Hepatobiliary MRI: Current Concepts and Controversies

Transcription

1 JOURNAL OF MAGNETIC RESONANCE IMAGING 25: (2007) Review Article Hepatobiliary MRI: Current Concepts and Controversies James F. Glockner, MD, PhD* Evaluation of the liver and biliary system is a frequent indication for abdominal MRI. Hepatobiliary MRI comprises a set of noninvasive techniques that are usually very effective in answering most clinical questions. There are significant limitations, however, as well as considerable variation and disagreement regarding the optimal protocols for standard hepatic MRI and magnetic resonance cholangiopancreaticography (MRCP). This review discusses pulse sequences most often used in hepatic MRI and MRCP, examines a few sources of controversy in the current literature, and summarizes some recent and future developments in the field. Key Words: hepatic MRI; MRCP; hepatic malignancy; image optimization; hepatic cirrhosis J. Magn. Reson. Imaging 2007;25: Wiley-Liss, Inc. efficacy for some of the most common indications. This review is an attempt to provide a broad overview of the current state of the art of hepatobiliary MRI. While nearly any of the topics addressed in this article would in itself constitute a suitable topic for a review article, general concepts, controversies, and comparisons with other imaging modalities are emphasized. Common indications for MRI and MRCP are discussed, followed by an examination of standard techniques for both of these examinations, emphasizing advantages and limitations of the most common pulse sequences. Recent technical and clinical developments are examined, and some of the current controversies in the field are analyzed. Finally, future developments in hepatobiliary MRI are assessed. INTRODUCTION HEPATOBILIARY MRI is very much a work in progress. Many positive developments have occurred in recent years, including the wide availability of high performance systems allowing rapid acquisition of high resolution images. New contrast agents and pulse sequences are available, which in many circumstances improve the flexibility, sensitivity, and accuracy of hepatic MRI and magnetic resonance cholangiopancreaticography (MRCP). On the other hand, protocols in private practice and academic institutions vary widely, and there is considerable disagreement as to the correct way to perform a hepatic MRI or MRCP examination, which pulse sequences are most useful, and whether MRI is truly more valuable than computed tomography (CT) or sonography in most cases. While the rapid proliferation of pulse sequences and protocols likely reflects a gradual evolution in the overall quality of these exams, it can be frustrating to busy clinical radiologists confronted with new techniques and sequences on a monthly basis, many of which have little documented Department of Radiology, Mayo Clinic, Rochester, Minnesota, USA. *Address reprint requests to: J.F.G., MD, PhD, Mayo Clinic, Dept of Radiology, 200 First Street SW, Rochester, MN glockner.james@mayo.edu Received November 14, 2005; Accepted September 28, DOI /jmri Published online 12 March 2007 in Wiley InterScience ( wiley.com). COMMON INDICATIONS Hepatic MRI for much of its existence has functioned primarily as a problem solving technique in cases in which sonography or CT is limited or indeterminate. In this capacity, for example, MRI can be very useful for confirming the diagnosis of hemangioma, focal nodular hyperplasia, complex cyst, etc. without the need for biopsy, surgery, or multiple follow-up examinations. Primary indications for hepatic MRI are somewhat more controversial, but include screening for hepatocellular carcinoma (HCC) or cholangiocarcinoma in patients with cirrhosis, evaluation of potential living related donor transplant candidates, and screening potential transplant recipients. Hepatic metastases are often assessed for staging or prognostic purposes, and MRI can detect metastases prior to surgical resection, radiofrequency (RF) ablation, or chemoembolization, and then assess response and monitor for recurrence. MRI is probably the test of choice to detect and characterize diffuse hepatic disease, including, cirrhosis, steatosis, and hemochromatosis. While evaluation for hepatic infection is often performed with CT, MRI may be more sensitive in some cases, such as fungal infection. MRCP is commonly employed as a noninvasive alternative to endoscopic retrograde cholangiopancreatography (ERCP) to evaluate the biliary tree for obstruction and/or to determine the cause of obstruction in patients with suspected choledocholithiasis or biliary or pancreatic mass. MRCP can assess the biliary anatomy of potential living related transplant donors and recip Wiley-Liss, Inc. 681

2 682 Glockner Figure 1. Utility of in-phase/out-of-phase images in lesion characterization. Arterial (a) and portal venous (b) phase images from CE 3D SPGR acquisition in patient with hepatic adenomatosis reveal early-enhancing lesions. In-phase (c) and out-of-phase (d) images reveal signal dropout in the lesions on out-of-phase images, confirming the presence of fat, and making adenomatosis the most likely diagnosis. ients. Patients with suspected primary sclerosing cholangitis, Caroli disease, choledochal cyst, or other biliary pathology can often be evaluated with MRCP without the need for an invasive ERCP. Complications of transplantation or cholecystectomy such as biliary stricture or leak can be detected with MRCP. MRI is obviously an excellent choice when there is a contraindication to CT such as a severe contrast allergy. MRI may be also preferred in patients who will have frequent follow-up examinations in order to prevent large cumulative radiation doses. STANDARD TECHNIQUES Hepatic MRI Most examinations include a T1-weighted in-phase/ out-of-phase spoiled gradient echo sequence, one or more T2-weighted sequences, and a dynamic contrastenhanced (CE) fat-saturated spoiled gradient echo sequence. In-phase/out-of-phase images are now usually obtained in a single breathhold as a double-echo acquisition, thereby reducing or eliminating misregistration between in-phase and out-of-phase images. These sequences are useful for detecting focal or diffuse fatty infiltration and identifying lesions containing fat (Fig. 1), as well as noting excessive iron deposition in patients with hemosiderosis or hemochromatosis. The