Quantitative Imaging Biomarkers Alliance (QIBA) Recommendations for Improved Precision of DWI and DCE-MRI Derived Biomarkers in Multicenter Oncology Trials

Abstract

Physiological properties of tumors can be measured both in vivo and non-invasively by diffusion-weighted imaging and dynamic contrast-enhanced magnetic resonance imaging. Although these techniques have been used for more than two decades to study tumor diffusion, perfusion, and/or permeability the methods and studies on how to reduce measurement error and bias in the derived imaging metrics is still lacking in the literature. This is of paramount importance because the objective is to translate these quantitative imaging biomarkers (QIB) into clinical trials, and ultimately in clinical practice. Standardization of the image acquisition using appropriate phantoms is the first step from a technical performance standpoint. The next step is to assess whether the imaging metrics have clinical value and meet the requirements for being a QIB as defined by the Radiological Society of North America’s Quantitative Imaging Biomarkers Alliance/(QIBA®). The goal and mission of QIBA and the National Cancer Institute Quantitative Imaging Network (QIN) initiatives are to provide technical performance standards (QIBA profiles) and QIN tools for producing reliable QIBs for use in the clinical imaging community. Some of QIBA’s development of quantitative diffusion-weighted imaging and dynamic contrast-enhanced QIB profiles has been hampered by the lack of literature for repeatability and reproducibility of the derived QIBs. The available research on this topic is scant and is not in sync with improvements or upgrades in magnetic resonance imaging technology over the years. This review focuses on the need for QIBs in oncology applications and emphasizes the importance of the assessment of their reproducibility and repeatability.

Introduction

In the last decade, there have been major rapid advances in the field of magnetic resonance imaging (MRI), including advancements in hardware, acquisition pulse sequences, image reconstruction, and data analysis algorithms (1–9). These technological advances have fostered to a timely focus on quantitative MRI (qMRI), which purports the ability to derive objective metrics from images that relate to specific physical or biophysical properties of the imaged tissue. Two prime qMRI examples are diffusion-weighted imaging (DWI) and dynamic contrast enhanced (DCE)-MRI, which allow characterization of tissue cellularity inferred from water mobility and microvascular properties, derived from exogenous contrast agent (CA) kinetics, respectively. Applications of qMRI include detection of disease and its evolution in progression or response to therapies that affect the relevant biophysical property of tissue (e.g., cytotoxic therapies that reduce cellularity) (10–13). These methods have been covered in excellent reviews detailing the technical aspects and their applications (5–9,14–17). As the quantitative measurements derived from DWI (e.g., mean diffusivity) and DCE (e.g., volume transfer constant) are utilized in clinical trials of new treatment strategies, or for precision medicine and personalized cancer care, the technical confidence of these measurements in repeatability and reproducibility is ever more critical (18–20). Expert task forces of the Quantitative Imaging Biomarkers Alliance (QIBA®) have devoted significant resources to write DWI and DCE profiles and review over 1000 scientific manuscripts in this area, but such literature review efforts have yielded few original manuscripts with adequately described test-retest data. The lack of repeatability and reproducibility literature in this area creates a roadblock for clinical translation of DWI and DCE. This review focuses on the clinical and technical needs for quantitative DWI and DCE derived imaging biomarkers and provides recommendations for image acquisition, analysis, and quality control relevant to improving precision and accuracy, or reducing measurement error for the derived quantitative metrics. This review is limited to the use of DWI and DCE for evaluation of tumors in the brain, prostate, breast, liver, and head and neck, recognizing that studies of different organs imply varying technical protocols and challenges.

1. Understanding Impact of Precision and Accuracy in Quantitative Imaging

Quantitative imaging metrics reflect relevant information about a biological process by measuring biophysical parameters that could be used as biomarkers, rather than relying solely on relative differences in image signal intensity (of arbitrary scale and units), as in routine diagnostic imaging. However, the quantitative images must be standardized and optimized to generate protocols for acquisition and analysis of these biomarkers. Kessler, et al. have defined the term Quantitative Imaging Biomarker (QIB) as “An objective characteristic derived from an in vivo image measured on a ratio or interval scale as an indicator of normal biological processes, pathogenic processes, or a response to a therapeutic intervention” (21). Unlike conventional diagnostic imaging, where sensitivity and specificity are used to describe the predictive power of the qualitative test for the patient population, the technical performance of a QIB in quantitative imaging, particularly its bias (accuracy), precision (variability), and linearity, determine its inherent reliability (confidence interval) to diagnose, monitor, and predict outcome.

The objective of a test-retest study is to measure the degree to which test results are consistent over time (22–26). The internal consistency is a measure of the correlation between two sets of imaging data performed on two occasions. In a test-retest study, subjects are scanned at least twice over a short period of time to ensure that no biological change has occurred. From each image, the biomarker measurement is derived completely and independently of the results from the other scan. Sometimes a subject is scanned twice in the same study session; it is important in these instances that the subject leaves the table and is repositioned for the second scan. In other situations, particularly those involving CA administration, a subject might be scanned a second time the next day or so. For example, administering a gadolinium-based CA to a patient in a clinical trial twice on the same day is not practical due to retention of the CA in a lesion and the concern for patient safety. For this reason, it is not surprising that test-retest data are limited in sample size. Although test-retest studies are ideally performed on clinical subjects, estimates of test-retest variability can be obtained from phantom studies (27,28). It is generally recognized that these estimates are likely to underestimate the true variability seen in clinical subjects. The variability could be due to the patient movement that adds to variation in the signal intensity measurement compared with the phantom study.

2. Precision Metrology

2.1. Repeatability and Reproducibility

Repeatability represents the measurement precision, or closeness of agreement, of replicate measurements made over a short period of time. These replicate measurements are made with the same measurement procedure, operator, measuring system, operating condition, and physical location (21). Reproducibility is similar to repeatability, except that in acquiring the measurements, some aspect of the procedure, or timing, differs (e.g., different operator, different scanner, etc.) (21). For instance, systematic measurement bias between different scanners would be expected to impact reproducibility compared with more controlled (single-system) repeatability values.

Repeatability is often quantified by the within-subject standard deviation (wSD) or variance. For example, for N subjects, each with replicate measurements, one could use Table 1 to estimate the wSD. Large values of wSD indicate that confidence in any single measurement of the biomarker should be minimal because a second measurement is likely to differ considerably. Small values of the wSD boost confidence in the reliability of the measurement. Of course, “large” vs. “small” should be interpreted relative to known, or anticipated, biomarker differences between normal/abnormal tissues, across patient groups, or change with time in an individual patient/lesion undergoing treatment.

