HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) All Versions of this Article: 2371041444v1 237/1/170    most recent Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Ward, J. Articles by Wyatt, J. I. Search for Related Content PubMed PubMed Citation Articles by Ward, J. Articles by Wyatt, J. I.

Liver Metastases in Candidates for Hepatic Resection: Comparison of Helical CT and Gadolinium- and SPIO-enhanced MR Imaging¹

¹ From the MRI Department, Clinical Radiology (J.W., P.J.R., J.A.G., S.D., D.W.), Hepatobiliary and Transplantation Unit (J.P.A.L., K.R.P., G.J.T.), and Department of Histopathology (J.I.W.), St James's University Hospital, Beckett Street, Leeds LS9 7TF, England. Received August 19, 2004; revision requested October 28; revision received December 8; accepted January 17, 2005. Address correspondence to J.W. (e-mail: janice.ward{at}leedsth.nhs.uk).

	ABSTRACT

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION References

 
PURPOSE: To prospectively compare accuracy of dynamic contrast material–enhanced thin-section multi–detector row helical computed tomography (CT), high-spatial-resolution three-dimensional (3D) dynamic gadolinium-enhanced magnetic resonance (MR) imaging, and superparamagnetic iron oxide (SPIO)-enhanced MR imaging with optimized gradient-echo (GRE) sequence for depiction of hepatic lesions; surgery and histologic analysis were the reference standard.

MATERIALS AND METHODS: Local ethics committee approval was granted, and written informed consent was obtained. Fifty-eight patients (45 men, 13 women; age range, 47–82 years) with hepatic metastases were imaged with multi–detector row CT (3.2-mm section thickness), 3D dynamic gadolinium-enhanced MR imaging (2.5-mm effective section thickness), and SPIO-enhanced MR by using an optimized T2-weighted GRE sequence. Images were reviewed independently by two blinded observers who identified and localized lesions with a four-point confidence scale. Accuracy of each technique was measured with alternative free-response receiver operating characteristic analysis. Results were correlated with findings at surgery with intraoperative ultrasonography or histopathologic examination. Statistical differences among techniques for each observer were measured.

RESULTS: Accuracy values for each observer for all metastases (n = 215) and 1.0-cm or smaller metastases (n = 80), respectively, follow: For CT, those for reader 1 were 0.82 and 0.65; for reader 2, 0.81 and 0.68. For gadolinium-enhanced MR imaging, those for reader 1 were 0.92 and 0.79; for reader 2, 0.90 and 0.76. For SPIO-enhanced MR imaging, those for reader 1 were 0.92 and 0.83; for reader 2, 0.92 and 0.81. For all metastases for both observers, there was no significant difference between MR techniques, but both were significantly more accurate than CT (P < .01). For metastases 1.0 cm or smaller and one observer, there was no significant difference between MR techniques, but both were more accurate than CT (P < .01); for the other observer, SPIO-enhanced MR imaging was more accurate than gadolinium-enhanced MR imaging (P < .05) and CT (P < .02), but there was no significant difference between gadolinium-enhanced MR imaging and CT (P = .2).

CONCLUSION: Accuracy for gadolinium-enhanced MR imaging and SPIO-enhanced MR imaging was similar. Both techniques were significantly more accurate than CT.

© RSNA, 2005

	INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION References

 
The detection and characterization of small metastases remains a major challenge for imaging. Successful diagnosis depends on high spatial resolution, good liver-to-lesion contrast, and specific signal intensity or attenuation and contrast enhancement characteristics. In patients with established malignant disease, dynamic contrast material–enhanced computed tomography (CT) still is the primary imaging modality for the detection of focal hepatic lesions, with reported sensitivity values for the detection of liver metastases that range from 73% to 85% (1–6) when lesions of all sizes are included in the analysis. Researchers in studies in which the detection rate for surgically confirmed small lesions has been specifically measured report that the majority of undetected lesions are smaller than 1.5 cm in maximum diameter measured in the axial plane (1–3,5,6).

Dynamic gadolinium-enhanced magnetic resonance (MR) imaging techniques with conventional two-dimensional sequences, compared with unenhanced MR imaging, helical CT, and CT arterial portography, have been shown to improve the detection of liver lesions and are of particular value in characterization of lesions on the basis of perfusion patterns (7–12). The detection of small lesions, however, is limited by partial volume averaging because of the relatively large section thickness and intersection gap associated with conventional two-dimensional gradient-echo (GRE) MR imaging. Superparamagnetic iron oxide (SPIO) has been developed as a liver-specific contrast agent that targets the Kupffer cells of the liver. Numerous studies have shown SPIO-enhanced MR imaging to be one of the most accurate techniques both in terms of the number of metastases depicted and in the size of the smallest lesion shown (13–24), but the detection of surgically confirmed subcentimeter metastases still is disappointing (3,10,13–15,18,20). Since enhancement with SPIO depends on susceptibility effects, which vary with different pulse sequences, diagnostic effectiveness is extremely dependent on the pulse sequence (20).

Recent technical advances have improved spatial and temporal resolution in both CT and MR imaging. Multi–detector row helical CT allows rapid imaging of the entire liver at narrower beam collimation with minimal signal-to-noise penalty. Results of studies in which different section thicknesses used at CT for lesion detection were compared have been conflicting (25,26), although improved sensitivity values for detection of small metastases with a section thickness of less than 5.0 mm have been reported (25).