inphase TE should be longer than the out-of-phase TE so that signal dropout on out-of-phase images is definitely the result of opposed phase fat signal and not substances with short T2* (such as iron). T2-weighted sequences have long been a source of controversy. Conventional spin echo sequences were the first widely available technique: these required long acquisition times (often more than 15 minutes) and were frequently degraded by motion artifact. Fast spin echo (FSE) sequences were then introduced, and for the most part rapidly adopted, due to their much shorter acquisition times and generally excellent image quality; however, some investigators found that conventional spin echo images were superior to FSE images for solid lesion detection (this is thought to be related to increased magnetization transfer and J-coupling effects with FSE sequences) (1), while others favored FSE images or found no significant differences (2,3). The practical disadvantages of conventional spin echo T2- weighted sequences have largely made this discussion moot, since they are rarely used today; however, similar debates are occurring regarding respiratory-triggered FSE sequences, breathheld FSE sequences, and singleshot FSE (SSFSE) sequences. Some authors have gone farther, and contend that T2-weighted images are largely superfluous, and seldom provide any information not evident on dynamic CE spoiled gradient recalled (SPGR) sequences (4 6). This is undoubtedly true in many situations; however, there are circumstances in which T2-weighted images are valuable (distinguishing regenerative nodules from HCC, for example), and it is worth remembering that dynamic

3 Hepatic MRI: Concepts and Controversies 683 Figure 2. T2-weighted imaging in hepatic metastatic disease. Fast recovery FSE image (a) in patient with metastatic smallcell lung cancer shows poor lesion-liver contrast. Respiratorytriggered FSE image (b) has improved SNR and lesion liver contrast. Diffusion-weighted echo planar image with low b- value (c) shows the best lesion-liver contrast, but with reduced SNR. sequences are not ideal for all patients. Poor bolus timing, poor breath-holding, and compromises between temporal and spatial resolution can all diminish image quality, and in these cases it is useful to have alternative techniques available that do not have similar limitations. Related arguments involve the question of which of the many available T2-weighted sequence is optimal for a particular indication. SSFSE and half-fourier acquisition single-shot turbo spin-echo (HASTE) sequences are generally half-fourier acquisitions in which all phase encoding steps are obtained following a single excitation. These images can be acquired quickly, and image quality is often adequate in patients not able to suspend respiration. SS- FSE sequences offer acceptable depiction of anatomy, and are quite sensitive for detection of cysts, hemangiomas, and other fluid-containing lesions. Limitations of SSFSE sequences for standard hepatic imaging include image blurring related to the very long echo train lengths (ETLs), which worsens as the resolution in the phase-encoding direction increases. Minimizing echo spacing as well as shortening ETL with strategies such as parallel imaging may aid in reducing these artifacts. SSFSE images are also probably less sensitive for detection of solid hepatic lesions (3,7,8), although opposing points of view are certainly represented in the literature (9 11). Some practices perform SSFSE sequences as the sole T2-weighted series, recognizing that, although this technique does have some limitations, the images are acquired quickly and reliably, and most potential missed findings are seen on dynamic CE SPGR images. Breathheld FSE, fast-recovery FSE, and fast inversion-recovery FSE sequences are probably the most commonly employed T2-weighted sequences in standard hepatic MRI examinations. Fast-recovery FSE sequences employ additional radiofrequency pulses at the end of each excitation to return residual magnetization to the z-axis, thereby allowing reductions in TR while still preserving T2 weighting. Inversion-recovery sequences, in which the inversion time (TI) is chosen to null the signal from fat, generally exhibit more uniform fat suppression at a cost of slightly lower signal-tonoise ratio (SNR) and slightly longer acquisition times. These breathheld sequences offer improved image quality and probably improved sensitivity for solid lesion detection in comparison to SSFSE sequences; however, the acquisition times are considerably longer, and a small but significant percentage of patients have difficulty holding their breath for the entire acquisition. Respiratory-gated FSE T2-weighted sequences represent an alternative to the breathheld sequences described above. Respiration is monitored by a respiratory belt that detects expansion of the chest or a navigator pulse (12) that detects the position of the diaphragm, and acquisition is prospectively or retrospectively gated to the most stable portion of the respiratory cycle (typically end-expiration). These sequences offer improved SNR and/or spatial resolution, since the number of excitations and phase encoding steps is not limited by breathhold capacity. In our experience, respiratory-triggered sequences offer improved visualization of solid lesions compared to the breathhold techniques (Fig. 2) (13,14), although several studies have reached opposite conclusions (15,16), and the improved image quality is appreciated only if the respiratory gating is effective at minimizing motion artifact. In addition to potential motion artifact from suboptimal respiratory gating (exacerbated in patients with ascites) these sequences require relatively long acquisition times, typically on the order of three to six minutes. T2*-weighted gradient echo sequences are used most frequently in combination with superparamagnetic iron