Table 1:

Steps for calculating the within-subject deviation

Steps	Method for calculating wSD
1	Calculate the variance for each of N subjects from their replicate measurements.
2	Take the mean of the variances over the N subjects. This gives an estimate of the within-subject variance.
3	Take the square root of the estimated within-subject variance to get an estimate of the wSD.

within-subject deviation (wSD)

When a patient is followed longitudinally to measure tumor treatment response or progression, there is a need to understand how to interpret observed differences in biomarker values. Small differences may be attributable to just measurement error whereas large differences exceed the expectation from simple measurement error. To reduce measurement error, standardized protocol should be implemented on pre-tested measurement device and kept consistent in the course of the study. If the wSD is known from relevant prior studies, a threshold can be calculated for when the difference between two longitudinal measurements is attributable to measurement error and when it can be confidently attributable to a true change. The “test-retest” procedure is designed to estimate wSD (22,23). From this, the repeatability coefficient (RC) is calculable (29) and represents the least significant difference between two repeated measurements taken under identical conditions, usually at a confidence level of 95%. It can be calculated as follows:

$$RC=1.96\times \sqrt{2wS{D}^2}=2.77\times wSD$$ [1]

For example, if a test-retest of DWI is performed and the wSD is estimated at 10, then RC = 27.7. This means that if the difference between a patient’s baseline and follow-up measurements is < −27.7, or > +27.7, then real change (outside of the measurement error) has occurred with 95% confidence. If the difference is between −27.7 and +27.7, it may be due to measurement imprecision.

When the repeatability varies with the magnitude of the QIB measurements, the within-subject coefficient of variation (wCV) is sometimes used to quantify the variability (29). The wCV is commonly used with imaging biomarkers because often the wSD is small for small QIB values but increases with larger QIB values. The steps for calculating the wCV are given in Table 2. Instead of the RC, the %RC is calculated as:

$$\%RC=2.77\times wCV$$ [II]

For example, if a test-retest of DCE is performed and the wCV is estimated at 10%, then the %RC = 27.7%. This means that if the percent change (e.g., difference between a patient’s baseline and follow-up measurements divided by the baseline measurement × 100) is < −27.7% or > 27.7%, then real change has occurred with 95% confidence. If the percent change is between −27.7% and +27.7%, it may be due to measurement imprecision.

Table 2:

Steps for calculating the within-subject coefficient of variation

Steps	Method for calculating wCV
1	Calculate the variance and mean for each of N subjects from their replicate measurements.
2	Calculate the wCV2 for each of the N subjects by dividing their variance by their squared mean.
3	Take the mean of the wCV2 over the N subjects.
4	Take the square root of the value in step 3 to get an estimate of the wCV.

within-subject coefficient of variation (wCV)

Based on the current literature, %RC values for tumor apparent diffusion coefficient (ADC) region of interest (ROI) measurements derived from monoexponential modeling of DWI data in three different organs are as follows: brain = 11% (30–32), liver = 26% (33–36), and prostate = 47% (37–40). This assumes the wCV for tumors in the brain is 3.97%, 9.38% for the liver, and 16.97% for the prostate. The claim statements for the tumor ROIs in these organs can be found in the DWI QIBAprofile(https://qibawiki.rsna.org/images/7/7d/QIBADWIProfilev1.45_20170427_v5_accepted_linenumbers.pdf).

The %RC values for volume transfer constant (K^trans) measurements in tumors, derived from pharmacokinetic modeling of DCE data, in two different organs is as follows: 21.3% for the brain and 55.7% for the prostate. This assumes that the wCV for tumors in the brain is 7.7% and 20.1% for the prostate (41,42). The claim statements for the tumor measurements in these organs can be found in the DCE QIBA profile (http://qibawiki.rsna.org/index.php/Profiles).

Figure 1 illustrates the impact of the wCV (or wSD) on this decision cut-point. It shows that when the imaging methods have good repeatability, earlier and more confident conclusions can be made about changes in patients’ QIB measurements. For example, when the wCV is low, e.g., 5%, a real change in K^trans of 14% (or larger) can be detected with 95% confidence. If the wCV is moderate, e.g., 15%, a change of over 40% is needed to rule out the measurement error. A doubling of Ktrans, equivalent to 100% change, would have to occur to detect a real change with 95% confidence when the wCV is 36%.

Figure 1:

The %RC is the cut-point where a change in the biomarker measurements is considered a real change, not merely a measurement error, with 95% confidence. The graph illustrates how this cut-point increases with the within-subject CV (wCV). When the wCV is small (i.e., high precision), very small changes in the biomarker can be detected. Whereas when the wCV is large (i.e., low precision), large changes in the biomarker are needed before one can be confident that a real change has occurred.

It is likewise important to know a QIB’s reproducibility when measuring the change in a patient from baseline. Often, slightly different imaging methods are used at baseline and follow-up, e.g., a different scanner, a different radiologist, or even a different facility. If the reproducibility of a QIB is known, then the minimum detectable difference can be calculated. The minimum detectable difference when different imaging methods are used is called the reproducibility coefficient and it is calculated similarly to the RC as described above (43). However, the reproducibility coefficient is often significantly larger than the RC because of the additional variance associated with the different imaging methods and their systematic biases. It is critical to recognize these sources of additional error and their effect on interpreting patients’ quantitative images.

The sample sizes of test-retest studies in the literature vary considerably, from a couple of subjects to the study by Petersen, et al. where 28 sites in Asia, Europe, and North America participated, and a total of 284 healthy volunteers were scanned (26). Usually, studies have included results from fewer subjects than that mentioned above (22–25). Obuchowski and Bullen have performed a simulation study to determine the minimum sample size needed in test-retest studies (44). They have determined that estimates of precision should be based on a sample size of at least N = 35 to provide true 95% confidence intervals for a patient’s QIB measurement and for change in the QIB over time. Note that the estimate of precision could come from a single test-retest study with N ≥ 35, or calculated as a summary measure from a meta­analysis of multiple test-retest studies (45) where the combined sample size is N ≥ 35.

