Functional MRI of the kidneys

Abstract

Renal function is characterized by different physiologic aspects, including perfusion, glomerular filtration, interstitial diffusion and tissue oxygenation. MRI shows great promise in assessing these renal tissue characteristics noninvasively. The last decade has witnessed a dramatic progress in MRI techniques for renal function assessment. This article briefly describes relevant renal anatomy and physiology, reviews the applications of functional MRI techniques for the diagnosis of renal diseases, and lists unresolved issues that will require future work.

INTRODUCTION

The accurate assessment of kidney function is important in both the clinic and research. Impairment in renal function precedes anatomic changes such as atrophy, and early detection of functional loss allows more options for treatment and potentially a delay or halt in disease progression. Functional assessment is also important in renal transplantation, the only effective therapeutic option after renal failure. Within the first two years, about a third of grafted kidneys manifest an episode of graft dysfunction due to various causes. Reliable functional monitoring for grafted kidney is desirable for both short-term and long-term management. The most commonly used measure of renal function is serum creatinine. However, serum creatinine has well known limitations, particularly insensitivity to early renal dysfunction, especially in chronic disease. Elevated serum creatinine is often discovered only in the late stages of disease, when substantial and sometimes irreversible renal damage takes place.

In the past decade, magnetic resonance imaging (MRI), conventionally used for tissue anatomic imaging, has been explored as an important and versatile tool for assessing the function of the kidneys. Techniques, such as dynamic contrast enhanced (DCE) MRI, diffusion-weighted MRI (DWI), blood oxygenation level dependent (BOLD) MRI and arterial spin labeling (ASL), enable noninvasive evaluation of various aspects of renal function ranging from perfusion to filtration to oxygenation. These techniques show promise in replacing the invasive techniques, e.g. oxygen micro-probe for oxygenation, which are currently limited to animal experiments.

In recent years, several reviews on the role of MRI in evaluating renal function have been published (1-6). This review article covers the recent progress in the area of functional MRI of the kidney. We begin with a short introduction of renal anatomy and physiology, followed by the overview of various MRI techniques (DCE, DWI, BOLD, ASL) that can assess renal function. We briefly discuss application of these techniques in evaluation of renal diseases and major directions for further development of renal functional MRI.

RENAL ANATOMY AND PHYSIOLOGY

Overall structure

The kidney consists of predominantly a peripheral cortex and more centrally, the renal medulla, which is configured as multiple internally situated cone-shaped tissues called pyramids (Figure 1a). The tips of each pyramid converge near the center of the kidney into the area known as a calyx. Urine collects in the renal pelvis and then flows through the ureter into the bladder. The space where renal pelvis resides, termed as renal hilum, contains the distal end of renal artery and the origin of renal vein. Each of the two (left and right) renal arteries receives ~10% of cardiac output. Within the kidney, blood perfusion in cortex is ~400-500 ml/min/100g, while medulla is less perfused, ~100-150 ml/min/100g, as measured by computed tomography technique (7,8).

Figure 1.

(a) Coronal section of the kidney; (b) schematic diagram of a nephron.

Nephron

A typical kidney contains about 1 million nephrons. Each nephron consists of a filtering component, called renal corpuscle, and a tubule where reabsorption and secretion take place (Figure 1b). The renal corpuscle, a ball-shaped structure, consists of its exterior component called a Bowman’s capsule that holds a compact bundle of capillary loops called a glomerulus. Blood in afferent arteriole enters glomerulus, gets filtered there, and exits through an efferent arteriole. Filtration results in about 20% of the incoming renal plasma passing through the glomerular membrane; first into Bowman’s capsule, then into the proximal tubule. This fluid is accompanied by solutes with molecular weight < 6000 Dalton. Glomerular filtration rate (GFR, ml/min) is normally around 60 ml/min, or a total of ~120 ml/min for both kidneys. Urine flow is only ~1 ml/min. This huge discrepancy is due to the reabsorption of glomerular filtrate back into the blood.