4 684 Glockner oxide (SPIO) contrast agents. Fast gradient echo or spoiled gradient echo breathheld sequences provide T2* weighting, emphasizing the negative contrast of SPIO agents. Lesions without reticuloendothelial cells do not take up contrast and therefore appear bright against the dark background of normal liver. Echo times (TEs) of 8 15 msec have been found to provide the best compromise between lesion/liver contrast and adequate SNR (17). Steady-state free precession (SSFP) sequences (fast imaging employing steady-state acquisition (FIESTA), true fast imaging with steady state precession (True- FISP), balanced fast field echo (FFE)) are fully-balanced fast gradient echo sequences providing T2/T1 contrast weighting. SSFP sequences were initially described in the 1980s (18); however, very short repetition times (TRs) are required to minimize artifacts, and gradient hardware required to accomplish this has been widely available for only a few years. An additional advantage of very short TRs is short acquisition times. SSFP images can be acquired even more quickly than SSFSE images, and are also sensitive for detecting cysts, hemangiomas, and other fluid-containing lesions, but provide relatively poor visualization of most solid lesions. SSFP sequences are useful for evaluation of hepatic vasculature, particularly the hepatic and portal veins. SSFP sequences may benefit from gadolinium-based contrast administration, since T1-shortening results in increased signal intensity. The benefits of contrast have been demonstrated in three-dimensional (3D) SSFP vascular sequences in coronary MR angiography (MRA) as well as in evaluation of portal vessels (19); however the role of contrast in hepatic parenchymal SSFP sequences has received relatively little attention (20). Echo-planar imaging (EPI) sequences offer excellent image contrast for both cystic and solid lesions and relatively short acquisition times. Several authors have demonstrated advantages of echo planar acquisitions in comparison to other standard T2-weighted sequences (1,7,13,21). Limitations of EPI sequences have long been noted, however, including significant susceptibility artifacts near the diaphragm and adjacent to gas-filled bowel loops. These can be reduced by using multishot techniques, high receiver bandwidths, and parallel imaging. Diffusion-weighted imaging using EPI sequences offers additional opportunities for characterization of diffuse hepatic disease as well as focal lesions. A recent article describing increased conspicuity of solid and cystic lesions with diffusion-weighted imaging using low b-values raises interesting possibilities with regard to the future role of these techniques (Fig. 2) (22). Dynamic fat-saturated CE SPGR sequences are often the most valuable in the hepatic MRI examination. Both two-dimensional (2D) and 3D techniques are widely used in clinical practice, with current trends favoring 3D acquisitions. 3D sequences offer advantages of seamless volumetric coverage without interslice gaps as well as improved SNR relative to 2D acquisitions (Fig. 3) (23,24). One disadvantage of 3D SPGR sequences is that image contrast is slightly diminished in comparison to 2D sequences, perhaps a result of the low flip angles and low TRs required for reasonable acquisition times. These sequences are performed in conjunction Figure 3. Dynamic CE fat-saturated 3D SPGR images in patient with large HCC. Parallel imaging is employed to reduce acquisition time and allow improved spatial resolution. Test bolus allows accurate determination of scan delay to obtain arterial phase images. Note subtle linear enhancement in portal vein, indicating tumor thrombus, in arterial phase image (a), with improved depiction of ill-defined tumor in right hepatic lobe and portal vein thrombus in portal venous (b) and equilibrium phase (c) images.

5 Hepatic MRI: Concepts and Controversies 685 with bolus injection of gadolinium-based contrast agents, typically at a dose of mm/kg and rate of 2 3 ml/second. Bolus timing can be a crucial element of a successful examination for lesions demonstrating arterial phase enhancement and rapid washout (HCC, neuroendocrine tumor metastases, focal nodular hyperplasia, etc.). While a generic delay of seconds following contrast injection generally yields acceptable results, variations in contrast circulation times can be quite large, and the use of a test bolus or fluoroscopic triggering is warranted in cases in which optimal arterial phase images are important. This is greatly simplified if an automatic injector is available. The art of planning dynamic SPGR acquisitions lies in selecting the correct compromise between temporal and spatial resolution: acquisition times must be short enough to be encompassed within a comfortable breath-hold as well as to allow visualization of lesions with rapid transient enhancement, while the spatial resolution must be high enough to detect reasonably small lesions. Fat Saturation Techniques Most T2-weighted sequences as well as dynamic CE sequences benefit from fat saturation. The primary benefit is an improvement in the conspicuity of solid lesions. Fat saturation also reduces phase ghosting artifact from subcutaneous and intraperitoneal fat, which is particularly important in respiratory-triggered sequences. Spectral fat saturation, in which frequency-selective pulses saturate the signal from fat prior to image acquisition, is the most common technique. Inversion recovery sequences employ an inversion pulse and TI chosen to null the signal of fat. Inversion recovery sequences offer improved fat saturation in the setting of poor magnetic field homogeneity or magnetic susceptibility artifact, but at the cost of slightly longer acquisition times. A more recent technique is Dixon fat saturation, in which in phase and opposed phase images are acquired and then added or subtracted to obtain separate fat and water images (25,26). A third acquisition is often added to help account for magnetic field and RF inhomogeneity. Finally, rather than suppressing fat, some sequences employ water-selective excitation. MRCP MRCP techniques have largely relied on SSFSE sequences until very recently. The underlying principle of MRCP is imaging fluid in the biliary tree while suppressing background signal from nonfluid structures, generally accomplished by the use of heavily T2- weighted sequences. SSFSE sequences are in many ways ideal for this task: the long ETLs are well suited for long TE acquisitions, acquisition times are short, even with relatively high spatial resolution, and these sequences are widely available on virtually all clinical systems (27 32). Limitations of SSFSE techniques include image blurring induced by long ETLs, flow artifacts within the biliary tree that can occasionally simulate stones or masses, and problems with saturation of adjacent slices when sequential acquisitions are performed (interleaved acquisitions alleviate this problem, but tend to introduce greater slice misregistration from motion artifact). Recently, 3D MRCP techniques have been introduced and are becoming widely available. 