2.2. Phantom-based Methodology to Improve qMRI Precision

2.2.1. DWI

Diffusion MRI assesses the Brownian motion of water molecules non-invasively. On the timescale relevant to clinical DWI, water mobility is obstructed by tissue microstructure, including extracellular tortuosity, cell membranes, organelles, and macromolecular components (46). This sensitivity to microstructure enables DWI to elucidate impediments on a micrometer-size scale to the usual random Brownian motion. The ADC value is the key quantification parameter of DWI images typically of interest for clinical decision making (3,26,30) and is derived from mono-exponential modeling of the signal intensity as a function of b-value minimizing effects of perfusion and restricted diffusion (46). The diffusion coefficient has temperature dependence in pure water of 2.4% per degree Celsius (47,48). This dependency is rarely of concern in vivo because body temperature is well regulated, and other biophysical properties have much greater influence on tissue water mobility. For instance, dense tumors typically exhibit lower values of ADC than benign tissues because of higher cell packing.

The primary use of a phantom is to standardize DWI acquisition schemes across multiple vendors, software, and hardware platforms, and certify proper calibration and performance of the systems to ensure adequate ADC measurement, accuracy, and reproducibility. The indispensable value of the phantom is in providing ground-truth parameter values fundamentally independent of measurement method both for acquisition and image analysis. Therefore, phantom measurements can be used to improve the quality of DWI images by minimizing artifacts and geometrical distortions, and ensuring a high degree of reproducibility across different sites and scanner platforms. For reproducibility evaluation, it is important that the field of view (FOV), b-values, imaging matrix, repetition time (TR), echo time (TE), parallel imaging factor, number of slices, slice positions, and slice thickness are held constant and match clinical protocols.

Many materials have been used in DWI phantoms such as aqueous solutions of polydimethylsiloxane, polyvinylpyrrolidone, sucrose, or polymers, liquid paraffin, alkanes, and pure water (28,49–51). The aqueous solutions and pure water are good choices for a phantom because of their non-toxicity and availability. Temperature dependence of ADC measures can be mitigated using an ice-water bath to ensure 0°C measurement across scanners (51). Moreover, the diffusion coefficient of water at 0oC is approximately 1.1 × 10–3 mm²/s, which is well within the tissue ADC range (39,52–56). However, the longitudinal relaxation time (T1) and transverse relaxation time (T2) for ice-water are much longer than most tissues. One recommended simple phantom design has been described by Chenevert, et al. (51) and consists of two cylinders of polypropylene, with the larger one containing ice and water. A smaller tube is filled with distilled water in thermal equilibrium at 0°C. Theoretically, the ADC value in this phantom should be independent of the acquisition protocol used at each site. This non-toxic and stable ice-water phantom can be readily manufactured on-site and has already been utilized for quantitative DWI quality assurance (QA) by several multi-site clinical trials (ACRIN 6698, 6701 and 6702, (57,58). Figure 2 shows representative intra-site ice-water phantom repeatability measurements acquired four times to calculate water ADC at 0° C in a multi-site setting using the same MRI imaging protocol. The CV for these 3 sites was 0.6%, 0.1%, and 1.1%. The biggest differences found between different MRI imaging protocols are in: FOV, number of slices, TR, and TE. Therefore, in practice, variability between DWI protocols influences the measured ADC value, and the differences found across measurements are about 10% from the literature value (51,59,60). The most significant differences were observed between MRI system manufactures due to distinct gradient designs leading to spatially-dependent bias in diffusion weighting b-values (28,58).

Figure 2:

Box-and-whisker plot demonstrating apparent diffusion coefficient (ADC) repeatability of water for multi-site results at 3T magnetic resonance imaging scanners using the ice water phantom. Note: Inset is the ADC map of the phantom. (Images compliments of authors from sites 1, 2, and 3: Memorial Sloan Kettering Cancer Center, Columbia University Medical Center, and University of Michigan.

The use of an application-specific phantom, such as that one developed by the National Institute of Standards and Technology (NIST) and (QIBA, to evaluate ADC measurement linearity for multi-b-value DWI studies using an array of ADC values is also recommended (61,62). The phantom is constructed using varying concentrations of polyvinylpyrrolidone (PVP, 0, 10, 20, 30, 40, and 50%) in an aqueous solution to generate physiologically relevant ADC values (63) and is available from High Precision Device, Inc (Boulder, CO). The space between the vials within the phantom can be filled with ice-water bath for temperature control. Figure 3 shows representative multi-site DWI data for 3T MRI scanners with repeatability measures. The recommended QIBA protocol for repeatability assessment with PVP phantom uses b-values of 0, 500, 900, and 2000 s/mm² and is repeated four times, based on the guideline. QIBA provides scan protocols and software for standard analysis of quantitative DWI phantom data on the Quantitative Imaging Data Warehouse (QIBA QIDW, rsna.org/qidw). ).

Figure 3:

Repeatability results obtained using the National Institute of Standards andTechnology/Radiological Society of North America Quantitative Imaging Biomarkers Alliance diffusion-weighted magnetic resonance imaging phantom containing vials with varying concentrations of polyvinylpyrrolidone (0–50%) to generate physiologically relevant apparent diffusion coefficient (ADC) values at different vial positions (c = central; o = outer; I = inner). The phantom and ADC image are shown as inserts in the graph. Graph showing ADC (mean±sd) values for each vial in 4 experiments performed at (A) site 1, (B) site 2, and (C) at site 3. (Images compliments of author from sites 1, 2, and 3: Memorial Sloan Kettering Cancer Center; Columbia University Medical Center, and University of Michigan.

2.2.2. DCE

DCE is a noninvasive technique that measures microvascular permeability and volume fractions of the extravascular extracellular space (EES) and blood plasma space. One of the technique, involves serial acquisition of T1-weighted images before, during, and after intravenous injection of gadolinium-labeled CAs. This review will focus on the DCE T1 measurement techniques. These CAs are low-molecular weight paramagnetic complexes that extravasate to EES through vascular space. DCE analysis allows measurement of signal change on a ROI or voxel. DCE time-course data have been analyzed using heuristic approaches and quantitative kinetic analysis based on the tissue compartmental models. The heuristic, semi-quantitative measurement of the blood-normalized initial-area-under-the-gadolinium curve (IAUGC_BN) has been described in the QIBA DCE profile (http://qibawiki.rsna.org/index.php/Profiles) (16,64). Areas showing rapid, high concentration uptake and fast wash-out of the CA are generally correlated with regions of malignancy within suspicious neoplastic lesions (65–67). The kinetics of the DCE time-course data depend on unique tumor vascular characteristics and thus the derived imaging metrics have found relevance in oncological applications (4,10,42,68).