Tubule and peri-tubular capillaries

The renal tubule extends from Bowman’s capsule, and consists of: proximal tubule (PT), loop of Henle (LH), and distal tubule (DT). The PT and DT are mostly in renal cortex, and LH, connecting PT and DT, is deep in the renal medulla (Figure 1b). DT joins a collecting duct which ultimately terminates in a calyx. In the renal cortex, peri-tubular capillaries are distributed around the PT and DT. Some efferent arterioles extend down to medulla, giving rise to descending vasa recta (DVR) (Figure 1b). Reabsorbed water and solutes from the tubules rejoins the circulation via the capillaries and vasa recta.

Water and sodium reabsorption

In the proximal tubule, a large fraction (~65%) of sodium is reabsorbed back via Na-H antiporter. Due to osmotic force, water is reabsorbed across highly permeable PT. The descending thin limb of LH is not permeable to sodium, but ~10% of water is reabsorbed there. In the thick ascending limb, active transcellular reabsorption by Na-K-2Cl symporter becomes a dominant factor, leading to ~25% of filtered sodium. In DT, about 5% of sodium is reabsorbed by Na-Cl symporter. The transporters for active transcellular sodium reabsorption are driven by Na-K-ATPase, or sodium pumps. The main function of the Na-KATPase is to pump sodium from epithelium into interstitium, at the expense of ATP. Note that in renal medulla where perfusion is low, the active transport of sodium results in low tissue oxygenation or hypoxia state (10-20 mmHg, compared with 50-60 mmHg in cortex)(9,10). Water does not permeate the ascending limbs and the distal tubule. In the collecting ducts, water reabsorption is controlled by antidiuretic hormone (ADH or vasopressin) secreted from the pituitary gland. The bulk flow of salt and water from interstitium to peritubular capillary results from hydrostatic and oncotic forces, also called Starling forces.

Renal autoregulation

Due to the myogenic mechanism and renin-angiotensin system, renal blood flow (RBF) and GFR are relatively stable in spite of variable blood pressure in renal artery. Decreased arterial pressure leads to reduced tubular Na concentration, which is sensed by specialized cells - macula densa. Resulting stimulation of renin secretion preferentially constricts efferent arterioles and leads to preserved GFR. Angiotensin-converting enzyme (ACE) inhibitors inhibit the effect of renin, and are useful in detecting renovascular hypertension.

Diuretics

A diuretic is any drug that elevates the rate of urine production. Different types of diuretics act on different tubular segments. Water loading is widely used in MRI research to suppress secretion of ADH and reduce water reabsorption along the collecting ducts. Furosemide, on other hand, blocks sodium reabsorption in the thick ascending limb of loop of Henle, thus reducing ATP consumption and counteracting medullary hypoxia.

FUNCTIONAL MRI TOOLS

Since the function of the kidney centers on the fluid regulation, MRI is naturally suited for in-vivo measurement of various aspects of renal function, such as perfusion, filtration, diffusion and oxygenation (Table 1).

Table 1.

MRI tools for renal functional assessment

Physiologic parameters	MRI tools
Perfusion	ASL, DCE, DWI(IVIM)
Glomerular filtration	DCE
Interstitial diffusion	DWI
Tissue oxygenation	BOLD
Tubular transit, Sodium concentration	Largely unexplored

Dynamic contrast enhanced (DCE) MRI

Before the advent of DCE MRI, dynamic renal scintigraphy using radioactive technetium-99m based tracers has been widely used for measuring renal function. DCE MRI for the kidney, also termed MR renography (MRR), has advantages over scintigraphy including no radiation exposure, inherently three-dimensional acquisition, and superior spatial resolution. Unlike scintigraphy, MRI can easily resolve the renal cortex from the medulla and collecting system. Through more realistic modeling DCE-MRI offers the potential to reveal in vivo renal properties that in the past could only be examined with biopsies, opening new windows into renal pathophysiology and clinical diagnosis.