3D fast recovery FSE sequences, for example, can be performed during suspended respiration (albeit with reduced spatial resolution) or as a respiratory-gated sequence. 3D SSFP sequences are another option. Both techniques offer advantages of volumetric acquisition with thin sections and high spatial resolution, excellent SNR, and reduced artifacts (Fig. 4). Limitations include motion artifacts from inadequate respiratory gating, relatively long breathheld acquisitions, and banding artifact near the diaphragm in SSFP acquisitions. Alternatives to heavily T2-weighted MRCP techniques include the use of contrast agents partially excreted via the biliary system in conjunction with fat-saturated 3D SPGR sequences (33 35). Contrast agents such as mangafodipir trisodium (Teslascan; Amersham Biosciences, GE Healthcare Technologies), gadobenate dimeglumine (Multihance; Bracco), and gadolinium-ethoxybenzyl-diethylene-triamine-pentaacetic acid (Gd-EOB- DTPA; Primovist, Schering-AG), exhibit varying degrees of biliary excretion (36), and offer some interesting advantages relative to traditional fluid-based techniques. Biliary function, for example, can be assessed more readily. Background suppression of ascites and bowel fluid is less problematic. Contrast excretion is more readily quantified in investigations of gallbladder and biliary function. Positive contrast agents are also useful for identifying potential biliary leaks following biliary surgery or transplantation (34). 3D fat-saturated SPGR sequences can also be performed after injection of standard extracellular gadolinium agents to generate negative contrast MRCP images: this can be accentuated by generating subvolume minimum intensity projection images (Fig. 4). Contrast Agents Gadolinium-based agents that rapidly equilibrate with the extravascular space following injection are by far the most common contrast agents employed in hepatic MRI. These agents have a proven track record of safety and efficacy, and are typically administered intravenously as a bolus at a rate of 2 3 ml/second, followed by acquisition of dynamic fat-saturated T1-weighted SPGR sequences to include arterial, portal venous, and equilibrium phase images. Dynamic imaging detects more lesions in comparison to unenhanced MRI, and also allows characterization of many benign and malignant lesions. Limitations of these agents have long been recognized, however. Dynamic information is crucial in many applications, and therefore data must be acquired rapidly in coordination with arrival of the contrast bolus. While this is usually straightforward, a small but significant percentage of patients have difficulty suspending respiration for the entire acquisition, resulting in varying degrees of motion artifact. Bolus timing is also occasionally problematic: arterial phase enhancing lesions can be missed when timing of the

6 686 Glockner Figure 4. 3D MRCP techniques. Maximum image projection (MIP) images from 3D FRFSE (a), 3D SSFP (b), and minimum intensity projection image from gadolinium-enhanced 3D SPGR (c) sequences in patient with Caroli disease and dilated intrahepatic ducts. Background suppression in the respiratory-triggered FRFSE sequence is superior to the breath-held SSFP sequence. Note that there is slight blurring of the FRFSE image, probably related to imperfect respiratory gating. Subvolume minimum intensity projection images from CE 3D SPGR sequences often yield excellent depiction of dilated ducts, although intraductal masses or other filling defects are poorly visualized. arterial phase acquisition is off by even a few seconds. Restricted temporal windows for data acquisition also place fundamental constraints on achievable SNR and spatial resolution: these limitations are most notable for detection of small lesions, which require both high spatial resolution and high SNR. Contrast agents allowing equilibrium phase imaging offer an alternative to dynamic acquisitions. SPIO agents, for example, are designed for equilibrium phase imaging. Breath-held T2*-weighted images are favored by most radiologists (37); however, respiratory-gated FSE images can also be obtained in patients with limited breathhold capacity, or in situations in which high spatial resolution and/or high SNR is desired (38). Mangafodipir trisodium (Teslascan) is another equilibrium agent, albeit with positive contrast, potentially allowing longer acquisitions, although 2D and 3D breath-held SPGR sequences are typically employed (Fig. 5). Dual action agents such as gadobenate dimeglumine and Gd-EOB-DTPA allow acquisition of both dynamic phases and hepatocyte-specific equilibrium phases. This flexibility offers some interesting advantages, and may also help to characterize lesions that are indeterminate on traditional dynamic imaging; contrast uptake in the equilibrium phase, for example, can be used to help distinguish focal nodular hyperplasia from other lesions. CONTROVERSIES Role of MRI MRI is often viewed as the most sensitive and specific technique for evaluation of the liver. While there is data in the literature supporting this contention, the evidence is by no means overwhelming, and not surprisingly there is relatively little information available on comparisons between state-of-the-art MRI and CT. The recent revolution in multidetector CT technology has allowed routine acquisition of high-resolution isotropic images with spatial resolution on the order of 0.6 to 1 mm and acquisition times considerably shorter than those typically achieved by MRI. Spatial resolution is not the only consideration, of course, but it is crucial for detecting very small lesions or observing subtle arterial and venous abnormalities. Most MRI acquisitions, with the exception of the dynamic CE series, are 2D sequences with relatively thick slices and significant interslice gaps. Even 3D SPGR sequences do not routinely approach the spatial resolution achievable with multidetector CT. On the other hand, lesion/liver contrast is higher for MRI, and the flexibility and range of pulse sequences available in MRI provide a significant advantage over CT. Multidetector CT radiation doses, particularly when high spatial resolution is required, can be quite signif-