With proper pharmacokinetic (PK) modeling of DCE time-course data, QIBs can be estimated. The most commonly used QIB for characterizing tumor vascular properties is the CA volume transfer constant (K^trans), which has been detailed in the DCE profile (http://qibawiki.rsna.org/index.php/Profiles). The most common PK models used for DCE data analysis are the standard Tofts model (TM), which estimates K^trans and volume fraction of EES (V_e) and the extended Tofts model (ETM) (14), which provides estimate of K^trans and V_e. and an additional metric, V_p. To estimate these QIBs, such models require additional information such as input of tissue native T1 values and arterial input function (AIF).

A DCE experiment is basically a measurement of T1 changes in a tissue during the passage of CA. A static phantom has been proposed in initial work by ACRIN CQIE (69) and then by QIBA to standardize DCE acquisition protocols across hardware platforms from various vendors to verify calibration and performance of the systems. Phantom quality control may ensure adequate precision and reproducibility in T1 measurements, a prerequisite for performing quantitative DCE analysis. It also allows for the measurement of the contrast response of the employed DCE acquisition sequence across a range of T1 relaxation rates observed in in vivo DCE studies and the stability of that contrast response over time and across system upgrades (70,71).

DCE acquisition parameters can vary significantly across vendors, scanner models, station software packages, and field strengths. Often, each vendor and platform has preferred acquisition protocols. The T1 values in clinics are influenced by B1 field inhomogeneity, incomplete spoiling of transverse magnetization and MR sequence used for the range of T1 values to be measured (70). One of the QIBA recommendations is to standardize acquisition parameters to reduce sources of variability for DCE imaging, possibly at the expense of moderate protocol capabilities for some systems. Before acquiring data from subjects, it is essential that the selected pre-DCE T1 mapping protocol be performed on the static standardized phantom multiple times (or on different days) with the phantom repositioned for each experiment. This provides an estimate of true scanner variance and bias for T1 values.

One recommended QIBA DCE T1 phantom contains vials of varying concentrations of nickel chloride solutions (70). Figure 4 shows the phantom design that consists of two sets of spherical inserts. The spheres were doped with nickel chloride to achieve T1 values spanning the range expected in vascular and tissue compartments during a DCE study. For the vascular input function spheres, the T1 values range between 0.75–41.6s⁻¹, and for the tissue spheres, the range is 0.67–7.5s⁻¹. To mimic the coil loading of a patient, the phantom was filled with a 30-mM sodium chloride (Sigma-Aldrich, USA) solution. The scan protocol for T1 measurement consists of acquiring coronal fast spoiled gradient echo sequences with variable flip angles (VFAs) of 30°, 25°, 20°, 15° 10°, 5°, and 2° to fully cover the range used for T1 mapping in clinical studies (70). Test-retest reliability and T1 accuracy evaluation using the QIBA DCE phantom should be considered for longitudinal studies. In addition to the DCE phantom, QIBA also provides an automated T1 quantification software application, DCETool, to analyze the data acquired from the QIBA DCE phantom (QIBA DCE-MRI WG at rsna.org/qidw ). This phantom and analysis software has been used for site qualification and requalification in support of ACRIN 6701, a DCE and DWI test-retest clinical trial in prostate cancer patients (unpublished data). Spatially-dependent B1 field inhomogeneity effects, which are more significant at higher field strengths, such as 3T, may confound VFA T1 data when acquired over large anatomic regions, necessitating B1 mapping and corrections to be included in the T1 measurement protocols. These effects will be addressed by version 2.0 of the RSNA QIBA DCE-MRI Profile that is currently under development.

Figure 4:

(A) The Quantitative Imaging Biomarkers Alliance dynamic contrast-enhanced phantom layout with 32 spheres, with different concentrations of NiCl2 solutions for varying T1 relaxation rates (R1). (B) T1-weighted magnetic resonance image of the phantom showing the 32 spheres, and (C) R1 values of the 8-vascular input function mimicking inserts compared with National Institute of Standards and Technology theoretical R1 values. (D) R1 values for the 24 tissue-mimicking inserts. (Images compliments of Edward Jackson, University of Wisconsin-Madison).

In addition to static T1 phantoms, some investigators have developed dynamic phantoms for MRI (Kim, et al. (71)), and CT (Driscoll, et al. (72)). The recent perfusion phantom was developed to correct MR scanner dependent variations in estimates of the tissue perfusion parameters in abdomen (71). The design of Kim, et al. is shown in Fig. 5A (71) is small enough to be imaged together with a patient for real-time quality assurance. Repeatability of the contrast enhancement curve of this phantom was measured using three phantoms placed at the isocenter of a 3T scanner, and the intra-class correlation coefficient was higher than 0.99 (Fig. 5B). Kim, et al. have demonstrated that this phantom significantly reduced the variation in quantifying perfusion parameters of various abdominal tissues across two different 3T scanners (71). However, the stability of this and other dynamic phantoms will need to be validated in longitudinal multi-site trials because such dynamic phantoms are difficult to produce in a manner that provides consistent results across phantoms and time.

Figure 5:

Portable perfusion phantom and its repeatability measurement. (A) Photograph of a portable perfusion phantom, and (B) contrast enhancement curves of three phantoms placed in a 3T magnetic resonance imaging scanner (temporal resolution = 2.9 s). Repeatability determined by the intraclass correlation coefficient is larger than 0.99. (Images compliments of Harrison Kim, University of Alabama at Birmingham).

2.3. Clinical data-driven Approaches to Improve qMRI Precision

2.3.1. DWI

The QIBA/Diffusion Biomarker Committee task force is dedicated to developing a DWI profile. The task force members reviewed over 1000 research articles to develop the profile claim statement based on clinical data from several organ systems, including the brain, liver, and prostate. Table 3 summarizes key scan protocol parameters for the brain (3A), prostate (3B), and liver (3D) from the QIBA profile (https://qibawiki.rsna.org/index.php/Profiles) to be relevant to a wider scientific audience. The tables have been adapted from the profile with permission from QIBA. Organs, such as breast (Table 3C) (8,73) and head and neck (Table 3E) (55,68,74), were not covered in the profile due to a lack of sufficient test-retest data. These review results point out the limited test-retest literature in various organs in the clinical oncology setting.