Tracer kinetics and modeling

One key aspect of DCE-MRI analysis is the choice of model used to describe the transit of contrast in renal tissue. As reviewed recently (11), at this time there is no consensus about the best model for MRR.

A two-compartmental model, restricted to vascular and tubular compartments, has been introduced by several groups. The major attraction of such model, especially its graphical Patlak-plot “upslope” implementation (12), is its simplicity. However, these two-compartmental models ignore the transit of the tracer out of the tubular compartment and strongly depend on the arbitrary time interval over which a slope is calculated. Annet et al (13) refined the two-compartment model by introducing the delay and dispersion of the bolus between the aorta and renal vasculature and accounting for the efflux from the tubular compartment as an exponential decay. Sourbron et al (14) extended this model for measuring both renal perfusion and GFR. A comparison study by Buckley et al (15) found that the refined model fitted experimental data better but it overestimated GFR. Koh et al (16) validated a more elaborate parallel two-pathway model, vascular (from renal artery to vein) and tubular (from renal artery to tubules).

More elaborate modeling is predicated on our ability to measure contrast in two subregions of the kidney, cortex and medulla (17). Such three-compartment models were developed by Lee and Zhang et al (18,19). Three compartmental models have the potential of providing functional information about tubules. The papers by Buckley et al (15) and Bokacheva et al (11) compared some of the models, and provide some recommendations for their use.

In addition to the kinetic modeling of DCE data, there are other considerations worth noting, including the effects of errors in arterial input function estimates and efforts to convert MR signal intensity into gadolinium concentrations for modeling, each of which are discussed briefly below.

Estimating arterial input function (AIF)

Kinetic analysis of DCE data requires the knowledge of AIF. MR signals from arteries are susceptible to several MR artifacts, including inflow effect, dephasing, partial-volume effect and flow pulsation. Figure 2 demonstrates the impact of inflow artifact. Mendichovszky and Cutajar et al (20,21) found that AIF strongly depends on the size of aortic ROI. Generally, two solutions have been proposed. Bauman and Rudin (22) used cortical signal as surrogate for arterial input and medullary signal as tissue retention function. Zhang et al (23) proposed a similar method but used the cortex-medulla difference as the tissue retention function for GFR estimation. This latter method utilized the fact that vascular and tubular pathways are coupled with each other and the subtraction of cortical and medullary signals gives roughly tracer retention in the collecting ducts. One drawback of these surrogate methods is that they do not provide information on renal blood flow.

Figure 2.

Variability of AIF in DCE MRI. Regions of interest at different levels of aorta (a) suffer from different degrees of inflow artifact (b), resulting in AIFs of different magnitudes (c).

Other studies have proposed to improve the measured AIF. Parker et al (24) and Wang et al (25) averaged AIFs obtained from a group of controls and applied this population-averaged AIF to the analysis of new patients. Since AIF partially depends on the cardiac output (CO) of the patient, this method assumes a relatively homogeneous population. Roberts et al (26) proposed a double-bolus technique to compensate mis-sampling of the first pass of AIF at low temporal resolution and showed that the method improved the precision of parameters. Zhang et al (27) developed a method for constraining AIF using CO of the patient, based on indicator dilution theory, which relates the injected dose, CO, and the area under the first pass of AIF. Using phase-contrast MRI, CO can be conveniently measured in <1 minute in the same imaging session as DCE MRI.

Quantification of tracer concentration

Quantitative analysis of MRR data typically requires image data to be converted to concentration of tracer in the tissue of interest. For widely used T₁-wieghted MRR, tracer concentration is related to tissue 1/T₁⁰ by the following linear relationship,

$$C^{Gd}=\left(\frac{1}{T_1}-\frac{1}{T_1^0}\right)\frac{1}{r_1}$$ (1)

where C^Gd is the time-varying concentration of contrast medium, T₁ is the time varying longitudinal relaxation, T₁⁰ is the spin-lattice relaxation time in absence of contrast medium, and r₁ is the T₁ relaxivity of the contrast medium. T₁ values can be derived from the acquired signals and the appropriate sequence-specific formula (28).

