| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 All authors: Department of Radiology/MRI B2B311, University of Michigan Health System, 1500 E. Medical Center Dr., Ann Arbor, MI 48109-0030.
Received June 19, 2002;
accepted after revision August 6, 2002.
Address correspondence to G. C. Mueller.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
2 cm) hepatic lesions made in a
routine clinical setting with the effectiveness of such characterization made
under standardized conditions by radiologists who are expert interpreters of
MR imaging.
MATERIALS AND METHODS. Forty-eight patients with 69 small (2 cm)
hepatic lesions considered indeterminate on a prior routine CT scan were
included in the study. The diagnosis for all lesions had been verified by
histology (n = 10), surgery and intraoperative sonography (n
= 5), imaging follow-up (n = 35), or clinical follow-up (n =
19). Using the initial radiology reports, the diagnoses based on MR imaging
were rated on a 5-point confidence scale. In addition, two radiologists
experienced in MR imaging who were unaware of the initial interpretations of
the images or the clinical histories of the patients independently analyzed
the MR imaging studies and characterized the lesions using the same 5-point
scale. The observer performance for the initial MR imaging interpretations and
the expert interpretations were measured using receiver operating
characteristic analysis. Interobserver agreement was determined with weighted
kappa statistics.
RESULTS. Fifty-eight lesions were benign (six cysts, 22 hemangiomas,
four regenerating nodules, two steatohepatitic lesions, one atypical blood
vessel, three focal fat and five focal fat-sparing lesions, 13 flow-related
pseudolesions, one diaphragmatic insertion, and one unspecified lesion), and
11 lesions were malignant (nine metastases and two hepatocellular carcinoma).
The areas under the curve were 0.94 (initial reports), 0.88 (observer 1), and
0.84 (observer 2). Substantial agreement was found between the expert
interpreters (
= 0.74), and moderate agreement, between the expert
interpreters and initial interpreters ( = 0.44 each).
CONCLUSION. MR imaging is an effective method of characterizing
small (2 cm) hepatic lesions in routine clinical practice.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The aim of our study was to compare the effectiveness of MR imaging
characterization of small ( 2 cm) hepatic lesions made in a routine
clinical setting with the effectiveness of such characterization made under
standardized conditions by radiologists who are expert interpreters of MR
imaging. Although our study did not compare MR imaging and CT, we chose
lesions that were considered indeterminate on previous CT scans because
patients with such lesions are commonly encountered in clinical practice and
are often referred for MR imaging for further characterization of the
lesions.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Forty-eight patients with 69 hepatic lesions were eligible. The size of the lesions ranged from 0.2 to 2.0 cm (mean, 1.14 cm); 18 lesions in 14 patients were smaller than 1 cm. The study population included 19 women and 29 men (age range, 30-83 years; mean, 55 years). Twenty-seven patients had a history of cancer (two of the patients had two malignancies each): Eight had a history of colon cancer; three, leiomyosarcoma; two, clear cell carcinoma of the kidney; two, esophageal cancer; one, head and neck cancer; one, gastrinoma; one, squamous cell carcinoma of the skin; three, pancreatic cancer; two, breast cancer; one, lung cancer; one, ovarian cancer; and four, hepatocellular carcinoma. Twelve patients had history of cirrhosis or a history of hepatitis B or C. One patient had sclerosing cholangitis. Two patients had undergone liver transplantation: one for cryptogenic cirrhosis and hepatocellular carcinoma, and the other for hepatitis C and hepatocellular carcinoma.
MR Imaging
MR imaging was performed on a 1.5-T unit (Signa; General Electric Medical
Systems, Milwaukee, WI) with high-performance gradients (maximum gradient
strength, 25 mT/m; rise time, 600 msec) and a phased array torso coil. The
sequences performed were as follows: a coronal heavily T2-weighted single-shot
fast spin-echo localizer sequence (TR/TE, infinite/180; slice thickness, 8 mm
with no interslice gap; matrix, 256 x 128 [frequency x phase];
breath-hold, 20-28 sec); an axial T1-weighted spoiled gradient-echo at in- and
out-of-phase TE sequences (150/2.1 [out-of phase]; 150/4.2 [in-phase]; slice
thickness, 6-8 mm with no interslice gap; matrix, 256 x 160;
breath-hold, 18-24 sec); and an axial T2-weighted fast spin-echo sequence with
fat suppression and respiratory triggering (>3000/96; slice thickness, 6-8
mm with no interslice gap; matrix, 256 x 224; acquisition time, 2-4
min).
