following sections.

Data Acquisition Methods

There have been many approaches to balancing requirements for high temporal versus high spatial resolution in DCE-MRI. Despite fast development of scanner hardware, current state-of-the-art hardware alone cannot fulfill the ever-increasing demand for higher spatial and temporal resolution. The approach that has been widely accepted now is to take advantage of information redundancy in images to extract higher spatial and temporal resolution information from under sampled data [55]. Some examples of this method include keyhole imaging [56], view sharing [57], and compressed sensing [58].

Over the last decade, compressed sensing [58] has emerged as a powerful tool for rapid imaging. It utilizes image compressibility/sparsity to reduce the number of samples required to reconstruct an image without loss of important information. It has been shown that compressed sensing can be combined with parallel imaging to further increase imaging speed by exploiting joint sparsity in the ensemble of images from multiple coil elements [59]. Parallel imaging provides additional encoding capabilities to reduce the incoherent aliasing artifacts and improve the performance of compressed sensing [59].

Furthermore, it has been demonstrated that radial k-space sampling is an attractive alternative to conventional Cartesian sampling for compressed sensing, due to the inherent presence of incoherent aliasing artifacts along all spatial dimensions, even for regular under-sampling [60]. A highly-accelerated dynamic MRI technique called Golden-angle RAdial Sparse Parallel (GRASP) MRI [61] is one of the examples that jointly combine compressed sensing, parallel imaging and golden-angle radial sampling for rapid imaging [62]. GRASP offers the flexibility of reconstructing temporal frames with user-defined temporal resolution, frame position, and number of frames, from the same raw data. GRASP has been successfully demonstrated in clinical dynamic MRI of the abdomen, breast (Fig. 5), head and neck, and heart with significantly improved imaging performance than conventional MRI [63]. The GRASP method has been also extended with a 3D ultra-short echo-time (UTE) pulse sequence which can provide an isotropic high-spatial resolution and negligible T₂* effect following contrast injection [64]. Recent development of deep learning-based imaging reconstruction [65] may provide further improvement in fast data acquisition of good quality images for DCE-MRI with sufficiently high spatial and temporal resolutions.

Figure 5.

A representative example of one frame from a breast DCE-MRI using GRASP that utilizes both compressed sensing and parallel imaging in order to provide dynamic images with high temporal (2.5 s/frame) and high spatial (1.1 × 1.1 × 1.1 mm³) resolutions. The images on the left column are example sagittal (top) and axial (bottom) images from one 3D frame for 3 min after injection. The images on the right columns are the maximum intensity projections on the sagittal (top) and axial (bottom) planes from the same 3D frame for 3 min after injection.

Signal-to-Concentration Conversion

Contrast enhancement in a tissue depends not only on the amount of contrast agent in the tissue, but also on the pre-contrast tissue longitudinal relaxation time T₁₀ [66, 67], such that analyses of DCE-MRI data based on time-intensity curves without taking into account of potential T₁₀ variability in lesions could result in a limited diagnostic accuracy [68]. For a quantitative contrast kinetic model analysis, DCE-MRI time-intensity curves are converted to contrast agent concentration curves. Such signal-to-concentration conversion process requires T₁₀ values which can be measured using various methods, such as inversion recovery [69] and variable flip angle methods [70], at a cost of extra scan time.

In addition, accurate T₁ mapping typically requires correction for the inhomogeneous radiofrequency (RF) transmit field (B₁) which determines the actual flip angle (FA) [71-73]. Various B₁ mapping methods have been developed based on either magnitude [72] or phase [74] images made sensitive to the B₁ field. Most of these B₁ mapping methods require either a long repetition time (TR) to minimize the influence of T₁) or an extra measurement of the B₀ field to minimize the off-resonance effect, leading to further increase of scan time. Furthermore, the actual FA can also be affected by other factors, such as slice profile [75] and RF amplifier nonlinearity [76].

Hence, the scan time required for these additional measurements of T₁₀ and B₁ can often be similar to or longer than the actual DCE-MRI scan itself [71], particularly when the target imaging volume is large as in body imaging and breast imaging. Given a limited scan time in most clinical scans, further development for fast B₁ and T₁₀ measurements is necessary to conduct a quantitative DCE-MRI experiment. Simultaneous B₁ and T₁) mapping using MR fingerprinting may provide a more practical solution to this challenge in near future [77].

Contrast Kinetic Model-based Analysis

Quantitative approaches fit various mathematical contrast kinetic models to the dynamically acquired tissue concentration curves [53] and describe the contrast exchange between the intra- and extravascular space with physiologically relevant parameters (Fig.4b) [78]. The Tofts model, also known as general kinetic model [78], and its extended version with a vascular component are the most commonly used ones in clinical applications [79] and have been applied extensively to characterize the brain [80], lung [81, 82], breast [83], prostate [84], liver [85], and colorectal tumors [86]. These two-compartment models provide estimates of the volume transfer constant (K^trans) as the leakage rate of GBCAs from the blood plasma towards extravascular extracellular space (EES), EES volume fraction (v_e) and fractional plasma volume per unit of tissue volume (v_p). The transfer constant from EES into blood plasma is defined as K_ep = K^trans/v_e [52, 79].

