Gastrointestinal Imaging
Original Research
Differences in Liver Imaging and Reporting Data System Categorization Between MRI and CT
Audio Available | 
Share
OBJECTIVE. The purpose of this study is to determine whether focal liver observations are categorized differently by CT and MRI using the Liver Imaging and Reporting Data System (LI-RADS).
MATERIALS AND METHODS. We performed a retrospective review of 58 patients at risk for hepatocellular carcinoma who underwent liver protocol CT and MRI within 1 month of each other. Two readers assigned a LI-RADS category for all focal liver observations in consensus. A significant category upgrade was defined as a change from LI-RADS categories 1 and 2 or nonvisualization to LI-RADS categories 3–5, from LI-RADS category 3 to category 4 or 5, from LI-RADS category 4 to category 5, or from any category to LI-RADS category 5V. A significant downgrade was defined as a change from LI-RADS category 5 to categories 1–4, from LI-RADS category 4 to categories 1–3, or from LI-RADS category 3 to categories 1 or 2.
RESULTS. The LI-RADS category was different between CT and MRI for 77.2% (176/228) of observations. A significant upgrade occurred on MRI for 42.5% (97/228) of observations because of nonvisualization by CT (n = 78), capsule (n = 8), arterial hyperenhancement (n = 4), intratumoral fat (n = 2), larger size (n = 2), tumor in portal vein (n = 2), and wash-out (n = 1). Of these 97 upgraded observations, two were upgraded to LI-RADS category 5V, 15 were upgraded to category 5, and 13 were upgraded to category 4. A significant downgrade occurred on MRI for 8.8% (20/228) of observations because of marked T2 hyperintensity (n = 14), smaller size (n = 2), wedge shape (n = 2), and marked T2 hypointensity (n = 2).
CONCLUSION. LI-RADS categorization of focal liver observations is dependent on imaging modality. MRI results in both upgraded and downgraded categorization compared with CT in an important proportion of observations.
Keywords: CT, hepatocellular carcinoma, LI-RADS, MRI
Hepatocellular carcinoma (HCC) is the most common primary liver cancer and the third leading cause of cancer mortality throughout the world [1]. CT and MRI play a crucial role in the diagnosis of HCC. To standardize the interpretation and reporting of CT and MRI of HCC, the Liver Imaging Reporting and Data System (LI-RADS) was created. The goals of LI-RADS are to reduce variability in interpretation and improve communication with referring clinicians [2]. LI-RADS provides a comprehensive algorithm that categorizes liver observations by their likelihood of being HCC from definitely benign (LI-RADS category 1) to definitely HCC (LI-RADS category 5).
LI-RADS is designed to be used with both liver protocol CT and MRI. However, it is unclear whether the interpretation and reporting of liver observations using LI-RADS differ depending on the imaging modality. The major imaging criteria described for CT are related to the appearance of the observation on multiphasic contrast-enhanced imaging and its growth. For MRI examinations, additional pulse sequences such as T2-weighted imaging and DWI can be used to further modify the observation's category. The detection of enhancement features, such as arterial phase hyperenhancement, capsule appearance, and delayed phase washout appearance, may differ between CT and MRI. In some studies, MRI has been shown to have higher accuracy than CT in detecting HCC and other cirrhotic nodules [3–5]. Therefore, the choice of imaging modality could affect the LI-RADS categorization and potentially clinical management. The purpose of this study is to determine whether focal liver observations are categorized differently by CT and MRI using LI-RADS.
| Materials and Methods |   |
|---|
This HIPAA-compliant study was approved by the University of California, Davis institutional review board, and the requirement for informed consent was waived because of the study's retrospective nature. A search of our single-institution radiology database was performed to identify patients who had undergone both liver CT and MRI between July 2005 and January 2014. Patients were included if they were 18 years old or older, were at risk for developing HCC, and had undergone liver protocol CT and MRI with an extracellular contrast agent within 30 days of each other. Patients were excluded for the following reasons: one imaging study was before and the second study was after an intervention (radiofrequency ablation, transarterial chemoembolization, or surgery) for a liver lesion (n = 18), severe motion artifact rendered part of the dynamic phases nondiagnostic quality (n = 4, all on MRI), and innumerable liver observations (n = 1). The final study group comprised 58 patients (43 men; mean age, 57 years). The clinical indications of all examinations were determined by a review of the electronic medical record.