In MR imaging, the implementation of high-performance gradient systems and body phased-array coils has resulted in improved temporal resolution with shorter acquisition times and high spatial and high contrast resolution. Consequently, it is now possible to image the entire liver with high lesion-to-liver contrast by using gadolinium-enhanced three-dimensional (3D) T1-weighted GRE MR sequences (27). Compared with two-dimensional methods, 3D T1-weighted sequences allow the liver to be imaged with thinner sections, no intersection gap, higher signal-to-noise ratios, and improved spatial resolution in a comparable breath-hold period. These features facilitate the detection of subcentimeter lesions by increasing in-plane resolution and eliminating partial volume averaging effects.

Improved pulse sequences for use with SPIO have also emerged with the introduction of high-performance gradients. The T2-weighted GRE MR sequences, with lower bandwidths to increase signal-to-noise ratio and flow compensation to reduce flow artifacts, can now be performed in acquisition times short enough for breath holding. In a recent study (28), an accuracy of 93% (calculated with alternating free-response receiver operating characteristic method) was achieved with an optimized SPIO-enhanced GRE sequence for the detection of surgically confirmed metastases. Conversely, a SPIO-enhanced fast spin-echo sequence, with which an accuracy of only 82% was achieved, offered no improvement in sensitivity when it was compared with unenhanced sequences.

To our knowledge, there are no published studies in which CT and MR imaging for detection of hepatic metastases with these current techniques have been compared. Thus, the purpose of our study was to prospectively compare the accuracy of dynamic contrast-enhanced thin-section helical CT, high-spatial-resolution 3D dynamic gadolinium-enhanced MR imaging, and SPIO-enhanced MR imaging by using an optimized GRE MR sequence for depiction of hepatic lesions, with surgery and histologic examination as the reference standard.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION References

 
Patients
Local ethics committee approval was granted, and written informed consent was obtained from each patient prior to entry into the study. For the purpose of this study, the SPIO contrast agent ferucarbotran was provided free of charge by the manufacturer; however, the authors had complete control of the data and the information contained within this article. From June 2001 to January 2003, 95 consecutive patients with hepatic metastases who were referred to our institution for consideration of surgical resection underwent CT and MR imaging following review of imaging findings from the referring hospital. Of these, 37 patients were subsequently regarded as unsuitable for surgery because of disease extent or cardiac status. These patients were not included in the analysis. The final study group comprised 58 patients with an age range of 47–82 years (mean age, 64 years); 45 of the 58 patients were men (age range, 47–82 years; mean age, 65 years) and 13 were women (age range, 47–74 years; mean age, 61 years). Fifty-six patients had colorectal metastases, one patient had hepatic metastases from a gastrointestinal stromal tumor, and one had hepatic metastases from primary breast cancer.

All patients underwent hepatic resection after intraoperative ultrasonography (US) on the basis of the findings that the disease was confined to the liver and that there was sufficient liver parenchyma to maintain liver function postoperatively; number, size, and distribution of lesions were not limiting factors. Twenty-nine patients underwent right hemihepatectomy; two patients, left hemihepatectomy; 11 patients, right trisectionectomy; four patients, left trisectionectomy; and one patient, left lateral sectionectomy. Eighteen of these 47 patients from the group of 58 patients also underwent metastatectomy of additional metastases, and one patient underwent radiofrequency ablation of three metastases in the residual liver. For the remaining 11 patients of 58, right posterior sectionectomy (segments VI and VII), with resection of segment III and metastatectomy of metastases in segments IV and VIII, was performed in one patient, and right posterior sectionectomy, with left lateral sectionectomy, was performed in another patient. More limited resections were performed in the remaining nine patients: Resection of segment VII was performed in three patients; metastatectomy of multiple metastases, in three patients; and metastatectomy of solitary metastasis, in three patients.

CT and both MR imaging examinations were performed at the same visit to minimize inconvenience to the patient.

CT Scanning
CT was performed prior to MR imaging in all patients. All examinations were performed with a multi–detector row CT scanner (MX8000; Philips Medical Systems, Eindhoven, the Netherlands). Four sections were acquired simultaneously with single-beam collimation of 2.5 mm to produce effective section thickness of 3.2 mm, tube voltage of 120 kVp, tube current–time product of 220 mAs, and pitch of 0.875. Images were reconstructed with a linear interpolation algorithm of 180° at 1.6-mm increments to achieve a 50% overlap, and 120–160 sections were obtained in 20–24 seconds (average, 21 seconds). Images were obtained at the portal venous phase of enhancement after injection of 150 mL of nonionic contrast medium (Ultravist 300; Schering, Berlin, Germany) injected with a power injector (Spectris SMR 200; Medrad, Ely, England) at a rate of 5 mL/sec. Imaging commenced 65 seconds after the start of the injection. Images were reviewed at the end of data acquisition to determine the need for delayed imaging to further characterize lesions with features suggestive of hemangioma.