Table 3:

Typical Diffusion-Weighted Imaging (DWI) and Dynamic Contrast Enhanced (DCE) Acquisition Details for (A) Brain, (B) Prostate, (C) Breast, (D) Liver, and (E) Head and Neck

(A) BRAIN
Parameters	DWI		DCE
Field Strength	1.5 T/3T		1.5T/3T
Acquisition Sequence	SS-EPI		3D SPGR
Receive Coil type	≥8channel head array coil		≥8channel head array coil
Lipid Suppression	On		On
Slice thickness (mm)	4–5		≤5
Gap thickness (mm)	0–2		0–1
FOV(mm)	220 −240		220 −240
Acquisition Matrix	160–256 × 160–256 or 1.5–1 mm in plane resolution		256×128–160
Plane Orientation	Axial		Axial
Phase/frequency encode direction	AP/RL		AP/RL
Receiver bandwidth (Hz/pixel)	Max possible in freq encoding direction (acceptable > 1000)		250
	DWI Sequence Class Monopolar Bipolar Bipolar Double Spin Echo		Contrast*	Pre-	Post-
	# b-values	≥ 2 (including b=0 s/mm2)	# Phases	≥5	40–80
Specific Parameters	Minimum highest b-value (s/mm2)	850–1000	# Averages	≥1	1
	# Averages	≥2	Flip Angles (FAs) (deg)	2–30*	25–30
	Diffusion Directions	3 orthogonal	# Flip Angles (FAs)	2–7	1
	TR (ms)	3000–5000	TR (ms)	3–8¥	3–8
	TE (ms)	Minimum	TE (ms)	≤3¥	≤3
	In plane parallel imaging	2	Temporal Resolution (sec) /Total Acquisition Time (min)	<10 (ideal 5)/5–10
	Total Acquisition Time (min)	3	#Contrast Dose and IV injection rate see references *Variable FAs for T10 measurement ¥Ensure TR/TE stays constant for all flip angles

(B) PROSTATE
Parameters	DWT		DCE
Field Strength	3T		1.5T/3T
Acquisition Sequence	SS-EPI		3D SPGR
Receive Coil type	> 8 channel torso array coil; pelvic phase array/endorectal coils; body array coil)		> 8 channel torso array coil; pelvic phase array coil/endorectal coils; body array coil)
Lipid Suppression	On		NA
Slice thickness (mm)	3–5		<5
Gap thickness (mm)	0–1		0–1
FOV(mm)	240–260		260–300
Acquisition Matrix	224–128 × 224–128; 1–2 mm in-plane resolution		<256×l60
Plane Orientation	Axial		Axial
Phase/frequency encode direction	AP/RL		RL/AP
Receiver bandwidth (Hz/pixel)	Max possible in freq encoding direction (acceptable > 1000)		250
	DWI Sequence Class Monopolar Bipolar Bipolar Double Spin Echo		Contrast#	Pre-	Post-
	# b-values	2 (including b < 50–100 s/mm2)	# Phases	≥5	Sufficient to allow acquisition of at least 5 min post injection; ≥ 30
Specific Parameters	Minimum highest b-value (s/mm2)	500–1500	# Averages	≥1	1
	# Averages	2-≥4	Flip Angles (FAs) (deg)	2–15*	10–15
	Diffusion Directions	≥ 3 orthogonal	#Flip Angles (FAs)	3–5	1
	TR (ms)	≤4000	TR (ms)	<5¥	<5
	TE (ms)	Minimum	TE (ms)	≤2¥	≤2
	In plane parallel imaging	2	Temporal Resolution (sec) /Total Acquisition Time (min)	<10/5–10
	Total Acquisition Time (min)	3	#Contrast Dose and IV injection rate see references * Variable FAs for T10 measurement ¥Ensure TRITE stays constant for all flip angles

(C) BREAST
Parameters	DWI		DCE
Field Strength	1.5T/3T		1.5T/3T
Acquisition Sequence	SS-EPI/SE-EPI		3D SPGR
Receive Coil type	≥ 4 channel breast phase array coil		> 4 channel breast phase array coil
Lipid Suppression	On		On
Slice thickness (mm)	4–5		<2.5
Gap thickness (mm)	0–1		0
FOV(mm)	260–360		To cover the entire breast whether uni- or bilateral
Acquisition Matrix	128–192 × 128–192		> 192×256; 1–1.5 mm in-plane resolution
Plane Orientation	Axial		Sagittal for single breast coverage; axial for bi lateral coverage
Phase/frequency encode direction	RL/AP		RL/AP for axial bilateral; HF/AP for sagittal unilateral
Receiver bandwidth (Hz/pixel)	Max possible in freq encoding direction		250
	DWI Sequence Class Monopolar Bipolar Bipolar Double Spin Echo		Contrast#	Pre-	Post-
	# b-values	≥2	# Phases	≥2	Sufficient to allow acquisition for ≥ 8 min post injection
Specific Parameters	Minimum highest b-value (s/mm2)	800	# Averages	≥1	1
	# Averages	≥2	Flip Angle (FA) (deg)	2–30*	10–30
	Diffusion Directions	3 orthogonal	# Flip Angles (FAs)	3–5	1
	TR (ms)	≥4000	TR (ms)	<8¥	<8
	TE (ms)	Minimum	TE (ms)	≤3¥	≤3
	In plane parallel imaging	2	Temporal Resolution (sec) /Total Acquisition Time (min)	≤ 20 /≥8
	Total Acquisition Time (min)	4–6	* Contrast Dose and IV injection rate see references * Variable FAs for T10 measurement ¥ Ensure TR/TE stays constant for all flip angles

(D) LIVER
Parameters	DWI		DCE
Field Strength	1.5T/3T		1.5T/3T
Acquisition Sequence	SS-EPI		3D SPGR
Receive Coil type	> 6–16 channel torso array coil		> 8–32 channel flexible or AP body array coil
Lipid Suppression	On		On
Slice thickness (mm)	5–7		4–5
Gap thickness (mm)	0–1		0
FOV(mm)	300–450		280–380
Acquisition Matrix	160–196 × 160–192 or 2 −3 in-plane resolution		320 × (160–192)
Plane Orientation	Axial		Oblique
Phase/frequency encode direction	AP/RL		AP/RL
Receiver bandwidth (Hz/pixel)	Max possible in freq encoding direction (acceptable > 1000)		250
	DWI Sequence Class Monopolar Bipolar Bipolar Double Spin Echo		Contrast#	Pre-	Post-
	# b-values	≥2 (including one b = 50–100 s/mm2)	# Phases	≥5	100
Specific Parameters	Minimum highest b-value (s/mm2)	600–800	# Averages	≥2	1
	# Averages	2-≥4	Flip Angles (FAs) (deg)	2–30*	20–30
	Diffusion Directions	3 orthogonal	# Flip Angles (FAs)	3–5	1
	TR (ms)	>2000	TR (ms)	3–7¥	3–7
	TE (ms)	Minimum	TE (ms)	≤5¥	≤5
	In plane parallel imaging	2–3	Temporal Resolution (sec) /Total Acquisition Time (min)	< 5(ideal~3)/5
	Total Acquisition Time (min)	~5	# Contrast Dose and IV injection rate see reference * Variable FAs for T10 measurement ¥Ensure TR/TE stays constant for all flip angles