Examinations were performed on a variety of MDCT equipment, including 4-, 16-, and 64-MDCT scanners and 64- and 128-MDCT scanners. All scans were obtained using a fixed voltage of 120 kV on all phases. With the exception of the GE Healthcare 4-MDCT scanner, a variable tube current–time product was used for all scans using automated dose modulation. The pitch varied across the scanners. All images were reconstructed using a slice thickness and interval of 5.0 mm. All CT examinations included four phases extending from the diaphragmatic dome to the iliac crest or the lower margin of the right hepatic lobe: an unenhanced phase, followed by late hepatic arterial, portal venous, and equilibrium phases. None of the patients received oral contrast media. For the contrast-enhanced phases, a total of 125 mL of iohexol (Omnipaque 350, GE Healthcare) was injected IV using a power injector at a rate of 4 mL/s. Bolus-tracking software was used to trigger the hepatic arterial phase scans at 6–8 seconds after contrast enhancement of the upper abdominal aorta to an attenuation threshold of 150 HU. The portal venous and equilibrium phases were timed to start at 80 and 160 seconds after the start of the contrast injection, respectively.
All patients were instructed to fast for 4 hours before MRI examination. Images were acquired on a 1.5-T MRI scanner system (Signa, GE Healthcare) with a phased-array torso coil. The imaging protocol varied because examinations spanned an approximately 10-year period, but all examinations included the following unenhanced pulse sequences: coronal T2-weighted single-shot fast spin-echo (SSFSE) (FOV, 38 cm; slice thickness, 5 mm; spacing, 6 mm; matrix, 288 × 192; TR/TE, 1141/90 ms; flip angle [FA], 90°), axial T2-weighted SSFSE (FOV, 40 cm; slice thickness, 5 mm; spacing, 6 mm; matrix, 288 × 192; TR/TE, 900/90; FA, 90°), axial fat-saturated T2-weighted fast spin-echo (FOV, 40 cm; slice thickness, 5 mm; spacing, 6 mm; matrix, 288 × 192; TR/TE, 2200/90; echo-train length, 25; FA, 90°), and axial 2D in- and out-of-phase T1-weighted imaging (FOV, 40 cm; slice thickness, 5 mm; spacing, 6 mm; matrix, 288 × 160; TR/TE, 150/2.2–4.4; FA, 90°). Unenhanced and contrast-enhanced axial T1-weighted fat-saturated 3D spoiled gradient-echo (FOV, 40 cm; slice thickness, 5 mm; spacing, 2.5 mm; matrix, 288 × 192; TR/TE, 3.1/1.4; FA, 12°) sequences were also performed. Contrast-enhanced images were acquired during the late hepatic arterial, portal venous, and equilibrium phases (2–3 minutes after injection) after IV administration of 0.1 mmol/kg of gadodiamide at a rate of 2 mL/s. The late arterial phase was timed by the use of either a 2-mL test bolus or an automated triggering technique (SmartPrep, GE Healthcare) with the ROI placed in the aorta at the level of the celiac axis. DW images were acquired with b values of 50 and 500 s/mm2 in the axial plane (FOV, 40 cm, slice thickness, 6 mm; spacing, 7 mm; matrix, 80 × 128; TR/TE, 11,250/58) for 40 patients and were not available for the others.
Two fellowship-trained abdominal radiologists with 5 and 2 years of postfellowship experience in liver MRI reviewed the cases in consensus. Any discrepancies in interpretation were resolved by a third fellowship-trained abdominal radiologist with 12 years of experience in liver MRI, which occurred in two cases. The readers were blinded to the original radiology reports and any clinical information but used the LI-RADS document for reference during the review. There were two review sessions separated by 1 month. During the first viewing session, only one imaging study was reviewed for each patient: the CT study for one half of patients and the MRI study for the other half. For the second review session, the other imaging study was reviewed for each patient.