MR Imaging
MR imaging was performed with a 1.5-T clinical imager (Symphony; Siemens, Erlangen, Germany) by using 30 mT/m gradients and a phased-array body coil. Prior to contrast agent administration, all patients were imaged with the following unenhanced sequences, according to our standard protocol: true fast imaging with steady-state precession sequence (repetition time msec/echo time msec, 5.57/2.79; flip angle, 80°; number of signals acquired, one; matrix, 148 x 256), half-Fourier rapid acquisition with relaxation enhancement single-shot fast spin-echo sequence (1100/95; echo train length, 128; echo spacing, 4.5 msec; number of signals acquired, one; section thickness, 6.0 mm with a 10% gap; matrix, 148 x 256), T1-weighted GRE in-phase sequence (172/4.7; flip angle, 75°; number of signals acquired, one; matrix, 134 x 256), and T1-weighted GRE out-of-phase sequence (144/2.4; flip angle, 75°; number of signals acquired, one; section thickness, 8.0–10.0 mm with a 10% gap; matrix, 115 x 256).

Prior injection of a gadolinium-based contrast agent does not affect the signal intensity change produced by SPIO on T2-weighted MR images, but prior injection of SPIO has a marked and prolonged effect on subsequent gadolinium enhancement. For this reason, gadolinium-enhanced MR images were obtained immediately after the performance of unenhanced sequences and approximately 4 hours before the acquisition of SPIO-enhanced MR images in all cases.

T1-weighted fat-suppressed 3D spoiled GRE MR images (29) were obtained immediately before and 15, 45, and 120 seconds after the start of bolus injection of the gadolinium-based contrast agent (Magnevist; Schering, Berlin, Germany) with the following parameters: 3.38/1.55; flip angle, 15°; bandwidth, 490 Hz/pixel; echo, slightly asymmetric (echo position, 38%); reconstruction, full Fourier in the phase-encoding direction with a phase resolution of 65%; field of view, 280–400 mm in the readout direction and 210–300 mm in the phase-encoding direction (75% rectangular field of view); and matrix size, 256 x 125 pixels. Interpolation in the phase-encoding direction was applied to make the pixel size in the phase-encoding and frequency directions the same. In the section direction, partial Fourier reconstruction and section interpolation allowed the reconstruction of 72–80 partitions, with a section thickness of 2.5–3.0 mm, which provided a slab thickness of 180–240 mm with full liver coverage in all patients. These parameters achieved actual and interpolated voxel dimensions of 1.6 x 2.4 x 4.0–4.8 mm and 1.6 x 1.6 x 2.5–3.0 mm, respectively. A chemically selective fat-saturation pulse was applied intermittently to provide fat suppression without a prohibitive time penalty. All patients received gadolinium-based contrast agent at a dose of 0.1 mmol per kilogram of body weight (dose range, 8–20 mL) followed by a 30-mL bolus of saline. Gadolinium-based contrast agent and saline were injected with the power injector at a rate of 4 mL/sec. Additional images were acquired approximately 10 minutes after gadolinium-based contrast agent injection if a benign lesion was suspected at performance of unenhanced sequences.

Approximately 4 hours after the acquisition of gadolinium-enhanced MR images, each patient underwent a second MR imaging examination. After standard localizing sequences were performed, the previously described T1-weighted fat-suppressed 3D spoiled GRE sequence was performed before and 15, 45, and 120 seconds after bolus injection of SPIO (ferucarbotran, Resovist; Schering). Patients with a body weight of 35–60 kg received 0.9 mL of SPIO, whereas patients with a body weight of more than 60 kg received 1.4 mL at a dose of 7–11 µmol/kg body weight. The contrast agent was administered by means of rapid hand injection and was immediately followed by a 30-mL saline flush.

Ten minutes after the end of injection, T2-weighted GRE MR images (185/15; flip angle, 30°; number of signals acquired, one; bandwidth, 65 Hz/pixel; section thickness, 6.0 mm with a 10% gap; matrix, 115 x 256) were obtained (28). Flow-compensation gradients were applied to minimize motion-related artifacts, and a frequency-selected fat-suppression pulse was applied to improve the detection of surface lesions. Despite manual and automated shimming of the magnetic field, fat suppression was inhomogeneous in four patients, so in these patients, T2-weighted GRE MR images were obtained without fat suppression. Three to four sections were acquired during a 17–20-second breath hold, and nine to 11 acquisitions were required to cover the entire liver.

For all MR imaging sequences, images were acquired in the transverse plane with a 280–400-mm field of view, depending on patient size, and a rectangular field of view of 75%; in all patients, these factors were the same with all sequences performed before and after gadolinium-based contrast agent and SPIO administration. All images were obtained during suspended expiration, with acquisition times of 16–20 seconds, depending on the patient's capacity for breath holding.

Qualitative Analysis
Three sets of images were analyzed in each patient: (a) CT images, (b) unenhanced MR images (acquired with T1-weighted in-phase and out-of-phase GRE and half-Fourier rapid acquisition with relaxation enhancement sequences) and gadolinium-enhanced T1-weighted 3D fat-suppressed GRE MR images, and (c) unenhanced MR images and SPIO-enhanced T2-weighted GRE MR images. The three sets of images were reviewed at a workstation (Radworks 5.1; Applycare Medical Imaging, GE Healthcare, Waukesha, Wis), with a minimum of 4 weeks between each viewing, and patient details were removed to minimize observer bias. The images were evaluated independently by two observers (J.W., P.J.R.) who were unaware of the results of the other imaging sequences, the results for the other observer, and the findings at surgery or histopathologic examination. Both observers had 14 years of experience in MR imaging of the liver.