(E) HEAD AND NECK
Parameters	DWI		DCE
Field Strength	1.5T/3T		1.5 T/3T
Acquisition Sequence	SS-EPI		3D SPGR
Receive Coil type	Neck array or neurovascular coil		Neck array or neurovascular coil
Lipid Suppression	On		On
Slice thickness (mm)	≥5		>5
Gap thickness (mm)	0		0
FOV(mm)	220–380		180–220
Acquisition Matrix	128×128		256×128
Plane Orientation	Axial		Axial
Phase/frequency encode direction	AP/RL		RL/AP
Receiver bandwidth (Hz/pixel)	Max possible in freq encoding direction		250
	DWI Sequence Class Monopolar Bipolar Bipolar Double Spin Echo		Contrast#	Pre-	Post-
	# b-values	≥3	# Phases	≥5	40–80
Specific Parameters	Minimum highest b-value (s/mm2)	1000	# Averages	≥ 1	1
	# Averages	>2	Flip Angles (FAs) (deg)	5–30*	15–30
	Diffusion Directions	3 orthogonal	# Flip Angles (FAs)	≥3	1
	TR (ms)	≥2000	TR (ms)	<9¥	<9
	TE (ms)	Minimum	TE (ms)	<2¥	<2
	In plane parallel imaging	2	Temporal Resolution (sec) /Total Acquisition Time (min)	≤6/~5
	Total Acquisition Time (min)	3	#Contrast Dose and IV injection rate see references *Variable FAs for T10 measurement ¥Ensure TR/TE stays constant for all flip angles

Clinical DWI is typically acquired using a diffusion-weighted, single-shot echo planar imaging (SS-EPI) sequence. The acquisition parameters are detailed in Table 3 for the five specific organs outlined in this review (3,8,10,74–85). Protocol optimization is a prerequisite for obtaining optimum signal-to-noise ratio for the DWI images because artifacts can be significant. Techniques that reduce the number of phase encoding steps and FOV, resulting in reduced artifacts, are preferred (86). There are newer developments in DWI building on single-shot EPI such as reduced FOV acquisition or multi-shot EPI (87,88). To reduce susceptibility artifacts and to improve spatial resolution compared to SS-EPI methods, the propeller/blade diffusion methods have also been used (89,90). Selection of optimum b-values for a specified organ is also an important parameter that should be optimized for signal-to-noise ratio (38,39). Another point in technique optimization is “landmark on the organ of interest” to confirm that organ position is close to the isocenter to minimize b-value non-uniformity across the organ (91).

Prior to the analysis, a radiologist draws a ROI on the DWI images guided by ancillary MR images, radiologic and clinical information. The ROI encompasses the entire tumor or tissue of interest. A DWI protocol includes producing an ADC map based upon a mono-exponential fit to images obtained using two or more b-values. Generally, at least three orthogonal diffusion directions are probed, with the resultant maps generated from combinations of the directional data, assuming isotropic diffusion (3,53).

$$S/S_0=\text{exp}\left(-b\times ADC\right),$$ [III]

where S and S₀ are the signal intensities with and without diffusion weighting, respectively, and b is the diffusion weighting factor (b-value, s/mm²).

Most MRI scanners have capabilities for automatically producing ADC maps from the DWI images using propriety software based on mono-exponential modeling of the data. Figures 6B, 7B, 8C, 9D, and 10B show representative ADC maps derived from patients with tumors in the brain, prostate, breast, liver, and head and neck, respectively.

Figure 6:

Representative pre-treatment magnetic resonance images of a patient with grade 4 brain tumor (65 years, female). (A) T2-weighted image. (B) Apparent diffusion coefficient × 10⁻³ (mm²/s) map generated using 3 b-values (b = 0, 100, 1000 s/mm²). (C) K^trans (min⁻¹) map generated from dynamic contrast enhanced data with insert of T1-weighted gadolinium contrast image. (Images compliments of Thomas Chenevert, University of Michigan).

Figure 7:

Representative pretreatment magnetic resonance images of a patient with prostate cancer—Gleason Score 4+3 (66 years, male). (A) T2-weighted image, (B) apparent diffusion coefficient × 10⁻³ (mm²/s) map generated using 2 b-values (i.e., b = 0, 600 s/mm²), and (C) K^trans (min⁻¹) map generated from dynamic contrast-enhanced data. (Images compliments of Susan M. Noworolski, University of California San Francisco).

Figure 8:

Representative magnetic resonance images from a breast cancer patient (34 years, female) with grade II invasive ductal carcinoma (IDC) in the right breast. (A) T2-weighted image with fat saturation, (B) color K^trans (min⁻¹) map of the tumor overlaid on T1-weighted dynamic contrast-enhanced image with fat saturation, and (C) Representative apparent diffusion coefficient × 10⁻³ (mm²/s) (ADC) map from a breast cancer patient (37 years, female) with grade II IDC in the right breast. Composite ADC map was generated from diffusion-weighted imaging images with b = 0 and 800 s/mm² showing decreased ADC in tumor (arrow). (Images compliments of Wei Huang, Oregon Health & Science University).

Figure 9:

Representative magnetic resonance images from a recurrent hepatocellular carcinoma patient (57 years, male) acquired on a 3T scanner. Dynamic contrast-enhanced magnetic resonance imaging scan showing (A) enhancing tumor and (B) contrast enhancement time course. (C) The gadolinium concentration time course and extended Tofts model fit and (D) Composite apparent diffusion coefficient map generated from diffusion-weighted imaging images, with b = 0, 600 s/mm² from same patient. (Images compliments of Sachin Jambawalikar, Columbia University Medical Center)

Figure 10:

Representative pretreatment magnetic resonance images of head and neck cancer patient (52 years, male). (A) T2-weighted image, (B) apparent diffusion coefficient ×10⁻³ (mm²/s) map overlaid on diffusion-weighted (b = 0 s/mm²) images generated using 10 b-values (0, 20, 50, 80, 200, 300, 500, 800, 1500, and 2000 s/mm²), (C) K^trans (min⁻¹) map overlaid on pre-contrast T1-weighted image. (Images compliments of Amita Shukla-Dave, Memorial Sloan Kettering Cancer Center).