An estimate of tracer concentration from Eq. [1] requires knowledge of r₁ and T₁⁰ in renal tissue and blood. For Gd-DTPA at 1.5T, Shuter et al (29) measured r₁ to be 1.2 s⁻¹mM⁻¹ for cortex and 1.3-1.6 s⁻¹mM⁻¹ for medulla. Morkenborg et al (30) measured cortical r₁ to be 1.1 s-¹mM⁻¹. These reduced values compared with plasma r (4-5 s⁻¹mM⁻¹ 1 (30)) may be attributed to compartmentalization between reabsorbed water and tubular water, as analyzed by Donahue et al (31). T₁⁰ can be measured by several rapid methods, such as variable flip angle method (32) and inversion recovery method (33,34).

Limitations and future needs of DCE-MRI 1

1. Quantification of tracer concentration in renal tissue relies on the determination of kidney in vivo T₁ values and on relaxivity of gadolinium, which may differ from in vitro estimates.

2. Over the time course of 3 – 5 min, when gadolinium contrast transits through the kidneys, there may be significant movement artifact in the MRI data. Manual correction of such movement artifact can be avoided by automatic registration techniques (35,36). The use of coronal imaging and the consequent restriction of movement to being mostly in-plane reduce the complexity of coregistration software.

3. Tracer kinetic analysis of DCE MRI data requires accurate segmentation of the kidneys and, for applying some complex models, separation of renal cortex and medulla (17,37,38). Mis-segmentation could cause partial volume effect that lowers the accuracy of GFR estimates (39).

4. Tracer kinetic models to interpret gadolinium concentration curves often depend on accurate measures of the arterial input function. Coronal acquisition helps minimize AIF inflow artifact. Moreover, some regularization of the AIF might give improvements (27).

5. Some Gd-based contrast agents may cause nephrogenic systemic fibrosis (NSF) in patients with diminished renal function (40,41). Non-contrast MRI techniques as we will discuss have been intensively explored in recent years.

Applications

With the capability of measuring single-kidney GFR, MR renography has the potential of diagnosing several renal diseases.

1. Renovascular hypertension (RVH): RVH as one type of correctable hypertension is caused by renal artery stenosis (RAS). Arterial narrowing could be due to atherosclerosis or fibromuscular dysplasia. Because patients can have idiopathic hypertension, it is important to identify correctly patients who have clinically significant RAS and who will benefit from revascularization surgery. Captopril renography using nuclear medicine techniques have been widely used in the past. This method can also be adapted to MRI where GFR measurementsbefore and after angiotensin-converting enzyme (ACE) inhibitor are compared to determine whether the RAS has activated a renin-mediated hypertension (42). In 1996, Grenier et al (43) first demonstrated the feasibility of ACE-inhibitor renography using MRI. Recently, Lee et al (44,45) improved the protocol by using low-dose contrast medium and shortening the procedure into <0.5hr, and analyzed the data with newly developed tracer kinetic modeling techniques.

2. Functional urinary obstruction: Urinary obstruction often causes hydronephrosis (a dilation of renal pelvis calyces) and eventually leads to renal failure. Renal function loss due to urinary obstruction can be assessed by measuring ‘differential renal function (DRF)’ from dynamic MR urography (46). Methods for calculating DRF are based on either the parenchyma volume (47), the area under tubular phase of signal vs. time curve (48), or Rutland-Patlak plot (49). In addition, renal transit time (RTT), defined as the time between tracer appearance in the kidney and in ureter (50), was shown to be useful in differentiating between normal and obstructive kidneys. The combination of high-resolution morphologic imaging, DRF and RTT, from a single MRI session, provides a comprehensive tool for diagnosing urinary obstruction.