The axial T1-weighted dynamic gadolinium-enhanced imaging performed in the early study period (studies performed between November 1998 and December 1999) was a two-dimensional (n = 27) spoiled gradient-echo sequence with fat suppression (<200/1.2; slice thickness; 6-8 mm with no interslice gap; matrix, 512 x 160; breath-hold, 24-28 sec). The axial T1-weighted dynamic gadolinium-enhanced imaging performed in the later period (studies performed between January 2000 and August 2000) was a three-dimensional (n = 21) axial spoiled gradient-echo sequence with fat suppression (TR range/TE range, 4-6/1-2; flip angle, 12°; section thickness, 3-4 mm with zero interpolation yielding an effective section thickness of 1.5-2 mm; matrix, 320 x 160; breath-hold, 24- 28 sec).
The gadolinium-enhanced imaging was performed in the arterial dominant, portal venous, and 2- and 5-min delayed phases of enhancement. Gadolinium was administered by a 20- to 22-gauge needle placed in the antecubital fossa and attached to an MR-compatible power injector (Spectris; Medrad, Pittsburgh, PA). All patients except one (who refused the gadolinium) received a bolus injection of 0.1 mmol/kg of gadopentetate dimeglumine (Magnevist; Berlex Laboratories, Wayne, NJ) at a rate of 2 mL/sec followed by 15 mL/sec of saline flush at the same rate. In the earlier studies (performed between November 1998 and December 1999), the arterial phase imaging was performed with a fixed delay of 15 sec after gadolinium injection. In the later studies (performed between January 2000 and August 2000), timing of the arterial phase imaging was selected using automated contrast-bolus detection (Smartprep; General Electric Medical Systems).
Verification of Diagnoses
The malignancy of 11 lesions was verified by histology (n = 3),
intraoperative sonography and biopsy of another lesion with a similar
appearance (n = 1), intraoperative inspection and biopsy of regional
lymph nodes (n = 1), and serial cross-sectional imaging (n =
6). Serial cross-sectional imaging was performed with CT (n = 4), a
combination of CT and MR imaging (n = 1), or a combination of CT and
sonography (n = 1). Criteria for the diagnosis of malignancy by
serial cross-sectional imaging were an increase in size and number of lesions.
The cross-sectional imaging follow-up period of malignant lesions ranged from
10 to 94 weeks (mean, 69 weeks).
The benignity of 58 lesions was verified by histology (n = 7), intraoperative sonography with additional CT follow-up (n = 3), serial cross-sectional imaging (n = 29), or clinical follow-up (n = 19). Intraoperative sonography did not include additional biopsy. Serial cross-sectional imaging was performed with CT (n = 13), MR imaging (n = 7), sonography (n = 6), a combination of CT and MR imaging (n = 2), or a combination of CT and sonography (n = 1). Criteria for the diagnosis of benignity by serial cross-sectional imaging were stability in both the number and the size of lesions. The serial cross-sectional imaging follow-up period of benign lesions ranged from 28 to 174 weeks (mean, 93 weeks). Clinical follow-up was defined as close medical observation of the patient for any clinical or laboratory signs of malignant disease; the lack of any of those signs was accepted as proof that the lesion was benign. The clinical follow-up period was 40-166 weeks (mean, 94 weeks).
Data Analysis
The initial MR imaging reports generated for each patient were reviewed.
For each eligible lesion, diagnostic confidence was rated using a 5-point
scale: 1, definitely benign; 2, probably benign; 3, equivocal; 4, probably
malignant; and 5, definitely malignant. Lesions not detected on MR examination
were given a rating of 1.
Reproducibility was tested by having two experienced MR radiologists independently assess the same lesions using the same 5-point scale under standardized conditions (i.e., unaware of the initial interpretations, patient history, or clinical records). The expert interpreters were fellowship-trained in body MR imaging and had 4-5 years of experience in MR imaging of the liver. Both expert interpreters worked in our department; they had rendered the initial MR imaging reports for 24 of the 69 lesions (observer 1, four lesions and observer 2, 20 lesions). The time interval between the rendering of the initial reports and the expert interpretation for those lesions was 15-36 months (mean, 23 months). The complete MR imaging data set was made available to each interpreter either on film or on a commercially available computer workstation (Advantage Windows, General Electric Medical Systems), and the lesions to be characterized were labeled on the CT images.