As discussed above, the QIBA/DWI Biomarker Task Force members performed an extensive literature search and found limited manuscripts with test-retest data and therefore reported the %RC for ADC in tumor ROIs derived from mono-exponential modeling of DWI data only in three different organs as follows: brain = 11% (30–32) liver = 26% (34–36,52) and prostate = 47% (37,40). The details on literature and assumptions used to inform these 95% confidence interval (CI) values are adapted from QIBA/DWI profile with permission. Specifically, meta-analysis was performed on the available test-retest study reports (e.g., 2–3 per organ) acquired with qualitatively similar acquisition protocols (detailed in the QIBA/DWI profile) to pool maximum sets of subjects (>30) sufficient to satisfy statistical significance. The estimated wCV for mean ADC in an ROI between 1–4 cm² was 3.97% for brain (30–32), 9.38% for liver (33–35), and 16.97% for prostate (37–40). The derived CIs could likely be improved by more advanced organ-specific acquisition protocols to achieve better QIB precision.

2.3.2. DCE

Similar to Diffusion, the QIBA/Perfusion Biomarker Committee Task Force has invested significant effort in updating the original DCE profile (https://qibawiki.rsna.org/index.php/Profiles). The new version of the DCE profile, version 2.0 (under development), includes the brain (Table 3A) (75,81,92), prostate (Table 3B) (38,39,76,78), and breast (Table 3C) (8,79,93). The tables have been adapted from the working document of the profile with permission from QIBA. Specific scan parameters for head and neck (10,74,94) and liver (80,95) were not included in the profile due to limited test-retest clinical data but are reflected in Table 3D and Table 3E, respectively. It is important to emphasize that even though the DCE literature consists of many studies of tumors in various organs, important repeatability and reproducibility data are lacking. The major challenge for such DCE test-retest studies is the need to repeat the CA injection. The retest for DCE should be performed after the first CA injection has been eliminated from the patient, which typically requires about 24 hours. The half-life of common low molecular-weight gadolinium CAs is approximately 90 min, retention in some tissues and lesions can be significantly longer. This creates issues with the logistics of repeating the experiments. Additional CA injections also require institutional review board approval, which is especially pertinent given the increasing awareness of nephrogenic systemic fibrosis in patients with abnormal renal function and potential brain deposition of gadolinium in patients with normal renal function (96).

DCE is typically acquired using a T1-weighted, fast spoiled gradient recalled echo sequence, and the temporal resolution is determined by the pulse sequence acquisition parameters and the spatial resolution and anatomic coverage required for the organ under study. The rate at which the CA extravasates from the vasculature depends on the molecular size of the CA. With low molecular weight agents, the temporal resolution required to observe microvessel permeability is typically on the order of 5–20s (14,18,97).

Similar to DWI, the ROIs for data analysis are often determined by experienced radiologists. DCE data are sometimes analyzed using qualitative or semi-quantitative methods. DCE profile details the heuristic approach using the IAUGCb n parameter (http://qibawiki.rsna.org/index.php/Profiles). Accurate AIF and native tissue T1 (i.e., T10) measurements are the first necessary steps for PK modeling of DCE data with TM and ETM (14). There are several ways to determine AIF (98–101), each with its pros and cons. The AIF should be measured from the feeding vessel of the tumor. In practicality, the AIF is often measured in a nearby major blood vessel within the vicinity of the tumor. For instance, the carotid arteries are commonly used for head and neck cancer and the aorta for liver cancer. However, due to the image volume coverage, angulation, or the absence of major vessels in the scan volume, direct measurement of AIF from the acquired images may not always be possible.

The time course of the CA concentration in the tissue, C_t(t), for the TM is based on the Kety exchange equations (14):

$$C_t\left(t\right)=K^{trans}{\int }_0^t{e}^{-k_{ep}\left(t-\tau \right)}{C}_p\left(\tau \right)d\tau$$ [IV]

where C_p(t) is the time course of the plasma concentration of the CA, K^trans (min⁻¹) is the volume transfer constant (vascular space to the EES), and k_ep = K^trans/V_e (min⁻¹) is the rate constant for CA transport from the EES to vascular space.

The ETM incorporates the vascular compartment in modeling the tissue CA concentration (14). For the ETM, C_t(t), can be expressed as:

$$C_t\left(t\right)=K^{trans}{\int }_0^t{e}^{-k_{ep}\left(t-\tau \right)}{C}_p\left(\tau \right)d\tau +v_p{C}_p\left(t\right)$$ [V]

Figures 6C, 7C, 8B, and 10C show representative K^trans maps derived from patients with tumors in the brain, prostate, breast, and head and neck, respectively. Figures 9A–C show CA uptake and CA time course of concentration curves for the aorta and tumors in the liver. The currently available test-retest DCE data has illustrated that the %RC for K^trans in a tumor ROI is 21.3% for brain and 55.7% for prostate (41,42). The statistical approach used to derive this performance claim information for DCE profile is similar to the one applied in the QIBA/DWI profile.

The DCE data acquisition in clinics for organs such as breast, prostate, liver is slightly challenging when compared to brain and head and neck. In this review, we highlight a few key acquisition aspects for these organs. The typical acquisition parameter range for breast DCE is detailed in Table 3C. There is a unique difference between DCE of the breast and that of other organs, largely due to the clinical need for bilateral scanning, which require full breast coverage with high spatial resolution because breast cancer has a high incidence of contralateral and multi-focal disease (102), and clear assessment of lesion morphology is essential for cancer diagnosis (103,104). For example, the American College of Radiology breast MRI lexicon recommends that image slice thickness no more than 2 mm or thinner. There is a trade-off between spatial and temporal resolutions, when using conventional gradient-echo pulse sequences, low temporal resolution breast DCE protocols on commercial scanner systems are commonly used in clinical settings and large-scale clinical trials such as the ISPY-1 (105,106) and ISPY-2 (107) trials. Due to poor accuracy in quantitative PK analysis of DCE data (67,108,109), semi-quantitative analyses (i.e., uptake slope, percent signal change, time to peak, signal enhance ratio, etc.) of DCE time course data are generally employed under this circumstance (110). The results of a simulation study by Henderson, et al. have shown that a temporal resolution of 16 s or less is preferred for PK analysis of breast DCE data (111). Parallel imaging acceleration together with k-space under-sampling acquisition and view sharing commercially available reconstruction methods, such as TWIST (time-resolved angiography with stochastic trajectories) (112–114), DISCO (Differential Subsampling With Cartesian Ordering) (115,116), and 4D THRIVE (T1 High Resolution Isotropic Volume Examination) sequences (117) have allowed for simultaneous acquisitions of 3D breast DCE data. For breast DCE, the AIF can be determined using direct measurement from an axillary artery (118), the reference-tissue method (119,120), or the population-averaged AIF (113,114,121).