For each set of images, each observer recorded the presence of all suspected lesions in each patient on the basis of a four-point confidence scale as follows: score 1, probably not a lesion; score 2, possible lesion; score 3, probable lesion; and score 4, definite lesion. To achieve an accurate comparison between the lesions that were assigned scores and true-positive lesions that were confirmed at surgery and histopathologic examination, at the time of assignment of scores the observers identified each lesion by means of a grid reference, image number segmental location, and size. Each lesion was also classified as benign, malignant, or indeterminate.

On MR images, characterization was based on lesion definition and signal intensity characteristics; on half-Fourier rapid acquisition with relaxation enhancement MR images, it was based on enhancement characteristics following gadolinium-based contrast agent administration. Lesions were classified as benign if they were sharply defined with a homogeneously high signal intensity similar to that of cerebrospinal fluid on half-Fourier rapid acquisition with relaxation enhancement MR images with no enhancement after gadolinium-based contrast agent administration and a near signal void appearance (simple cyst) or with discontinuous peripheral nodular enhancement progressing to involve most or all of the lesion on delayed-enhancement images (hemangiomas). Lesions were classified as malignant if they were ill defined with a slight to moderately high signal intensity relative to background liver on half-Fourier rapid acquisition with relaxation enhancement MR images and continuous rim enhancement with central progression, borders that became indistinct over time, or peripheral washout after gadolinium-based contrast agent administration.

Lesions were characterized as benign on CT images if they were homogeneous, sharply defined, and hypoattenuating, with an attenuation similar to that of bile, and as malignant if they were ill defined, heterogeneous, and had a higher attenuation than bile and some degree of enhancement. If lesions did not show characteristic benign or malignant features on CT or MR images, they were regarded as indeterminate.

Comparison with Surgical Findings
At the time of surgery, the surgical and intraoperative US findings were carefully compared with the findings on CT and MR images by an author (S.D.) who was different from those who performed the blinded reading; this author had 8 years of experience in liver MR imaging. CT and MR imaging findings were also compared with the histopathologic findings after dissection of the resected specimen, which was sectioned at 3-mm intervals in the transverse plane by a pathologist with 18 years of experience in hepatobiliary pathology (J.I.W.). Surgery was performed by one of three hepatobiliary surgeons (with 12, 10, and 5 years of experience) who were aware of the imaging findings, and intraoperative US was performed by one of three experienced sonologists (two of whom had 7 years of experience and one of whom had 12 years of experience) in conjunction with the surgeon. Hepatic resection was performed between 1 and 12 weeks (mean, 5.2 weeks) after imaging. To identify small lesions missed at preoperative imaging, intraoperative US, and surgical palpation, we also reviewed CT or MR images obtained at least 6 months after surgery. In patients with additional lesions at follow-up imaging, the preoperative images were reevaluated by one author (J.W.) to see if any of the new lesions were visible retrospectively on initial images.

After identification and localization of all true-positive lesions, an assessment of all false-positive interpretations determined with confidence scores of 3 or 4 was undertaken by the same author by means of review of all preoperative imaging results, histologic findings, and follow-up CT or MR imaging results.

Statistical Analysis
Alternating free-response receiver operating characteristic analysis (Analyze-it; General & Clinical Laboratory Statistics, 1999) was used to determine the accuracy of each technique. Alternating free-response receiver operating characteristic analysis allows positional information to be recorded, and the recording of this information enables all of the observer's responses to be correlated with all of the lesions present. The alternating free-response receiver operating characteristic curves were calculated for each observer and each technique by plotting the true-positive fraction against the likelihood of obtaining a false-positive image (ie, an image with one or more false-positive lesions) at each confidence score. The area under each curve was used to indicate the overall performance rating of each technique and each observer. The statistical significance of any difference between the techniques assessed by an individual observer was determined according to the method of McNeil and Hanley (30). In addition, the sensitivity of each technique for each observer was assessed by using only those lesions assigned confidence scores of 3 or 4. The McNemar test was used to assess the statistical significance of differences in performance between each technique within an individual observer. Separate analyses of all malignant lesions and of malignant lesions of less than 1.0 cm were undertaken. In 54 of 58 patients, lesions were categorized according to size on the basis of findings at intraoperative US or histopathologic examination. In four patients, each with multiple lesions ranging in size from less than 1.0 cm to 3.5 cm, there had been a substantial increase in the size of all lesions in the time between imaging and surgery (4–8 weeks), so lesion size was recorded on the basis of preoperative imaging results. A difference with a P value of less than .05 was regarded as significant.

An analysis of all false-negative and false-positive observations with a confidence score of 3 or 4 also was undertaken. Malignant lesions that were assigned scores and were classified as malignant or indeterminate were regarded as true-positive observations, malignant lesions that were assigned scores and were classified as benign were regarded as false-negative observations, and benign lesions that were assigned scores and were classified as malignant were regarded as false-positive observations.

Interobserver variability for lesion detection for each technique was assessed with the statistic analysis. A value of 0.40 or less was considered to indicate poor correlation; 0.41–0.75, good correlation; and more than 0.75, excellent correlation.

	RESULTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION References

 
Malignant Lesions
Two hundred fifteen lesions were present in 58 patients. All patients had at least one metastasis, and the maximum number of malignant lesions in any patient was 21. Eighty-five lesions were larger than 2.0 cm, 50 were 1.1–2.0 cm, and 80 were 1.0 cm or smaller.