Statistical Analysis
The obtained data were assessed using receiver operating characterristic
(ROC) curves analysis [25,
26,
27,
28] (JROCFIT software; Eng J,
Johns Hopkins University, Baltimore, MD). Observer performance was measured by
calculating the areas under the ROC curves. The observer performance of the
initial interpretations and the observer performance of the expert
interpretations were evaluated by comparing the resulting areas under the
curve.
To evaluate the effect of equivocal ratings on the sensitivity and specificity of MR imaging, we separated the data using two different thresholds for assigning lesion malignancy and subsequently calculated cumulative sensitivities and specificities and the corresponding positive and negative predictive values. Threshold 1 defined ratings of 3-5 as positive diagnoses for malignancy, and threshold 2 defined only ratings of 4 and 5 as positive diagnoses.
To assess whether a history of cancer or chronic liver disease correlated with the frequency of occurrence of malignant lesions, we analyzed the data separately for two groups of patients, applying the two previously described thresholds. Group A consisted of patients with a history of cancer, cirrhosis, or sclerosing cholangitis (37 patients with 49 lesions); group B were those patients without a history of such diseases (11 patients with 20 lesions). We also performed descriptive statistical analysis on lesions smaller than 1 cm. Agreement between the two expert interpreters and between initial interpretations and expert interpretations was determined using weighted kappa statistics [29].
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Observer Performance
For all lesions, observer performance of the initial interpretations and
expert interpretations yielded the following areas under the curve: 0.94 for
the initial reports, 0.88 for observer 1, and 0.84 for observer 2
(Fig. 1). ROC analysis was
repeated for group A but not group B because the latter group had no malignant
lesions. For group A, the area under the curve was 0.94 for the initial
interpretations, 0.91 for observer 1, and 0.92 for observer 2. The differences
between the performance of initial interpreters and the performance of each of
the expert interpreters were statistically not significant (p >
0.05).
|
Table 2 provides cumulative sensitivities and specificities for two different thresholds for assigning lesion malignancy, and Table 3 shows the corresponding positive and negative predictive values. As expected, with a shift from threshold 1 to threshold 2, the cumulative specificities increased, and the cumulative sensitivities decreased (Table 2). For the overall population and for group A, the best compromise between sensitivity and specificity was obtained by applying threshold 1 and regarding equivocal ratings as positive diagnoses. Given the low prevalence of malignant lesions even in group A, the positive predictive values were low, but the negative predictive values were high when threshold 1 was applied (Table 3).
|
|
Because there were no malignant outcomes in group B, sensitivities and predictive values could not be calculated. For group B, the highest specificities were obtained by applying threshold 2. No lesion in group B had a rating higher than 3; thus, the cumulative specificities increased to 100% for all observers when threshold 2 was applied.
Eighteen lesions in 14 patients were smaller than 1 cm: four were cysts; six, flow-related pseudolesions; four, hemangiomas; three, metastases; and one, regenerating nodule. For these subcentimeter lesions, the sensitivities, specificities, and positive and negative predictive values were as follows: If threshold 1 was applied, the initial interpretations obtained a sensitivity of 66.7%, a specificity of 93.3%, a positive predictive value of 66.7%, and a negative predictive value of 93.3%; the expert interpretations obtained a sensitivity of 66.7% ± 0.0% (mean ± SD), a specificity of 86.7% ± 9.4%, a positive predictive value of 53.3% ± 18.9%, and a negative predictive value of 92.8% ± 0.7%.
With threshold 2, the initial interpretations obtained a sensitivity of 33.3%, a specificity of 100.0%, a positive predictive value of 100.0%, and a negative predictive value of 88.2%; the expert interpretations obtained a sensitivity of 50.0% ± 23.6%, a specificity of 100.0% ± 0.0%, a positive predictive value of 100.0% ± 0.0%, and a negative predictive value of 91.0% ± 3.9%.
Interobserver Agreement
Interpretation of the kappa values was based on the method described by
Landis and Koch [30].
Substantial agreement was found between the expert interpreters ( =
0.74), and moderate agreement was found between the expert interpretations and
the initial interpretations (observer 1: = 0.44; observer 2: =
0.44).