The reviewers identified all liver observations on both CT and MRI examinations and assigned a LI-RADS category to each using the LI-RADS version 2013.1 algorithm [2]. Each observation was recorded as iso-, hyper, or hypoenhancing for the unenhanced and each contrast-enhanced phase compared with background liver. Additional features, including size and the presence of a delayed enhancing capsule, were recorded. The presence or absence of all ancillary features, as defined by LI-RADS version 2013.1, was also recorded for CT and MRI, as applicable. Twenty-six patients had applicable prior imaging, and for those patients, comparison was made to determine whether threshold growth was present. Per LI-RADS version 2013.1, threshold growth is defined as a diameter increase of a mass by a minimum of 5 mm and a diameter increase of 50% or more in 6 months or less, or 100% or more in more than 6 months.
A significant upgrade of category was defined as a change from LI-RADS categories 1 and 2 or nonvisualization to categories 3–5, from LI-RADS category 3 to category 4 or 5, from LI-RADS category 4 to category 5, or from any category to LI-RADS category 5V. A significant downgrade was defined as a change from LI-RADS category 5 to categories 1–4, from LI-RADS category 4 to categories 1–3, or from LI-RADS category 3 to categories 1 and 2. The reviewers assigned a specific imaging diagnosis for all LI-RADS 1 or 2 observations.
One reviewer assessed the adequacy of the late arterial phase for every study as follows: too early was defined as arterial but no portal venous enhancement, adequate was defined as portal but no hepatic venous enhancement, and delayed was defined as both portal and hepatic venous enhancement.
| Results | |
|---|
The clinical histories of the patients are presented in Table 1. CT was performed before MRI for 44 patients, with a mean (± SD) time between studies of 14.8 ± 9.1 days, and MRI was performed before CT for 10 patients, with a mean time between studies of 14.5 ± 10.4 days. CT and MRI were performed on the same day for four patients. The indications for the 44 MRI scans performed after the CT were as follows: to characterize an observation seen on prior CT (n = 22), elevated α-fetoprotein level (n = 9), presurgical planning (n = 2), and unknown (aside from the clinical histories putting the patients at risk for HCC) (n = 11).
There were a total of 228 observations. The numbers of observations in each LI-RADS category for CT and MRI are presented in Table 2. The sizes of the observations based on LI-RADS category are presented in Table 3. The LI-RADS category was different between CT and MRI for 176 (77.2%) observations. MRI identified 111 observations not seen on CT, and CT identified five observations not seen on MRI. The category with the largest number of observations seen on MRI and not CT was LI-RADS category 3 (n = 64; mean diameter, 0.8 ± 0.3 cm). Fifty-eight of these 64 observations showed arterial phase hyperenhancement without washout and were isointense on all other image sets.
A significant upgrade was seen on MRI compared with CT for 42.5% (97/228) of observations (Table 4 and Figs. 1 and 2). Of these 97 upgraded observations, two were to LI-RADS category 5V, 15 were to category 5, and 13 were to category 4. For these 30 observations, CT was performed after MRI in four cases, with a mean interval between examinations of 17.5 ± 10.4 days. In the 26 cases for which CT was performed before MRI, the mean interval between examinations was 10 ± 8.5 days.