Accuracy
For each technique the areas under the alternating free-response receiver operating characteristic curves and the 95% confidence intervals (CIs) for each observer are shown in Figures 1 and 2. The results for all malignant lesions are shown in Figure 1, and the results for malignant lesions that were 1.0 cm or smaller are shown in Figure 2. In 35 patients, all true-positive lesions as determined by the reference standard (75 of 215) were assigned a confidence score of 3 or 4 by both observers with all three techniques. In the other 23 patients, at least one lesion was missed or was assigned a low confidence score of 1 or 2 by at least one observer with at least one technique. Regardless of lesion size, SPIO-enhanced MR imaging was the most accurate technique for one observer. The second observer achieved equal highest accuracy values for SPIO-enhanced MR imaging and gadolinium-enhanced MR imaging for all malignant lesions, but SPIO-enhanced MR imaging was the most accurate technique for lesions that were 1.0 cm or smaller.

View larger version (30K): [in this window] [in a new window] [Download PPT slide]   Figure 1. Histograms show the area under the alternating free-response receiver operating characteristic curve for each observer and each technique for all 215 malignant lesions. The 95% CIs for observers 1 and 2 were 0.88 to 0.95 and 0.86 to 0.94, 0.88 to 0.96 and 0.89 to 0.96, and 0.76 to 0.87 and 0.75 to 0.88 for gadolinium-enhanced MR imaging, SPIO-enhanced MR imaging, and CT, respectively. For both observers, there was no significant difference between gadolinium-enhanced MR imaging and SPIO-enhanced MR imaging, but both MR imaging techniques were significantly more accurate than CT (P < .01).

View larger version (29K): [in this window] [in a new window] [Download PPT slide]   Figure 2. Histograms show the area under the alternating free-response receiver operating characteristic curve for each observer and each technique for all 80 malignant lesions that were 1.0 cm or smaller. The 95% CIs for observers 1 and 2 were 0.71 to 0.87 and 0.67 to 0.84, 0.75 to 0.90 and 0.75 to 0.90, and 0.55 to 0.75 and 0.59 to 0.78 for gadolinium-enhanced MR imaging, SPIO-enhanced MR imaging, and CT, respectively. For observer 1, there was no significant difference between gadolinium-enhanced MR imaging and SPIO-enhanced MR imaging (P = .4), but both MR imaging techniques were significantly more accurate than CT (P < .01). For observer 2, SPIO-enhanced MR imaging was significantly more accurate than gadolinium-enhanced MR imaging (P < .05) and CT (P = .01), but there was no significant difference between gadolinium-enhanced MR imaging and CT (P = .2).

Sensitivity
For each technique, the sensitivity values of each observer and the 95% CIs for the differences are shown in Figures 3 and 4. The results for all malignant lesions and for malignant lesions that were 1.0 cm or smaller are shown in Figures 3 and 4, respectively.

View larger version (29K): [in this window] [in a new window] [Download PPT slide]   Figure 3. Histograms show the sensitivity of each observer with each technique for all 215 malignant lesions. The number of lesions detected by each observer is indicated above the bars. The differences in the sensitivity values of SPIO-enhanced MR imaging and gadolinium-enhanced MR imaging were 0.037 (95% CI: –0.004, 0.064; P = .07) and 0.051 (95% CI: 0.003, 0.085; P = .03) for observers 1 and 2, respectively. The differences in the sensitivity values of SPIO-enhanced MR imaging and CT were 0.144 (95% CI: 0.100, 0.161; P <.001) and 0.084 (95% CI: 0.032, 0.118; P < .01) for observers 1 and 2, respectively. The differences in the sensitivity values of gadolinium-enhanced MR imaging and CT were 0.107 (95% CI: 0.065, 0.123; P < .001) and 0.033 (95% CI: 0.014, 0.069; P = .19) for observers 1 and 2, respectively.

View larger version (26K): [in this window] [in a new window] [Download PPT slide]   Figure 4. Histograms show the sensitivity of each observer with each technique for all 80 malignant lesions that were 1.0 cm or smaller. The number of lesions detected by each observer is indicated above the bars. The differences in the sensitivity values of SPIO-enhanced MR imaging and gadolinium-enhanced MR imaging were 0.075 (95% CI: –0.022, 0.134; P = .15) and 0.125 (95% CI: 0.017, 0.184; P < .05) for observers 1 and 2, respectively. The differences in the sensitivity values of gadolinium-enhanced MR imaging and CT were 0.238 (95% CI: 0.137, 0.262; P < .001) and 0.138 (95% CI: 0.015, 0.219; P < .05) for observers 1 and 2, respectively. The differences in the sensitivity values of gadolinium-enhanced MR imaging and CT were 0.163 (95% CI: 0.058, 0.206; P < .01) and 0.013 (95% CI: 0.088, 0.108; P > .99) for observers 1 and 2, respectively.