MR Imaging Characteristics
Of the 69 lesions, 58 (84.1%) were benign
(Table 1). Each lesion had
three ratings, one from the initial interpretations and one from each expert
observer. Therefore, we tabulated the lesions according to the highest (or
most suspicious) rating received. The most common benign lesions were
hemangiomas (n = 22) and flow-related pseudolesions (n = 13)
(Fig. 2A,
2B,
2C). Two hemangiomas were rated
malignant by at least one of the expert observers. One hemangioma was 1.7 cm
and was detected in a patient with cirrhosis and a history of hepatocellular
carcinoma. This lesion was shown to be a sclerotic hemangioma at explantation
and displayed an atypical imaging appearance with central hyperintensity on
T1-weighted imaging and a corresponding hypointensity on all the other
sequences. The second hemangioma was a 1-cm lesion in a patient with hepatitis
C and an elevated 
-fetoprotein level. This lesion showed a moderately
high signal on T2-weighted imaging, had a single peripheral puddle on the
arterial phase imaging, and remained unchanged on the portal venous phase and
delayed phase imaging. This lesion appeared unchanged on sonography at the
32-month follow-up examination.
|
|
|
Two regenerating nodules in one patient were given malignant ratings by at least one observer. One of these lesions is shown in Figure 3A, 3B, 3C, 3D. The second lesion was isointense relative to the liver on the T1- and T2-weighted images and on the arterial phase images and became hypointense on the 2-min delayed phase images. The patient underwent hepatic transplantation, and no dysplasia or malignancy was found in the explanted liver.
|
|
|
|
As shown in Table 1, 11 (15.9%) of 69 lesions were malignant. These lesions were tabulated according to the lowest (or least suspicious) rating received. Figure 4A, 4B, 4C shows a 1-cm metastasis from colon cancer that was correctly assigned by all observers. A 9-mm metastasis from a pancreatic cancer that was revealed on CT was not detected by any observer on the MR images. Two months after undergoing MR imaging, the patient underwent follow-up CT that showed that the lesion had doubled in size and had developed ill-defined margins and that additional liver lesions had also developed. After data collection, both expert observers reviewed the MR images with the knowledge of the history of pancreatic cancer and agreed that the lesion was not visible even in retrospect. A 1.4-cm metastasis of gastrinoma was regarded as a probable hemangioma by one observer. Later review of the MR images by the expert radiologists with knowledge of the patient's history found that the imaging features of this lesion were not characteristic of either a metastasis or a hemangioma (Fig. 5A, 5B, 5C, 5D).
|
|
|
|
|
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The lesion types encountered in our study were different from those described in previous studies, particularly studies that focused mainly on distinguishing cysts and hemangiomas from malignancies [6, 7, 8, 10, 11, 17, 18, 19, 20]. Although we found a relatively large number of hemangiomas, we encountered a relatively high number of pseudolesions but only a few cysts. The increased frequency of pseudolesions in our population is not surprising because pseudolesions often resemble malignancies on CT [31] and, in our population, pseudolesions were more frequently judged to be indeterminate on CT. Furthermore, even small hepatic cysts are usually easier to characterize on routine CT.
Only 10 of 22 hemangiomas in our study received benign ratings by all observers. The small size of the lesion may have contributed to this finding, as has been suggested in previous reports [10, 17]. Furthermore, the selected lesions were indeterminate on contrast-enhanced CT, and we speculate that gadolinium enhancement patterns were also not specific. Thus, the observers may have been unable to definitively characterize these lesions on the T2-weighted MR images alone.
The seemingly low numbers of malignant lesions in our patient population (15.9% of 69 lesions in all patients; 22.4% of 49 lesions found in patients with known malignancy or chronic liver disease) is in accordance with the literature [32, 33]. Jones et al. [32] found 55 malignant lesions (22%) in 254 patients who had masses measuring 1.5 cm or smaller detected on abdominal CT. All 55 malignant lesions occurred in patients with history of cancer (209/254). No patient without underlying malignancy had a malignant lesion. Schwartz et al. [33] reported finding 378 patients with lesions smaller than 1 cm or too small to characterize on CT among 2978 patients with cancer referred for CT. The rate of malignancy in these patients was 11.6%. Because we focused on CT-indeterminate lesions and included only those with verification of the lesion type, malignancy might be overestimated—rather than underestimated—in our study population. Furthermore, our findings support those of Jones et al. [32] and suggest that for small hepatic lesions detected incidentally on CT in patients without a history of cancer or chronic liver disease, the probability of malignancy is extremely low. Decisions concerning further diagnostic procedures in such patients should be made judiciously.