3. Renal transplant: Complications in the early post-transplantation period (~1-2 weeks) could lead to delayed graft function and impact the long-term outcome of renal grafts. The major complication is acute tubular necrosis (ATN), which is essentially an ischemic injury to the tubules. ATN is characterized by reduced blood flow, reduction in GFR and tubular dysfunction. Figure 3 compares perfusion maps of a healthy kidney and an ischemic transplant. Other causes include acute rejection (AR) of different types, arteriole or venous thrombosis, nephrotoxicity and ureteric obstruction.

Figure 3.

Comparison of DCE MRI for a healthy kidney and a partially ischemic transplant. (a) One image of the healthy kidney from post-contrast vascular phase; (b) the perfusion map of the healthy kidney generated from vascular-phase images; (c) one image of the renal transplant from post-contrast vascular phase; (d) the perfusion map of the transplant showing partial ischemia.

In 1997, Szolar et al (51) aimed to differentiate ATN, AR, normal graft cases using a semi-quantitative analysis of MR enhancement curves and found that, compared to normal cases, AR cases displayed lower cortical and medullary enhancement curves, while ATN cases showed distinct medullary enhancement curves. Agildere (52) demonstrated the potential of DCE MRI in differentiating AR and cyclosporine nephrotoxicity. Wentland et al (53) estimated cortical and medullary perfusion for renal transplants, and found that cortical and medullary perfusion of AR was significantly lower than that of normal and ATN cases. Their results were in agreement with Szolar et al (51). However, perfusion information alone could not differentiate ATN and normal cases. Realizing that ATN might affect tubular transit, Yamamoto et al (54) estimated mean transit times (MTT) using a three-compartment model and found a higher ratio of tubular MTT over whole-parenchyma MTT than successful transplant.

Diffusion-weighted MRI

Diffusion-weighted MRI (DWI) is designed to probe random microscopic motion, or diffusion, of water in tissue, and does not require exogenous contrast agent. The diffusion sequence incorporates a bipolar gradient into a spin echo sequence (Figure 4).

Figure 4.

Diagram of typical DWI sequence. δ is the duration of diffusion-encoding gradient lobe, Δ is the time between the gradient lobes, G is the magnitude of the gradient.

The dependence of measured signal S on the b values is given as:

$$S\left(b\right)=S\left(0\right)\cdot e^{-b\cdot \mathit{ADC}}$$ (2)

where ADC is apparent diffusion coefficient. ADC is one of the simplest parameters that reflect tissue microstructure. The estimation of ADC can be done by fitting signals acquired at different b values by Eq. [2]. In a detailed review, Kim et al (55) suggested a standardized sequence: breath-hold or respiratory triggered (using a navigator echo), single-shot echo-planar imaging, tri-directional gradients using b-values: 0, 400, and 800 sec/mm².

Applications of ADC in renal fibrosis and dysfunction

Normal renal function involves multiple processes of water transport, while worsening renal function (reduction in GFR) should lead to a decrease in water reabsorption, i.e. a lower rate of water transfer across interstitial space. This would potentially reduce diffusion. In addition, fibrosis formed in the process of chronic renal dysfunction may also restrict water diffusion. Several researchers have shown a positive correlation (R=~0.8) between ADC and GFR, estimated either with serum creatinine level (56,57) or renal scintigraphy (58,59). Recently Togao et al (60) studied the relationship between renal fibrosis and ADC in an animal model of unilateral ureteral obstruction (ureteral obstruction induces fibrosis in ~1 week), and found that the ADC of the obstructed kidneys decreased to 0.7×10⁻³ mm²/sec on day three, and to 0.57×10⁻³ mm²/sec on day seven (Figure 5). ADC of contralateral kidney remained stable at 0.9-1.0×10⁻³ mm²/sec. As speculated by the authors and Thoeny et al (61), the decrease in ADC could be due to the increase in cell density in interstitial space.

Figure 5.

Coronal ADC maps of the kidneys in a mouse before UUO (day 0) and on days 3 and 7 after UUO. Reprinted with permission from RSNA (60).