7 Hepatic MRI: Concepts and Controversies 687 Figure 5. Hepatobiliary contrast agents. Fat-saturated 3D SPGR images from patient with right hepatic lobe focal nodular hyperplasia before (a) and after (b) administration of Mn-DPDP. Note homogeneous administration of contrast in the lesion, with the exception of the central scar, indicating the presence of hepatocytes. icant; this is an important consideration, particularly in radiation-sensitive patients as well as those likely to undergo multiple examinations. Hepatic Metastases A few recent investigations have compared MRI and CT for detection of hepatic metastatic disease. Bartolozzi et al (39), for example, compared unenhanced MRI, mangafodipir trisodium (Mn-DPDP)-enhanced MRI, and spiral CT in 44 patients with suspected colorectal metastases. Sensitivity of CT and unenhanced MRI was similar at 72%, while enhanced MRI had a sensitivity of 90%. This study, however, was limited by the lack of a standardized CT protocol. Kim et al (40) also investigated Mn-DPDP-enhanced MRI vs. CT in 69 patients with suspected colon cancer metastases, and found no significant difference in the sensitivity of four-row multidetector CT and MRI: both had sensitivities in the range of 90%. MRI was superior, however, in distinguishing benign from malignant lesions. This is an important consideration, since as many as 90% of hepatic lesions 1 cm in patients with a known primary malignancy are benign (41). Dromain et al (42) examined hepatic metastases from endocrine tumors in 64 patients, comparing somatostatin receptor scintigraphy, CT, and MRI. MRI had the highest sensitivity, at 95%, for detection of metastases, vs. 78% for CT and 49% for somatostatin receptor scintigraphy. It should be noted, however, that the CT protocol consisted of single-detector spiral CT with slice thickness of 5 mm. It is generally accepted that CE MRI increases sensitivity for detection of metastases; the question of which contrast agent to use is more difficult. Del Frate et al (43), for example, compared gadobenate dimeglumine and ferumoxide-enhanced MRI for detection of metastases in 20 patients, and found a significantly higher sensitivity with ferumoxide-enhanced MRI; they also noted that subcentimeter lesions were more successfully detected with ferumoxides. Kim et al (44), on the other hand, evaluated 23 patients with 59 hepatic metastases and found no difference in the performance of SPIO and gadobenate dimeglumine-enhanced MRI. Schneider et al (45) compared low dose gadobenate dimeglumine and standard dose gadopentetate dimeglumine (Gd-DTPA) in 43 patients with suspected hepatic malignant lesions (predominantly metastases), noting a tendency toward more accurate results with gadobenate dimeglumine. In summary, then, MRI often compares favorably, or performs as well as CT in detection of hepatic metastases (46 49). Whether this situation changes when high-resolution CT is performed with 64-row systems remains an open question. It does seem likely, however, that the significant advantage enjoyed by MRI in soft tissue contrast will likely remain important for many situations. Choosing the best MR contrast agent to use for a particular situation remains difficult. The recent introduction of hepatobiliary agents with the potential for both dynamic and hepatocyte specific imaging offers intriguing possibilities, and may improve overall sensitivity for lesion detection and characterization. In addition to the opportunity for lesion characterization, the hepatocyte phase allows respiratory-gated imaging without time constraints: these may occasionally be the only useful images acquired in patients with poor breath-holding ability, and also allow the acquisition of high-resolution images with excellent SNR, an opportunity that has probably not yet been fully exploited. Logistical problems should not be underestimated, however: gadobenate dimeglumine requires a delay of minutes for optimal hepatocyte-specific imaging, which makes routine inclusion of both dynamic and delayed imaging somewhat difficult. Gd-EOB- DTPA has a shorter delay time of 20 minutes, which may facilitate examinations with this agent. Ferumoxide agents also require slow infusion over 30 minutes, which can interfere with patient throughput in busy practices. In our practice, dynamic imaging with a gadolinium contrast agent is most often performed, with ferumoxides or delayed imaging following administration of a hepatobiliary agent reserved for difficult cases. Hepatocellular Carcinoma Screening cirrhotic patients for HCC is a subject of considerable interest and debate. While many experts advocate MRI as the best available technique, the evidence supporting this contention is mixed. Teefey et al (50), for example, evaluated 25 patients prior to hepatic

8 688 Glockner transplantation with sonography, CT, MRI, and positron emission tomography (PET), comparing the imaging results with explant pathology. Sonography was superior to CT and MRI on a patient by patient basis for two independent readers with sensitivities of 0.89 for sonography, 0.67 and 0.56 for CT, and 0.56 and 0.50 for MRI (PET sensitivity was 0). Similarly, an explant study published by Krinsky et al (51) evaluated 24 patients with MRI prior to transplantation, and detected 39 of 118 lesions for an overall sensitivity of 33%. MRI detected all lesions 2 cm in diameter, 50% of lesions 1 2 cm, 4% of lesions 1 cm, and only one out of nine patients with carcinomatosis. While many other authors have reported much better results (37,52 59), usually in studies where pathologic confirmation was obtained from biopsy of visualized lesions, explant pathology remains the gold standard for these studies. There are many potential explanations for the relatively poor performance of MRI. Patients with advanced cirrhosis have distorted anatomy and altered hepatic perfusion. Often ascites is present, which can generate significant ghosting artifacts in T2-weighted images. Breathhold capacity may be limited as well, so that dynamic CE sequences are frequently degraded by motion artifact. The low sensitivity for detection of small subcentimeter lesions is not surprising, given the relatively limited spatial resolution