For prostate DCE, data acquisition details are given in Table 3B (122), which suggests at least a 10-stemporal resolution and 30 time points, resulting in a five-minute total scan time. Spatial resolution is approximately 1 mm in-plane with 3-mm thick sections. These values are based on the PI-RADS_v2 recommendations for clinical acquisition (123). Studies from the literature generally meet these criteria with more variation in spatial resolution (0.7 −1.9 mm in-plane resolution and 3–4 mm slices) and, for some studies, higher temporal resolution (3–10 s) (122,124). Best spatial or temporal resolutions and/or increased coverage can be obtained via the use of key-hole imaging, parallel imaging, or compressed sensing (122,125). Usually, the AIF is directly measured from iliac arteries.

For liver DCE, data acquisition details are provided in Table 3D (126–132). One of the major challenges in acquisition of liver DCE data is the respiratory motion of the abdomen. Keeping this in mind, DCE images are mostly acquired with a series of multiple breath-holds and/or shallow breathing. In order to achieve the optimum tradeoff between temporal and spatial resolution, the procedure commonly used in the clinical setting is to coach the patients to hold their breath for approximately 15 s (expiratory phase), followed by a 5s break and then repeat the acquisition multiple times for a total of 2­5 min (129,132). This allows to acquire high temporal resolution DCE data <5s (ideal ~3s) at the first breath hold for the accurate quantification of the AIF. Generally, 10 −12 coronal slices are acquired to bisect both the lesion and aorta ensuring that the lesion is in the center of FOV in the superior-inferior direction. In addition, on a 3T scanner, a B1-mapping sequence for correction of the T10 mapping and DCE data is recommended (126,130,131). Even though liver has dual blood supply inputs (portal vein and aorta), and tumors are highly vascular, quantitative PK analysis software applications generally use a single input TM. However, recent work for liver DCE analysis has evaluated hepatic perfusion quantification using a dual-input kinetic model (127,128).

3. Pearls and Pitfalls as qMRI Precision is Improved

The role of qMRI in clinical oncology settings has been elegantly reviewed in the past (1,10,133–138). This review is focused on the need for more test-retest studies, defining repeatability and reproducibility, and determining the extent of repeatability and reproducibility determinations that have been performed in phantoms and patients and reported in the literature with adequate technical data and details to allow a statistically robust meta-analysis. Although there are about 1000 publications for human subject DWI and DCE studies in oncology in the literature, the test-retest pool of papers is still quite limited. In the development of a QIB that can be used in clinical trials or practice, a critical step is understanding the test-retest precision for a specific acquisition and analysis protocol; hence this dearth of test-retest data is limiting both imaging biomarker discovery and clinical application of more advanced quantitative imaging methods.

4. Proposal fo r qMRI Precision Studies

QIBA (http://www.rsna.org/qiba/) seeks to improve the value and practicality of QIBs by reducing variability across devices, patients, sites, and time (43). The common platform for communicating strategies to improve technical performance for QIB applications is through QIBA Profiles. Profiles are developed using published data to generate evidence-based performance claims that inform users about what quantitative results can be achieved by following the profile guidelines (139). For example, in longitudinal claims, the Profile provides a cut-point for when a true change has occurred as well as a range of values for the true change in the biomarker (139). For each of these claims, data on the imaging procedure’s precision is critical to obtain, particularly its ability to provide repeatable measurements when there has been no biological change in the subject (21). The estimate of precision is then used to construct the cut-point to distinguish true change in the biomarker from measurement error.

The precision estimate is also essential in planning clinical trials. Whether the QIB is being used as an integrated or integral biomarker, the estimate of its precision is needed to predict the required trial size. Underestimating the wSD or wCV will lead to a trial that lacks adequate statistical power, and overestimating wSD or wCV will lead to a trial with a larger N (and higher cost) than needed. Thus, a reliable estimate of the precision is critical to clinical decision-making as well as understanding the potential role of the QIB in diagnosis, prediction, and treatment monitoring. These general guidelines are relevant both for the studies using conventional acquisition protocols and for more advanced quantitative imaging techniques (e.g. MR fingerprinting (140,141)) seeking translation in to clinical practice. Such methods hold promise for providing multiple quantitative MRI parameters from a fast (single) acquisition, although their specific implementations for quantitative DWI and DCE are currently sparse. Prior to test-retest precision studies, these advanced multi-parametric model-based methods would also need to demonstrate level of accuracy with respect to conventional acquisition and image analysis techniques (that quantify individual diffusion and perfusion parameters) using physical and digital phantoms that provide ground-truth parameter values.

There is a paucity of studies assessing the repeatability of imaging procedures for measuring QIBs. The published test-retest studies are often small (e.g., < 10), poorly designed (e.g., changing protocols, varying times between imaging), and their results are presented using metrics that are neither generalizable to other sites (e.g., intraclass correlation coefficients) nor lend themselves to meta-analyses.

In conclusion, QIBA recommends reproducibility and repeatability of DWI and DCE studies in phantoms and patients for identification of QIBs to be used in multicenter oncology trials.

List of Aberrations:

AIF

arterial input function

AP

anterior-posterior

CA

contrast agent

FOV

field of view

DCE

dynamic contrast enhanced

DWI

diffusion-weighted imaging

ETM

extended Tofts model

PVP

polyvinylpyrrolidone

qMRI

quantitative magnetic resonance imaging

QIBA

quantitative imaging biomarkers alliance

RC

repeatability coefficient

RL

right-left

RSNA

radiological Society of north America

SS-EPI

single-shot echo planar imaging

SPGR

spoiled gradient recalled echo

QIB

quantitative imaging biomarkers

PK

Pharmacokinetic model

wCV

within-subject coefficient of variation

wSD

within-subject standard deviation

TM

Tofts model