Intra-voxel incoherent motion (IVIM) effect

IVIM extends the mono-exponential model from Eq. [2] by including microscopic motion attributable to microcirculation of blood (62,63),

$$S\left(b\right)=S\left(0\right)\cdot [F_p\cdot e^{-b\cdot D_p}+(1-F_p)\cdot e^{-b\cdot D_r}],$$ (3)

where F_P is fast-component fraction, D_T is tissue diffusion coefficient, and D_P is pseudo-diffusion coefficient due to IVIM effect. With high capillary density and high blood flow, the kidney is expected to show a strong IVIM effect. Zhang et al (64) found that ADC values from mono-exponential fitting depend heavily on the b values used (Figure 6), and concluded that that mono-exponential model is not adequate for renal DWI data. The advantage of bi-exponential model in fitting renal DWI data was confirmed by Wittsack et al (65) who compared mono-exponential, bi-exponential and a distribution function model.

Figure 6.

ADC value from mono-exponential analysis depends heavily on the b values used: (a) ADC map from b values 0 and 400 s/mm²; (b) ADC map from b values 0 and 800 s/mm²; (c) ADC map from all b values from 0 to 800 with interval of 50 s/mm². For comparison, D_T map from bi-exponential analysis is shown in (d).

Limitations and future needs of IVIM

While mounting evidences suggest that the IVIM parameters may provide a non-invasive biomarker for renal perfusion, more work is needed.

1. DWI data of high quality is needed to improve the estimation precision of IVIM parameters, since error propagation from signal to F_P and D_P is much higher than that to ADC (66). Respiratory motion is a major problem that degrades DWI image quality.

2. The relative contributions of blood flow (perfusion) and tubular flow to the IVIM parameters are unclear. Quantitative measurement of perfusion based on DWI would require the separation of these different flow components that may contribute to the fast diffusion-decay component.

Blood oxygen level dependent (BOLD)

Although the kidney receives a high fraction of cardiac output (~10%), the renal medulla especially the inner medulla works in a state of hypoxia (O₂ partial pressure pO₂ ~10-20 mmHg). The hypoxic state of medulla is caused by 1) low oxygen delivery due to low vascular density in medulla and arterial-venous shunting, and 2) high oxygen consumption for active transcellular transport of sodium and chloride. Since renal tissue oxygenation is an important marker of renal dysfunction, there is considerable interest in noninvasive techniques to estimate intrarenal oxygenation. One possible approach is based on BOLD MRI. The presence of deoxyhemoglobin generates a susceptibility difference that locally perturbs the magnetic field and increases the spin-spin relaxation rate R₂* of surrounding water (67). High R₂* estimate corresponds to low tissue oxygenation. R₂* is usually derived from multi-echo gradient recalled echo (mGRE) sequence (68).

Diuretic-enhanced BOLD

The frequently used diuretic in BOLD kidney experiments is furosemide, which acts by inhibiting Na-K-2Cl symporter in TAL of loop of Henle, thereby decreasing the activity of sodium pumps, and thus decreasing oxygen consumption. In 1996, Prasad et al (69) showed that both furosemide and water loading induced a decrease in medullary R₂* (increase in pO₂). A time-resolved approach (70) was then implemented to record the temporal response of medullary R₂* after water loading. Reasonable reproducibility of R₂* measurement (3%-12%) has been evaluated by multiple studies (71-73).

Limitations and future needs of BOLD

First, high field strength such as 3T induces strong BOLD contrast, but may also cause severe bulk susceptibility artifact, e.g. due to the presence of bowel gas. Second, BOLD effect (or R₂*) relates to tissue oxygenation, but also depends on other variables, such as perfusion, vascular volume fraction and intrinsic spin-spin relaxation R₂. Better quantification of renal tissue oxygenation from BOLD data requires separating these other effects out.