 View larger version (310K) | Fig. 1A —57-year-old man with hepatitis C, alcohol abuse, and cirrhosis. A, Observation (2.6 cm) is isoenhancing on arterial phase CT (A), but is noted to wash out on delayed-phase CT (arrow, B) without capsule. |
 View larger version (339K) | Fig. 1B —57-year-old man with hepatitis C, alcohol abuse, and cirrhosis. B, Observation (2.6 cm) is isoenhancing on arterial phase CT (A), but is noted to wash out on delayed-phase CT (arrow, B) without capsule. |
 View larger version (256K) | Fig. 1C —57-year-old man with hepatitis C, alcohol abuse, and cirrhosis. C, On MRI, observation (arrow) shows arterial hyperenhancement (C) and delayed washout and capsule (D). Lesion was Liver Imaging and Reporting Data System category 4 on CT and category 5 on MRI. |
 View larger version (252K) | Fig. 1D —57-year-old man with hepatitis C, alcohol abuse, and cirrhosis. D, On MRI, observation (arrow) shows arterial hyperenhancement (C) and delayed washout and capsule (D). Lesion was Liver Imaging and Reporting Data System category 4 on CT and category 5 on MRI. |
 View larger version (306K) | Fig. 2A —59-year-old woman with hepatitis C and cirrhosis. A, Arterially hyperenhancing observation (1.1 cm) is seen on arterial phase CT (arrow, A), but is not noted to washout or have capsule on delayed phase CT (B). |
 View larger version (298K) | Fig. 2B —59-year-old woman with hepatitis C and cirrhosis. B, Arterially hyperenhancing observation (1.1 cm) is seen on arterial phase CT (arrow, A), but is not noted to washout or have capsule on delayed phase CT (B). |
 View larger version (211K) | Fig. 2C —59-year-old woman with hepatitis C and cirrhosis. C, Observation shows arterial hyperenhancement on arterial phase MRI (arrow, C) as well as both washout and enhancing capsule on delayed phase MRI (arrow, D). Observation was new and therefore rated Liver Imaging and Reporting Data System category 4 on CT and as category 5 on MRI. |
 View larger version (307K) | Fig. 2D —59-year-old woman with hepatitis C and cirrhosis. D, Observation shows arterial hyperenhancement on arterial phase MRI (arrow, C) as well as both washout and enhancing capsule on delayed phase MRI (arrow, D). Observation was new and therefore rated Liver Imaging and Reporting Data System category 4 on CT and as category 5 on MRI. |
There were significant downgrades of observations by MRI compared with CT for 8.8% (20/228) of observations because of marked T2 hyperintensity (n = 14), smaller size on MRI compared with CT (n = 2), wedge shape (n = 2), and marked T2 hypointensity (n = 2). Most of these downgrades (95%; 19/20) were from LI-RADS category 3 on CT to categories 1 and 2 on MRI. One downgrade was from LI-RADS category 4 to category 3 because of marked T2 hypointensity. The 14 observations downgraded because of marked T2 hyperintensity were diagnosed as hemangiomas (Fig. 3).
 View larger version (334K) | Fig. 3A —60-year-old man with hepatitis C and cirrhosis. A, Arterial phase CT reveals 0.7-cm hyperenhancing observation (arrow, A) that is isoenhancing on delayed phase CT (B). |
 View larger version (363K) | Fig. 3B —60-year-old man with hepatitis C and cirrhosis. B, Arterial phase CT reveals 0.7-cm hyperenhancing observation (arrow, A) that is isoenhancing on delayed phase CT (B). |
 View larger version (230K) | Fig. 3C —60-year-old man with hepatitis C and cirrhosis. C, MRI shows arterial hyperenhancement (arrow, C) as well as persistent hyperenhancement that follows blood pool on delayed phase (arrow, D). |
 View larger version (279K) | Fig. 3D —60-year-old man with hepatitis C and cirrhosis. D, MRI shows arterial hyperenhancement (arrow, C) as well as persistent hyperenhancement that follows blood pool on delayed phase (arrow, D). |
 View larger version (213K) | Fig. 3E —60-year-old man with hepatitis C and cirrhosis. E, Observation (arrow) is markedly hyperintense on T2-weighted image and is consistent with hemangioma. Observation was categorized as Liver Imaging and Reporting Data System category 3 by CT and as category 1 by MRI. |
The arterial phase timing was adequate on MRI and CT for 43 patients. It was early on both MRI and CT for six patients. It was early on CT and adequate on MRI for four patients and early on MRI and adequate on CT for five patients. Of the cases where the CT was early and MRI was adequate, there were two significant upgrades due to arterial hyperenhancement visualization on MRI but not CT, both to LI-RADS category 3.
| Discussion | |
|---|
Our study shows that an important proportion of liver observations are categorized differently by CT and MRI using LI-RADS. The choice of imaging modality may affect the management of patients who are at risk for HCC.