Although differences between the ROC curves for the initial radiology reports and the expert interpreters in our study were not significant, the initial interpreters tended to have better performance. The initial reports were not based on imaging characteristics alone and may have been influenced by knowledge of patient history and clinical information. The studies by Hamm et al. [9] and Tello et al. [14] indicate the importance of patient history for the assessment of hepatic lesions. Hamm et al. evaluated the diagnostic accuracy of unenhanced and gadolinium-enhanced MR imaging for the characterization of hepatic lesions and found that knowledge of the clinical data improved the performance of experts interpreters in differentiating benign from malignant lesions. Tello et al. found that a known patient history of malignancy significantly increases the odds of a given hepatic lesion being malignant. Our finding that the initial interpreters tended to have higher effectiveness than the blinded expert interpreters supports the results of these studies.
Because clinical decision making is influenced by the certainty of radiologic interpretations, lesions with equivocal ratings are particularly problematic for the referring physician [34]. In patients with a history of cancer or chronic liver disease or with an unknown clinical history, equivocal lesions should be regarded as probably malignant to obtain the best compromise between sensitivity and specificity. Given the relatively low positive and extremely high negative predictive values obtained by applying threshold 1, we found that MR imaging was more useful for excluding the possibility of malignancy than for establishing proof of malignancy in this clinical setting.
In patients with no history of cancer or chronic liver disease, the probability of malignancy is extremely low. In this clinical setting, patients with equivocal lesions may benefit from less aggressive clinical management.
Although the expert interpreters had to make their decisions by assessing combinations of imaging criteria that were not clearly defined, agreement between the two experts was substantial, indicating the reproducibility of their diagnoses. The moderate agreement between initial interpretations and the expert interpretations is best explained by the initial interpreters' knowledge of the patient history and clinical data.
There are recognized limitations to our study. Inclusion of more than one lesion in some patients might have introduced a bias resulting in better observer performance. On the other hand, the high prevalence of benign liver lesions, often coexisting in the same patients with malignancies, may have introduced a bias resulting in worse observer performance. For some patients with benign lesions, only clinical follow-up was available. We decided to include these patients in our population because our study was designed as an audit of our clinical work. Many of our patients, especially those with nonmalignant findings, receive clinical follow-up, and exclusion of those patients would have caused selection bias that would have resulted in a higher frequency of malignant outcomes. For those patients with clinical follow-up, the lack of any clinical or laboratory signs of malignant disease over a sufficiently long period of time allowed the clinicians to make patient management decisions.
Assignment of a rating of 1 (definitely benign) to lesions not detected on MR imaging might have caused overestimation of the area under the curve. On the other hand, the effect of the diagnosis "no lesion detected" on patient management is similar to the effect of the diagnosis "definitely benign." Therefore, our categorization of these lesions approximated clinical practice. Finally, the interpreters might have been influenced by the presence of additional lesions, either in the liver or in other abdominal organs.
In conclusion, the results of our study show that MR imaging is an effective method for characterizing small hepatic lesions in routine clinical practice.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  J. M. Willatt, H. K. Hussain, S. Adusumilli, and J. A. Marrero MR Imaging of Hepatocellular Carcinoma in the Cirrhotic Liver: Challenges and Controversies Radiology, May 1, 2008; 247(2): 311 - 330. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. A. Patterson, H. I. Khalil, and D. M. Panicek MRI evaluation of small hepatic lesions in women with breast cancer. Am. J. Roentgenol., August 1, 2006; 187(2): 307 - 312. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. I. Khalil, S. A. Patterson, and D. M. Panicek Hepatic Lesions Deemed Too Small to Characterize at CT: Prevalence and Importance in Women with Breast Cancer Radiology, June 1, 2005; 235(3): 872 - 878. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
; area
under curve [Az] = 0.94) and either of the expert
interpreters (observer 1, •; Az = 0.88; observer 2, 
;
Az = 0.84) (p > 0.05).