Applications

1. Renal artery stenosis (RAS)

There is increasing evidence that BOLD is sensitive to subtle intrarenal change induced by RAS, enabling the prediction of possible renal failure at an earlier stage. Juillard et al (74) observed increased R₂* in a pig model during progressive constriction of renal artery, indicating the important role of renal blood flow in maintaining renal tissue oxygenation. With the decrease in renal blood flow, oxygen content was consistently lower in the cortex than in the medulla, while in the contralateral kidney without artery constriction, oxygen content was slightly lower in medulla. This suggests that in progressive RAS, blood flow is better preserved in the medulla than in the cortex.

Textor et al (75) compared cortical and medullary R₂* values and their change after furosemide in human kidneys without RAS, with RAS but preserved kidney volume (‘viable’), and with total artery occlusion and renal atrophy (‘nonviable’). Normal kidneys experienced a rapid fall in R₂* values after furosemide, similar to Prasad’s study, due to the inhibition of Na reabsorption in TAL. The kidneys with RAS but preserved function had elevated R₂* (lower oxygenation) values at baseline, which dropped after furosemide. The atrophic kidneys with total arterial occlusion had reduced R₂* values and minimal change after furosemide, which was similar to renal allografts with acute dysfunction.

2. Renal transplant

The noninvasiveness feature of BOLD and its sensitivity to multiple physiologic factors make it suitable for the assessment of renal transplants. Current applications focus on the differentiation between acute rejection (AR) and acute tubular necrosis (ATN) as the cause for delayed graft function. Sadowski et al (76) showed that the medullary R₂* of AR cases was lower than that of normal or ATN cases, and cortical R₂* of ATN cases was higher than that of normal or AR cases. Djamali et al (77) obtained similar results, except that a significant reduction in medullary R₂* was detected in ATN, compared with normal cases. Compared to normal kidney, the ratio (M/C) of medullary R₂* to cortical R₂* was reduced in ATN and lowest in AR. The decreased M/C in ATN and AR are probably due to decreased oxygen consumption in delayed graft function. However, Han et al (78) observed abnormally higher medullary R₂* and M/C in ATN. This discrepancy was attributed to the shorter surgery to scan time (~10 days in Han et al compared with 29 days in Djamali et al). In the early stage of ATN, acute ischemia compromises oxygen delivery while oxygen consumption by the reabsorption is maintained. This results in a temporary decrease of oxygen bioavailability.

3. Diabetes

Diabetic patients are at high risk for developing chronic kidney disease. Renal damage occurs in multiple stages (79). Stage 1: at the onset of diabetes, there is an increase in the size and in glomerular filtration (hypertrophy and hyperfiltration); Stage 2: at 2-5 years after onset, if blood sugar level remains high, damage to glomerulus and thickening of the basal membrane occurs; Stage 3: at 5-15 years, if blood sugar level remains high, albumin appears in urine, and GFR and RBF decrease; Stage 4: urine albumin exceeds 200mg/L, indicating a serious damage to glomerulus. Accumulation of albumin and other protein blocks blood flow to capillaries, leading to end stage renal failure.

Epstein et al (80) compared mild diabetic patients (presumably stage 1 diabetes) and normal subjects of similar age with BOLD MRI. During water diuresis, medullary R₂* decreased in the normal subjects, but remained unchanged in the diabetic patients. Since a decrease in R₂* represents an increase in tissue oxygenation, this result presumably reflects a deficiency in medullary vasodilatation in diabetic subjects. Inducing diabetes in rats, dos Santos et al (81) detected a progressive decrease in pO₂ of both the cortex and medulla from day 0 to day 28 after the induction of diabetes, suggesting increased activity of sodium pumps. Similar result was found by Ries et al (82) (Figure 7). These studies support the use of BOLD to monitor changes in renal hemodynamics in diabetic patients at an early stage.

Figure 7.

T₂* maps of (a) healthy and (b) diabetic kidneys for comparison. In both cases, the lowest T₂* values clearly are found within the outer stripe of the outer medulla. Note that low T₂* value corresponds to low oxygenation. Reprinted with permission from John Willy and Sons (82).