One of the key assumptions used in the Tofts model is that the contrast concentration in the tissue capillary bed is same as that in the artery where the arterial input function (AIF) is measured. This could be a valid assumption when the blood plasma flow F_p is high. But in some part of a tumor, F_p can be lowered due to increased interstitial fluid pressure [87]. Hence, one can use more generalized model that include F_p as a parameter to determine how GBCA is transferred from the artery to the capillary compartment, and vascular permeability surface area product PS as a parameter for GBCA transfer between the capillary compartment and EES. Such model is known as the two compartment exchange model [88] (Fig.4b). It has also been shown that the change of GBCA concentration within the capillary bed can be included by using more sophisticated models, such as the adiabatic approximation to the tissue homogeneity model [89].

Along with a proper selection of contrast kinetic model, it should also be noted that accurate measurement of AIF is crucial for the estimation of contrast kinetic model parameters. A recent study by the Quantitative Imaging Network (QIN) founded by the National Cancer Institute [90] well demonstrated the impact of AIF on contrast kinetic analysis of prostate cancer DCE-MRI data from 11 subjects. The variability among the centers was substantially decreased when the reference-tissue adjusted AIF was used. This study substantiates the need to have an accurate estimation of AIF, particularly in terms of the GBCA concentration in the plasma.

It is also noteworthy that estimation of contrast kinetic model parameters can be performed simultaneously with estimation of T₁₀ and B₁ if the dynamic scan were performed with appropriate contrast encodings with multiple flip angles and repetition times [91]. This technique is referred to as active contrast encoding (ACE)-MRI. In this framework, estimation of T₁₀ and B₁ maps from ACE-MRI data can be performed based on the assumption that the contrast concentration curve is continuous regardless of the contrast encoding. The actual flip angle achieved in a DCE-MRI scan and T₁ mapping can be affected by various factors, such as RF coil transmit field B₁ inhomogeneity [72], RF pulse profile [75], nonlinearity of RF amplifier [76], tissue properties [73], and imperfect spoiling in steady-state imaging [92]. The B₁ estimation in ACE-MRI can include all these effects since the estimation is performed with the actual DCE-MRI data, as opposed to using a separate pulse sequence for B₁ mapping (Fig. 6). ACE-MRI effectively introduced the MR fingerprinting concept in DCE-MRI and has a potential to simplify the complex procedure of DCE-MRI data acquisition and analysis, while providing accurate and quantitative contrast kinetic model parameters.

Figure 6.

Active Contrast Encoding (ACE)-MRI can measure T1 and B1 maps simultaneously with contrast kinetic parameters as shown by a representative example of a 4T1 tumor on the flank. (A) Axial slice with ROIs for reference muscle (blue) and enhancing lesion rim (red). (B) AIF generated from ACE- MRI (red full line) and DCE- MRI using the measured T₁/B₁ for the reference muscle region. (C) Lesion (black crosses) and muscle (blue crosses) enhancement curves and the corresponding fits (green full lines). (D, E) Comparison of parameter color maps between DCE- MRI and ACE- MRI methods. For the DCE- MRI method, T₁₀ and B₁ were measured separately from the RARE- VTR sequence and signal null method. Adapted from [91] with permission.

Water Exchange Effect in DCE-MRI

As discussed above for the water exchange in diffusion MRI, the cellular-interstitial water exchange can be a sensitive indicator of cellular energy turnover [38-40]. DCE-MRI is another potential tool to measure such water exchange effect. One of the challenges in accurate analysis of DCE-MRI data is that the contrast agent concentration is indirectly measured via tissue MR relaxivity measurement and the complex nature of the tumor microenvironment. Tissue relaxivity is a function of GBCA exchange between the vascular and extravascular-extracellular compartments, as well as of water exchange between the extravascular-extracellular and intracellular compartments. Water exchange terms have previously been included in the analysis with the Tofts models [67, 93-96]. However, it has been shown that the uncertainty in the estimation of the water exchange effect can be considerably high with conventional scan protocols [97]. Furthermore, a recent study also demonstrated that the vascular contribution could be misinterpreted as an effect of transcytolemmal water exchange such that a proper contrast kinetic model needs to be used in order to estimate the water exchange effect adequately [98]. Spencer and Fishbein [99] showed that the effect of water exchange depends on flip angle and repetition time. Hence, having more than one scan protocol (i.e., flip angle and repetition time) for a dynamic scan session could make it possible to measure water exchange effect from the unique contrast pattern depending on the choice of flip angle and repetition time [100]. Such method would make DCE-MRI a comprehensive imaging tool for both vascular and cellular structural properties simultaneously.

CONCLUSION

In conclusion, dMRI and DCE-MRI have become important techniques with applications in various areas of cancer imaging including diagnosis, tumor grading, and treatment response evaluation and prediction. The rapid development of new diffusion and perfusion techniques along with the advances in MR hardware and emerging new microstructure models have shown a promising trend to provide dMRI and DCE-MRI as quantitative and clinical imaging tools to study tumor heterogeneity, vascularity, cellularity, and microstructural properties.