of MRI. Additional difficulties arise in distinguishing well-differentiated HCC from regenerative or dysplastic nodules. HCC classically demonstrates increased signal intensity on T2- weighted images relative to liver, variable T1 signal intensity, arterial phase enhancement, and hypoenhancement on equilibrium phase images. Regenerative nodules are rarely bright on T2-weighted images and do not usually demonstrate arterial phase enhancement, but there are exceptions Hussain et al (6), for example, recently reported that 52% of non-hcc nodules in cirrhotic livers were hyperintense on T2-weighted sequences and many HCCs show little increased T2 signal intensity and/or little arterial phase enhancement. Frequently, patients with cirrhosis have transient foci of enhancement on arterial phase images that cannot be visualized on any other pulse sequence (38,60) (Fig. 6). Jeong et al (61) evaluated 68 small arterial phaseenhancing nodules and found that only 13% of these were HCCs. Holland et al (62) correlated small lesions seen only during the arterial phase with explant pathology in 16 patients undergoing hepatic transplantation and found that only 7% were malignant. These have been attributed to variant arterial and venous anatomy, small arteriovenous shunts, and dysplastic nodules, but small HCCs can occasionally exhibit an identical imaging appearance. The dilemma, then, is whether to disregard these early arterial lesions without corresponding findings on other sequences: this improves the specificity of MRI, but probably at a cost of the already low sensitivity. Some authors have investigated the efficacy of multiple arterial-phase acquisitions. Ito et al (63), for example, acquired six arterial phase acquisitions, each lasting five seconds, with relatively low spatial resolution. Rapid central washout after early enhancement and coronal enhancement were highly specific findings of Figure 6. Cirrhotic liver with arterial-phase enhancing nodules. Respiratory-triggered FSE image (a) reveals cirrhotic liver with heterogeneous signal intensity but no discrete hyperintense nodules. Arterial phase 3D SPGR image (b) reveals numerous subcentimeter enhancing nodules, which could not be identified on portal venous (c) or delayed images. Explanted liver revealed no evidence of malignancy. small hypervascular HCCs, although these characteristic findings were seen in only 53% of HCCs. On the other hand, Mori et al (64) performed triple arterialphase dynamic imaging in 32 patients with 102 HCCs, and concluded that the middle arterial phase (acquired at 12.6 seconds after peak aortic enhancement) was optimal for detecting HCC, and showed diagnostic ac-

9 Hepatic MRI: Concepts and Controversies 689 curacy equivalent to that of the entire triple arterialphase acquisition. Close interval follow up of arterial-phase enhancing lesions remains a reasonable strategy (61), although small variations in bolus timing can sometimes have a significant effect on visualization of these lesions from exam to exam. The use of a test bolus or fluoroscopic triggering helps to select the correct scan delay for achieving optimal arterial phase images and reduces some of this variability. Another strategy for increasing sensitivity and specificity for detection of hepatomas is to improve the quality of the T2-weighted images. An arterial-phase hyperenhancing lesion in a cirrhotic liver with increased T2 signal intensity is usually HCC. T2-weighted sequences generally perform poorly, however, in detecting these small lesions, and in fact the utility of performing any T2-weighted sequences in evaluating cirrhotic patients has in fact been questioned. Hecht et al (65), for example, in a retrospective study of 18 patients undergoing hepatic transplantation, found that the addition of T2- weighted images to the interpretation of dynamic CE 3D SPGR sequences provided no benefit in sensitivity or specificity in detection of HCC. The use of SPIO contrast agents has been advocated to improve lesion conspicuity, and some success has been achieved with a dual contrast-agent approach incorporating both ferumoxides and gadolinium-based agents (58,59). A recent study by Qayyum et al (66) examined the benefit of adding dynamic gadolinium-enhanced imaging to the interpretation of ferumoxide-enhanced sequences in cirrhotic patients, and found no significant difference in sensitivity or specificity for detection of HCC, but did note that reader confidence in diagnosis of lesions increased. In our practice, we rely predominantly on 3D dynamic gadolinium CE images for detection of HCC in cirrhotic patients. Small lesions detected only on arterial phase images remain problematic; while most of these lesions are benign, a malignancy rate of 7% to 13% is quite significant, particularly given the rapid doubling time of most HCCs. At present, close interval follow-up of these lesions seems to be the best alternative. T2-weighted imaging is usually less important, but can be useful, particularly if high-resolution respiratory-gated fat-saturated FSE sequences are performed with relatively short echo trains. This is particularly true for the small percentage of patients in whom dynamic imaging is limited by poor breath-holding, poor bolus timing, or other factors. Double-contrast exams using ferumoxides and gadolinium agents, because of the added expense and imaging time, are reserved for problematic cases. MRCP There is little doubt that MRCP is an effective noninvasive alternative to ERCP. A recent meta-analysis, for example, concluded that MRCP had excellent overall sensitivity and specificity for demonstrating the level and presence of biliary obstruction, with slightly lower sensitivity for detecting stones or differentiating malignant from benign obstruction (67). Evaluation of nondilated biliary ducts, however, remains problematic. The recent emergence of CT cholangiography using biliary contrast agents has emphasized the importance of spatial resolution as well as contrast in detecting subtle anatomic and pathologic abnormalities. Schroeder et al (68), in evaluating 25 potential living liver donors found 17 biliary variants with CT cholangiography and only four with MRCP. Similarly, evaluation of patients with suspected primary sclerosing cholangitis can be difficult when there is no ductal dilatation. Silva et al (69) have