4. Other applications

BOLD imaging has the potential for a broad range of other clinical indications. With unilateral ureteral obstruction (UUO), an animal study with pigs after 24-hour UUO has shown increased cortical R₂* (i.e. lower oxygenation) and decreased medullary R₂* (i.e. higher oxygenation) (83). In human subjects with UUO (with duration 5-47 hours), Thoeny et al (84) observed decreased medullary R₂* and decreased cortical R₂*. The different change in cortical R₂* between the two studies could be due to the race difference, or the different durations of UUO. In patients with acute renal failure, Xiao et al (85) observed decreased cortical R₂* and even more decreased medullary R₂*, indicating increased oxygenation level in both cortex and medulla.

Arterial spin labeling (ASL) – non-contrast perfusion imaging

Originally developed for measuring cerebral perfusion (86,87), ASL uses magnetically labeled protons in arteries as endogenous tracer and is therefore completely non-invasive. Renal ASL has been validated by several studies, most of which used FAIR (flow sensitive alternating inversion recovery) for perfusion preparation and True-FISP (true fast imaging with steady precession) for data acquisition. Because SNR is low in ASL, multiple repetitions and high field (3T) are needed. ASL signal decays with blood T₁, so acquisitions must be performed within 1-2 sec of arterial labeling. Because acquisitions typically require labeled and unlabeled measurements and their subtraction, motion artifacts can impede accurate measurements. Song et al (88) demonstrated high-quality ASL images with free breathing by incorporating dual-navigator technique into FAIR and correcting mismatch between labeling slice and image slice (Figure 8). Most of the studies use a one compartment ASL model. Kiefer et al. (89) recently reported on a two-compartment model that accounts for delayed exchange between vascular and extra-vascular spaces. Such models require higher SNR and acquisition of multiple inversion time points.

Figure 8.

Axial perfusion rate maps obtained by FAIR True-FISP with (a) and without (b) the dual navigator-gated technique in the same scale. The kidneys are clearly delineated in the dual navigator-gated FAIR True-FISP perfusion image (a). The perfusion values in (b) (without navigator gating) are overestimated because of respiratory motion. Reprinted with permission from John Willy and Sons (88).

In agreement with more established invasive techniques, ASL MR studies have estimated renal perfusion at ~300 ml/min/100g for cortex and ~100 ml/min/100g for medulla of healthy kidneys. Artz et al (90) showed good correlation between ASL measured perfusion and fluorescent microsphere method in a swine model. However, ASL values were consistently lower than microsphere estimates. More research is needed to validate ASL against reference technique in human subjects.

Other MRI techniques

We have focused on some of the main areas of functional renal MRI studies. Many other approaches are also in the exploratory phase including diffusion tensor imaging (DTI) to study the directionality of water diffusion (91-93), Na MRI to study corticomedullary sodium gradient (94-96) and MR elastography to evaluate tissue mechanical property (97,98).

FUTURE DIRECTIONS

MRI tools appear promising in estimating physiologic parameters for the diagnosis and monitoring of renal diseases. To achieve the promise, future studies are needed to resolve the following issues.

1. Given the challenges of respiratory and bulk motion, adjacent bowel gas, and spatial resolution demands, technical improvements in image quality are critical to advances in functional renal MRI.

2. Validation of quantitative models is critical for relating measured MRI data with specific and meaningful physiologic parameters. Examples include (1) the interpretation of BOLD measurements for oxygenation and perfusion, and (2) the differentiation of perfusion, filtration, and resorption changes on DWI and IVIM measurements.

3. With the broad array of tools available in renal MRI, we must combine multiple (especially non-contrast) MRI protocols in a same scan. Clinical studies are needed to refine multimodality applications.

4. Robust image registration and segmentation. Image registration to correct for motion artifacts and the segmentation of renal cortex and medulla have been major issues in the analysis of renal images. Readily available tools for co-registration of images from different MR sequences are also needed.

The trajectory of functional renal imaging over the past decade and the growth in interest in the field promise a bright future ahead. Without question, research advances in this field will benefit substantial patient populations with high impact.