The most important finding of our study was that nearly half (42%) of observations were significantly upgraded on MRI compared with CT. Although most were upgraded to LI-RADS category 3, approximately one third of upgrades were to category 4, 5, or 5V. The most common reason for the upgrade to these categories was nonvisualization of the observation on CT. For observations seen on both CT and MRI, the most common reasons for the upgrade by MRI were the visualization of arterial hyperenhancement or a delayed enhancing capsule not seen on CT. We think that this is due to the superior contrast enhancement of gadolinium-enhanced liver MRI compared with CT. Our results are in concordance with those of a prior study by Hayashida et al. [6] that showed improved visualization of arterial hyperenhancement of small HCCs with multiphasic MRI compared with CT (80% on MRI vs 67% on CT). Older studies by Tomemori et al. [7] and Yamashita et al. [8] also showed improved detection of HCC with arterial phase MRI compared with CT. A prior report by Honda et al. [9] also showed greater accuracy in detecting a capsule with MRI compared with CT (92% vs 82%). This is important because visualization of these two features is highly likely to upgrade an observation's LI-RADS category because they are considered classic features of HCC. LI-RADS category 5 observations are considered to be HCC with nearly 100% certainty, and category 4 observations are likely HCC such that they may be treated as HCC, biopsied, or closely followed depending on the clinical situation. Therefore, improved detection of these imaging features with MRI may lead to earlier diagnosis and treatment of early HCCs.
Only one observation in our study was upgraded because of improved visualization of delayed washout, and none was upgraded to LI-RADS category 4, 5, or 5V on this basis. This finding is similar to the results of the study by Hayashida et al. [6], which found improved visualization of washout with CT compared with MRI. It is possible that the difference in contrast media volume between CT and MRI played a role in improved visualization of arterial enhancement on MRI but did not affect the detection of washout. The relatively compact bolus of gadolinium used for MRI allows better isolation of the arterial phase in relation to the imaging time, as well as greater peak enhancement in that phase by arterial supply–dominant observations. In contradistinction, the larger volume of contrast media used for CT causes peak enhancement of arterial supply–dominant observations to occur later, in the portal venous phase [10]. This factor would not be expected to affect delayed washout assessment and may explain why MRI showed little advantage compared with CT for this feature.
MRI can also reveal more ancillary features than CT (e.g., mild-to-moderate T2 hyperintensity and restricted diffusion), and this could theoretically lead to higher categorization of observations by MRI. However, only two observations were upgraded to LI-RADS category 4 solely on the basis of ancillary features seen only on MRI; both showed intratumoral fat. Therefore, we suggest that these features are not major drivers of changes in categorization of observations between MRI and CT. There are already inherent limitations on the use of ancillary features because they cannot be used to upgrade to category 5 according to LI-RADS. In addition, the precise effect of ancillary features is difficult to judge because there are not strict guidelines as to when ancillary features should be used to upgrade an observation.