False-Negative Lesions and Changes in Surgical Treatment
At confidence levels of 3 or 4, 19 lesions in 11 patients were not detected by either observer with any technique; all were 1.0 cm or smaller. Eleven of 19 lesions were detected with intraoperative US, three were detected with surgical palpation, and five were detected only at histopathologic inspection. One of 19 lesions was visible on SPIO-enhanced MR images only, another was visible on SPIO-enhanced and gadolinium-enhanced MR images, and a third was visible on gadolinium-enhanced MR images, but these lesions were all assigned low confidence scores of 1 or 2 by one observer. One further lesion was identified retrospectively with both MR imaging techniques, but it was not visible at CT. One of the 19 missed lesions occurred in a patient with seven lesions in total; six of the seven lesions were visible with both of the MR imaging techniques, and four of the seven, with CT. Although there was an interval of only 4 weeks between imaging and surgery, all of the lesions had doubled in size in the interval, and the missed lesion, which also was visible at intraoperative US, was only 5.0 mm at histologic examination. The other 10 patients also had multiple lesions, but there was no apparent increase in the size of the detected lesions between imaging and surgery, so it is unlikely that the interval had a substantial influence on our results.

In six of 11 patients, the location of the missed lesions did not cause an alteration in surgical treatment. In three patients, however, metastatectomy was performed in addition to planned hepatic resection, and in a fourth patient who underwent a right hemihepatectomy and planned radiofrequency ablation to a single lesion in the left lobe, two further lesions in the left lobe that were undetected at preoperative imaging also were treated with radiofrequency ablation. In one further patient with two missed lesions at preoperative imaging and in whom the surgical approach was not altered as a result of the missed lesions, a planned right hemihepatectomy was extended to a right trisectionectomy because of an abnormality in segment IV, and this abnormality was diagnosed at MR imaging as a focal fatty change but was thought to be a metastasis at intraoperative US; histopathologic examination results confirmed the abnormality as focal fatty change.

All lesions that were larger than 1.0 cm were detected by using at least one observer with at least one technique. At confidence scores of 3 or 4, however, eight lesions detected with SPIO-enhanced MR imaging were missed with gadolinium-enhanced MR imaging and CT (six were <1.0 cm, and two were 1.1–2.0 cm) by both observers (Fig 5). Two subcentimeter lesions were missed with gadolinium-enhanced MR imaging, and 12 lesions (seven were <1.0 cm, and four were 1.1–2.0 cm) were missed with CT by both observers (Figs 6, 7). Two further lesions (one was 1.1–2.0 cm, and one was <1.0 cm) were recorded with gadolinium-enhanced MR imaging and CT but not with SPIO-enhanced MR imaging (Fig 8), and a subcentimeter lesion was recorded with CT but not with either MR imaging technique by both observers. None of the lesions in our study were visible only on unenhanced MR images.

View larger version (156K): [in this window] [in a new window] [Download PPT slide]   Figure 5a. False-negative lesions on CT scan and gadolinium-enhanced MR image in a 77-year-old man with multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A 1.0-cm metastasis (arrow) that is clearly seen in the tip of the left lobe on c is not visible on a or b.

View larger version (145K): [in this window] [in a new window] [Download PPT slide]   Figure 5b. False-negative lesions on CT scan and gadolinium-enhanced MR image in a 77-year-old man with multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A 1.0-cm metastasis (arrow) that is clearly seen in the tip of the left lobe on c is not visible on a or b.

View larger version (146K): [in this window] [in a new window] [Download PPT slide]   Figure 5c. False-negative lesions on CT scan and gadolinium-enhanced MR image in a 77-year-old man with multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A 1.0-cm metastasis (arrow) that is clearly seen in the tip of the left lobe on c is not visible on a or b.

View larger version (125K): [in this window] [in a new window] [Download PPT slide]   Figure 6a. False-negative lesions on CT scan in a 57-year-old woman with multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A subcentimeter metastasis (arrow) in segment VIII that was assigned high confidence scores by both observers on b and c was not assigned a score by either observer on a.

View larger version (141K): [in this window] [in a new window] [Download PPT slide]   Figure 6b. False-negative lesions on CT scan in a 57-year-old woman with multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A subcentimeter metastasis (arrow) in segment VIII that was assigned high confidence scores by both observers on b and c was not assigned a score by either observer on a.

View larger version (144K): [in this window] [in a new window] [Download PPT slide]   Figure 6c. False-negative lesions on CT scan in a 57-year-old woman with multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A subcentimeter metastasis (arrow) in segment VIII that was assigned high confidence scores by both observers on b and c was not assigned a score by either observer on a.

View larger version (118K): [in this window] [in a new window] [Download PPT slide]   Figure 7a. False-negative lesion on CT scan in a 65-year-old woman with a diffusely fatty liver and multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). The presence of fat on a may have obscured visualization of a 1.2-cm metastasis (arrow), which was located on the border of segments II and IV and was clearly seen on b and c.

View larger version (132K): [in this window] [in a new window] [Download PPT slide]   Figure 7b. False-negative lesion on CT scan in a 65-year-old woman with a diffusely fatty liver and multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). The presence of fat on a may have obscured visualization of a 1.2-cm metastasis (arrow), which was located on the border of segments II and IV and was clearly seen on b and c.

View larger version (141K): [in this window] [in a new window] [Download PPT slide]   Figure 7c. False-negative lesion on CT scan in a 65-year-old woman with a diffusely fatty liver and multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). The presence of fat on a may have obscured visualization of a 1.2-cm metastasis (arrow), which was located on the border of segments II and IV and was clearly seen on b and c.

View larger version (111K): [in this window] [in a new window] [Download PPT slide]   Figure 8a. False-negative lesion on SPIO-enhanced MR image in a 67-year-old man with multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A 3.0-mm metastasis (arrow) in segment VIII clearly seen on a and b was not assigned a score by either observer on c. The lesion, which was identified on c at retrospective review, was incorrectly interpreted as a vessel by both observers at the time they assigned scores.