suggested using intravenous morphine to improve ductal distention in MRCP, noting benefits in visualization of nondilated intrahepatic ducts. Ductal distention may be an important factor in visualizing subtle biliary abnormalities, particularly given the relatively limited spatial resolution of MRCP techniques. While the reliability of respiratory-gated 3D fast recovery fast spin echo (FRFSE) or FSE sequences is currently somewhat limited relative to 2D SSFSE sequences, when they are successful, the image quality is generally superior to SSFSE images. A flexible, multisequence approach to MRCP is probably the ideal one. RECENT AND FUTURE DEVELOPMENTS Parallel Imaging Parallel imaging comprises a set of techniques (sensitivity encoding [SENSE], simultaneous acquisition of spatial harmonics [SMASH], generalized autocalibrating partially parallel acquisitions [GRAPPA], array spatial sensitivity encoding technique [ASSET], etc.) in which half or fewer of the usual number of lines of k-space are sampled while simultaneously preserving spatial resolution and reducing the field of view, which ordinarily would result in severe aliasing artifacts. Aliased images (or corresponding data in k-space) can be unwrapped using algorithms based on a priori knowledge of the spatial sensitivity profile of the individual receiver elements of the phased array coil (70,71). The ability to reduce acquisition times by half or more has obvious implications for hepatic imaging, particularly dynamic, time-resolved sequences, or any sequences requiring relatively long breathholds. The reduced acquisition times can also be used to improve spatial resolution and/or spatial coverage. Limitations of parallel imaging include a substantial reduction in SNR, which can be problematic in inherently low SNR sequences. Reconstruction artifacts are also a common problem, to some extent alleviated by the k-spaced based methods such as SMASH and GRAPPA. Parallel imaging is already frequently employed in hepatic MRI (in the multiarterial phase techniques described above, for example, as well as in most of the dynamic SPGR images included in this article). A cautionary note has been sounded by Vogt et al (72), however, who compared 3D volumetric interpolated breathhold examination (VIBE) sequences performed without and with parallel imaging in 49 patients. Acquisition times were substantially reduced; however, image quality was rated lower in the parallel imaging sequences, and fewer lesions were detected in comparison to the unaccelerated images. The correct balance between the

10 690 Glockner Figure 7. 3T MRI in patient with focal nodular hyperplasia. Respiratory-triggered FSE images with high spatial resolution and excellent SNR (a and b) reveal a mildly hyperintense mass with a central scar. Arterial (c) and equilibrium (d) phase images from fat-saturated 3D SPGR sequence with 2-mm section thickness reveal intense arterial enhancement with rapid washout and enhancement of the central scar on the delayed image. conflicting requirements of temporal resolution, spatial resolution, SNR, and image quality has always been difficult to achieve, and will likely vary from patient to patient and also depend on the pathology under investigation. While parallel imaging has primarily been used to reduce acquisition times or improve spatial resolution without affecting fundamental image contrast, it is also possible to change sequence parameters to improve contrast that ordinarily would result in prohibitively long acquisitions (for example, increasing TR or flip angle in 3D SPGR sequences). Parallel imaging can be useful in T2-weighted sequences. The reduction of ETL in single-shot sequences results in less image blurring, albeit at a cost of reduced SNR. Similarly, phase error accumulation and consequent image artifact is reduced in echo planar sequences, including diffusion-weighted images. 2D and 3D T2-weighted sequences employed in MRCP enjoy similar benefits from parallel imaging techniques. 2D parallel imaging (performed along both phase-encoding directions) can be applied to 3D sequences; this offers additional reduction of acquisition times or improvement in spatial resolution, although the SNR loss is correspondingly higher. Parallel imaging has fundamental advantages and limitations, as with any other technique, and is perhaps best thought of as an additional degree of freedom in planning acquisitions (73); it can be used with great flexibility to adjust acquisition time, resolution, or image contrast. The possibilities in hepatobiliary MRI are just beginning to be explored. High-Field Imaging 3T clinical systems were initially introduced primarily for applications in neuroradiology; however, there has been great interest recently in extending these benefits to body MRI (Fig. 7). In theory, high-field systems offer a clear benefit in SNR, since this is directly proportional to field strength. In practice, however, a number of additional factors must be taken into consideration, including specific absorption rate (SAR) limitations, increased artifacts, and alterations in image contrast due to changes in T1 and T2 relaxation times. Modification of pulse sequences for 3T systems will in most cases involve more than a change in the center frequency: the nuances of pulse sequence development and modification for 3T and higher field systems are likely to prove challenging, and will likely require more effort than was initially envisioned. Nevertheless, early results are promising, demonstrating clearly improved SNR and image quality in selected applications (74). The combination of high-field systems with parallel imaging is particularly attractive: the reduction in SNR due to parallel imaging may be partially or completely alleviated by the higher field strength, thereby allowing improved temporal and/or spatial resolution without the significant penalties noted above at 1.5T. Likewise,