The most common reason for different LI-RADS categorization in our study was that an observation was seen on MRI but not CT, with approximately one half of the discrepancies due to this finding. LI-RADS category 3 observations accounted for the largest number of observations seen on MRI but not CT. The reason these lesions are visualized on MRI and not CT is again likely related to both the superior image contrast and advantages in arterial phase imaging for MRI. A prior study by Pitton et al. [4] reported increased tumor nodule detection with MRI compared with CT, with the highest sensitivity for arterial phase MRI. The precise features of those nodules were not described in detail in that study, but most of the nodules seen with MRI and missed by CT were small (< 15 mm) and may have been similar to our LI-RADS category 3 lesions. We hypothesize that many of the LI-RADS category 3 lesions are transient hepatic intensity differences or pseudolesions. These lesions are related to arterioportal shunts frequently seen in cirrhotic livers. This is supported by the fact that most of these observations showed arterial phase hyperenhancement without washout. Our findings are consistent with those of a recent report from Choi et al. [11], which found that the most common cause of LI-RADS category 3 observations on MRI was hypervascular pseudolesions. That study also reported that most of the indeterminate observations seen on MRI were stable at follow-up imaging. Holland et al. [12] also found that most (93%) small (< 2 cm) arterially enhancing lesions seen on MRI that were occult on other phases and pulse sequences were nonneoplastic, even in patients with proven HCC elsewhere in the liver. Kim et al. [13] found that, although gadobenate dimeglumine–enhanced MRI had higher sensitivity for HCC than did CT, increased false-positive findings were seen with MRI. The authors suggested that the false-positives were likely arterioportal shunts. This increased detection of indeterminate observations by MRI has the potential to lead to increased follow-up of benign lesions that would not have been visualized with CT. This is a potential disadvantage of MRI compared with CT. The distinction between LI-RADS category 3 and categories 1 and 2 observations is relevant, however, because a small but important proportion of LI-RADS category 3 observations will progress to HCC and therefore should be followed by CT or MRI, whereas LI-RADS categories 1 and 2 observations require no follow-up [11].
It is important to note that a downgrade in category was observed with MRI for approximately 9% of observations. Most of these cases were downgraded because of marked T2 hyperintensity, which is an ancillary feature that favors benignity, and allowed the diagnosis of hemangioma to be made with confidence in those cases. Although historically thought to be less commonly than in noncirrhotic livers, hemangiomas do occur in cirrhotic livers, and a recent study suggested that the prevalence of hemangiomas seen on MRI is the same in cirrhotic and noncirrhotic livers [14–17]. Hemangiomas in patients with diffuse liver disease are also reported to have an appearance virtually identical to those in the normal liver, and the availability of T2-weighted imaging in MRI is a key advantage over CT in making this diagnosis [18]. These lesions were categorized as indeterminate (LI-RADS category 3) on CT, whereas a confident benign categorization (LI-RADS category 1) was possible on MRI.
The different categorization of liver observations by CT and MRI has important implications. According to the Organ Procurement and Transplant Network guidelines, patients with chronic liver disease and HCC meeting certain criteria receive automatic priority status for liver transplantation [19, 20]. The diagnosis of HCC can be made by imaging studies alone without biopsy. The Organ Procurement and Transplant Network imaging criteria for the diagnosis of HCC (class 5) are nearly identical to the LI-RADS category 5 classification. However, as with LI-RADS, CT and MRI are both considered acceptable and no distinction is made between the two modalities. The results of our study show that the same patient could be diagnosed with HCC (Organ Procurement and Transplant Network class 5 or LI-RADS 5) by MRI but not by CT, drastically affecting the chances of liver transplantation.
Our study has limitations. First, more MRI examinations were performed after CT, which could potentially introduce bias toward more progressed lesions being visualized on MRI. However, the mean time interval between CT and MRI in the cases where MRI findings led to an upgrade of an observation to LI-RADS category 4, 5, or 5V was only 10 days, and it is unlikely that the lesions changed significantly during that period. Many of the MRI scans were performed to further evaluate an observation seen on CT, which could also introduce bias. We did not have pathologic proof or follow-up of the lesions in this study. However, the focus of the study was to compare the two imaging modalities with respect to categorizing observations under the LI-RADS criteria, not the underlying pathologic abnormality. The LI-RADS categorization was performed in consensus and not by independent reviewers, and interreader agreement was not assessed. The MRI examinations in this study were performed with an extracellular gadolinium contrast agent, so we could not assess the effect of using a hepatobiliary agent. We also used the now-prior version of LI-RADS (version 2013) because this was the most recent version available at the time of image review. Given the relatively long study period, the image quality of the MRI examinations was likely variable and improved with later cases. A minority of studies showed suboptimal arterial phase timing; however, the number of suboptimal cases was similar between CT and MRI and did not lead to an important number of significant upgrades.
In conclusion, the LI-RADS classification of focal liver observations is dependent on the imaging modality. More observations are visualized with MRI, and an important proportion of observations are significantly upgraded or downgraded into categories that would alter patient management.
| References | |
|---|