View larger version (108K): [in this window] [in a new window] [Download PPT slide]   Figure 8b. False-negative lesion on SPIO-enhanced MR image in a 67-year-old man with multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A 3.0-mm metastasis (arrow) in segment VIII clearly seen on a and b was not assigned a score by either observer on c. The lesion, which was identified on c at retrospective review, was incorrectly interpreted as a vessel by both observers at the time they assigned scores.

View larger version (117K): [in this window] [in a new window] [Download PPT slide]   Figure 8c. False-negative lesion on SPIO-enhanced MR image in a 67-year-old man with multiple metastases. Transverse images obtained with (a) thin-section helical CT, (b) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (c) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A 3.0-mm metastasis (arrow) in segment VIII clearly seen on a and b was not assigned a score by either observer on c. The lesion, which was identified on c at retrospective review, was incorrectly interpreted as a vessel by both observers at the time they assigned scores.

Interobserver Variability
The best agreement between the two observers was achieved with SPIO-enhanced MR imaging ( = 0.90 for all malignant lesions and = 0.91 for malignant lesions that were 1.0 cm or smaller). Excellent interobserver agreement was also obtained with CT ( = 0.81 for all malignant lesions and = 0.83 for malignant lesions 1.0 cm or smaller). The values of 0.74 for all malignant lesions and 0.67 for malignant lesions of 1.0 cm or smaller also indicated good agreement between the two observers for gadolinium-enhanced MR imaging; however, although both observers achieved similar accuracy values, sensitivity values were more variable, particularly for lesions that were 1.0 cm or smaller, and the actual lesions recorded by each observer also varied.

False-Positive Findings
At confidence scores of 3 or 4, a total of 1072 true-positive interpretations and 20 false-positive interpretations were recorded to yield a false-positive rate of 1.9%. Substantially more false-positive interpretations were determined at CT (n = 15) than were determined at gadolinium-enhanced MR imaging (n = 3) or SPIO-enhanced MR imaging (n = 2). Six of 15 false-positive interpretationss with CT were caused by the incorrect interpretation of benign lesions as metastases (Fig 9). The six interpretations arose from single lesions in four patients (ie, four lesions); two of four were characterized as malignant by both observers (four false-positive interpretations), and each of the other two were characterized as such by a different single observer (two false-positive interpretations). At retrospective review, four further false-positive interpretations were attributed to thrombosed vessels (all <1.0 cm) (two interpretations were determined by one observer each, and one was determined by both observers), one was attributed to partial volume averaging (one observer, 1.2 cm in size), and one was attributed to a focus of more severe fatty change in a diffusely fatty liver (one observer, 2.0 cm in size). Three false-positive interpretations (all <1.0 cm), which were assigned scores by one observer, were unexplained. All three false-positive interpretations determined with gadolinium-enhanced MR images (<1.0 cm) were determined by one observer each and were attributed to patent (2) and thrombosed vessels (1). The two false-positive interpretations determined with SPIO-enhanced MR images were attributed to a vessel in one false-positive interpretation (<1.0 cm) and to an unexplained subcentimeter area of increased signal intensity on the surface of the liver in the other. This unexplained subcentimeter area of increased signal intensity was also identified at CT, intraoperative US, and surgical palpation but was not confirmed at histopathologic examination.

View larger version (121K): [in this window] [in a new window] [Download PPT slide]   Figure 9a. Coexistent benign and malignant lesions in a 52-year-old man. Transverse images obtained with (a) thin-section helical CT, (b) unenhanced half-Fourier rapid acquisition with relaxation enhancement MR imaging (1100/95), (c) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (d) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A large metastasis (short arrow) in segment II and a simple cyst (long arrow) in segment IV have markedly different signal intensity on b and c but similar characteristics on a and d. With combined review of pre- and postcontrast MR images, both observers classified the lesions correctly on c and d, but the cyst was incorrectly classified as a metastasis on a by both observers.

View larger version (136K): [in this window] [in a new window] [Download PPT slide]   Figure 9b. Coexistent benign and malignant lesions in a 52-year-old man. Transverse images obtained with (a) thin-section helical CT, (b) unenhanced half-Fourier rapid acquisition with relaxation enhancement MR imaging (1100/95), (c) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (d) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A large metastasis (short arrow) in segment II and a simple cyst (long arrow) in segment IV have markedly different signal intensity on b and c but similar characteristics on a and d. With combined review of pre- and postcontrast MR images, both observers classified the lesions correctly on c and d, but the cyst was incorrectly classified as a metastasis on a by both observers.

View larger version (121K): [in this window] [in a new window] [Download PPT slide]   Figure 9c. Coexistent benign and malignant lesions in a 52-year-old man. Transverse images obtained with (a) thin-section helical CT, (b) unenhanced half-Fourier rapid acquisition with relaxation enhancement MR imaging (1100/95), (c) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (d) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A large metastasis (short arrow) in segment II and a simple cyst (long arrow) in segment IV have markedly different signal intensity on b and c but similar characteristics on a and d. With combined review of pre- and postcontrast MR images, both observers classified the lesions correctly on c and d, but the cyst was incorrectly classified as a metastasis on a by both observers.