11 Hepatic MRI: Concepts and Controversies 691 Figure 8. Dual contrast evaluation of cholangiocarcinoma. Ferumoxides infused prior to the examination provide suppression of normal liver and improved lesion conspicuity on respiratory-triggered FSE (a) and gradient echo (b) T2 and T2*-weighted images. 2D SPGR image acquired 10 minutes after gadolinium contrast injection (c) reveals late enhancement in the hilar cholangiocarcinoma. Maximum image projection (MIP) image from respiratory-triggered 3D FRFSE MRCP (d) reveals peripheral biliary dilatation and central obstruction (Note incidental pancreatic head cyst). high SAR T2-weighted sequences such as SSFSE and SSFP sequences useful both for hepatic MRI and MRCP will benefit from reduced echo trains and reduced SAR at 3T provided by parallel imaging. The future role of hepatobiliary MRI at 3T is uncertain. The 3T platform could well become the standard for most MRI examinations, in much the same manner that1.5t systems replaced earlier low-field magnets. However, in an era of increasing scrutiny of medical costs by both government and private insurers, the additional expense of 3T systems may need to be justified by data demonstrating improvements not only in image quality but also in diagnostic performance and patient outcomes. Contrast Agents There are already several contrast agents available for hepatic MRI, including standard gadolinium-based extracellular agents, SPIO agents, and agents with varying degrees of hepatocyte uptake and excretion. All have well-documented benefits in comparison to noncontrast MRI, but they also have limitations. Currently, there is relatively little information available regarding direct comparison of different classes of contrast agents for specific applications. The clinical utility of these agents will also depend to some extent on the willingness of radiologists to tolerate their less than ideal logistics: will all patients who would benefit from hepatobiliary phase imaging, for example, return for this procedure, and how will this fit into a standard schedule? Ferumoxides also require relatively long infusions, and in addition lesion characterization can also be problematic, although some recent work has helped in classifying the appearance of benign and malignant lesions on T2*-weighted gradient echo sequences (75). Several authors advocate the use of both SPIO and gadolinium contrast agents for certain indications, such as cholangiocarcinoma (Fig. 8) and HCC, and have demonstrated improved sensitivity for lesion detection in some cases (58,59). A recent work by Aguirre et al (76) used a similar double contrast technique to identify patients with significant hepatic fibrosis with high accuracy. The recent introduction of hepatocyte-specific agents with biliary excretion also offers a positive contrast agent for MRCP. 3D SPGR sequences performed during the excretory phase allow visualization of the biliary tree with excellent SNR and good spatial resolution.

12 692 Glockner Figure 9. Multiecho gradient echo sequence in patient with cirrhosis and siderotic nodules. Note increasing visualization of small hepatic nodules as the TE is increased with each echo (TE 2.1, 7.0, 12.0, and 16.9 msec). These agents have clear advantages for certain indications, in particular suspected biliary leaks, but may also prove useful in assessing biliary function and in characterization of anatomy in patients with nondilated ducts who are potential candidates for living related donor transplantation. USPIOs, agents with smaller particle size than conventionally used for hepatic imaging, have recently been shown to greatly aid in discriminating benign lymph nodes from those invaded by malignant cells. Although primarily used thus far for pelvic malignancies, these agents may prove very useful in staging patients with hepatic malignancies and adenopathy, a frequent occurrence in patients with primary sclerosing cholangitis, cirrhosis, and other diffuse inflammatory disorders. Blood-pool contrast agents remain under consideration for U.S. Food and Drug Administration (FDA) approval. These agents, with larger molecules and much longer intravascular half-life periods compared to conventional extracellular agents, may have applications in evaluation of hepatic arteries and veins, particularly in situations in which high spatial resolution is called for. Additionally, qualitative assessment of perfusion, both in diffuse hepatic diseases as well as in hepatic malignancies, should be more straightforward with these agents. Novel Pulse Sequences New pulse sequences are introduced in nearly every issue of this and other journals, and many of these have implications for hepatobiliary imaging. SSFP sequences, for example, are not only valuable in cardiac imaging, but also are quite useful for evaluation of the hepatic vessels without the need for contrast as well as assessment of the biliary tree. Recent articles have evaluated echo planar sequences with low b-value diffusion-weighting. This technique seems to enhance the conspicuity of many solid lesions. Multiecho gradient echo sequences can be performed to generate a parametric map of R2* values, which have been shown to correlate well with the amount of intracellular iron (77). This technique offers the potential for qualitative or semiquantitative noninvasive monitoring of hepatic (and other organ) iron content in patients with hemochromatosis. This pulse sequence may also be a useful adjunct when imaging with SPIO or USPIO contrast agents: the optimal TE in such cases is not always clear, and this sequence offers a range of TEs to choose from for optimal lesion/liver contrast and image quality. Long TEs also greatly increase the conspicuity of regenerative nodules containing hemosiderin, and therefore aid in the detection of patients with cirrhosis (Fig. 9). Quantitative assessment of fat content in patients with hepatic steatosis is a subject of some clinical interest, and has been approached with a variety of methods, including Dixon techniques (78). Quantitative assessment of hepatic perfusion is somewhat difficult due to the dual vascular supply of the liver, but there is increasing interest in this topic: it may prove useful in assessing therapeutic response of antiangiogenic therapy for hepatic metastases, and might also be a means of assessing and quantifying diffuse hepatic disease (79,80). MR elastography, a technique in which the elastic properties of tissue are assessed using an external transducer to generate low frequency longitudinal mechanical waves, which are coupled to a motion sensitive pulse sequence, has recently become feasible in abdominal imaging. Huwart et al (81) studied 25 patients undergoing hepatic biopsy and found a good correlation between mean elasticity and the extent of fibrosis.