View larger version (130K): [in this window] [in a new window] [Download PPT slide]   Figure 9d. Coexistent benign and malignant lesions in a 52-year-old man. Transverse images obtained with (a) thin-section helical CT, (b) unenhanced half-Fourier rapid acquisition with relaxation enhancement MR imaging (1100/95), (c) portal phase gadolinium-enhanced 3D T1-weighted GRE MR imaging (3.8/1.55; flip angle, 15°), and (d) SPIO-enhanced T2-weighted GRE MR imaging (185/15; flip angle, 30°). A large metastasis (short arrow) in segment II and a simple cyst (long arrow) in segment IV have markedly different signal intensity on b and c but similar characteristics on a and d. With combined review of pre- and postcontrast MR images, both observers classified the lesions correctly on c and d, but the cyst was incorrectly classified as a metastasis on a by both observers.

	DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION References

 
Most lesions that are larger than 1.0 cm are detected with current imaging methods, but the detection of smaller lesions still is poor. Also, studies with comparison of different imaging techniques for the detection of hepatic metastases have produced conflicting results (3,10,12–14,16,19,21,22,24,31), and all were performed with imaging strategies that now would not be considered state of the art. A critical feature of our current study was the use of three contemporaneous techniques that reflect the application of recent advances in CT and MR technology.

Our CT technique employed portal venous phase imaging with multi–detector row technology. A small minority of colorectal hepatic metastases may only be seen on arterial phase images (4,32), but most authorities agree that the improvement in detection is too small to justify the additional radiation exposure induced by the addition of an arterial phase acquisition, so we obtained images at only the portal phase of enhancement. In one patient with primary breast cancer in the current study, the lack of arterial phase images at CT did not bias our results in favor of MR imaging, since a solitary large metastasis was well seen with all three techniques. Two subcentimeter lesions were undetected only at CT in our patient with a primary gastrointestinal stromal tumor, but arterial phase gadolinium-enhanced MR imaging showed no evidence of lesion hypervascularity.

Our gadolinium-enhanced technique employed a 3D T1-weighted GRE MR sequence with imaging at arterial and portal phases of enhancement. The arterial phase is helpful for distinguishing benign and malignant lesions; even hypovascular lesions, which are best seen at the portal phase, frequently exhibit a characteristic rim enhancement that may be seen at only the arterial phase. Because of the superior contrast resolution of MR imaging, rim enhancement in metastases usually is more marked at MR imaging than it is at CT.

Our choice of T2-weighted MR sequence after administration of SPIO was based on data from a recent study in which the researchers evaluated four breath-hold T2-weighted sequences by using optimized parameters to establish the most effective SPIO-enhanced MR sequence for detection of hepatic metastases (28). In that study, the highest mean sensitivity was achieved with a GRE sequence with an echo time of 15 msec, which allowed depiction of 70% of lesions that were smaller than 1.0 cm. The same GRE sequence was used in our current study, but fat suppression was applied to improve the conspicuity of surface lesions, and the section thickness was reduced from 10.0 mm to 6.0 mm to improve the detection of small metastases. We used the contrast agent ferucarbotran because it is administered as a bolus injection with relatively few and minor side effects (33). Also, because it is injected as a bolus, ferucarbotran provides the opportunity to obtain dynamic T1-weighted MR images immediately after injection (33). Although our imaging protocol included a series of T1-weighted MR images obtained in the first 2 minutes after ferucarbotran injection, only delayed T2-weighted GRE MR images were included in our analysis, and further studies will evaluate the added value provided by dynamic ferucarbotran-enhanced T1-weighted MR imaging.

In accordance with results of previous studies, all three of the techniques evaluated in our present study were successful for detection of lesions that were larger than 1.0 cm. All lesions that were larger than 2.0 cm were detected with all techniques, and only one, two, and six lesions in the 1.1–2.0-cm size range were not detected at SPIO-enhanced MR imaging, gadolinium-enhanced MR imaging, and thin-section helical CT, respectively. All lesions that were missed with all three techniques were 1.0 cm or smaller, and our results for the detection of small lesions compare favorably with those of earlier studies (1,5,13,20,24,28).

Nevertheless, since 16% (nine of 58) of our patients had evidence of previously undetected lesions at 4–6-month follow-up imaging, our results probably caused overestimation of the accuracy of preoperative imaging and should be regarded with a degree of caution. Regeneration of the residual liver after surgery is believed to produce a growth spurt in metastases in some patients (34), so it is not surprising that lesions that were invisible at preoperative imaging were well seen at follow-up imaging. To verify all true-positive lesions, particularly those of subcentimeter size, the pathologist sectioned the liver at 3-mm intervals, and we performed a meticulous comparison between the lesions identified at imaging and those identified at surgery with intraoperative US and histopathologic examination. Despite this verification, our follow-up results suggest that a proportion of small metastases remain undetected at imaging, intraoperative US, and surgery. Relative to our reference standard, however, SPIO-enhanced MR imaging depicted substantially more small lesions than did gadolinium-enhanced MR imaging and thin-section helical CT, and this result increased detection of small lesions by 10% and 19%, respectively.

Matsuo and colleagues (31) compared gadolinium- and SPIO-enhanced MR imaging for the detection of malignant liver tumors and found gadolinium-enhanced MR imaging to be the more sensitive technique. When only subcentimeter lesions were considered, the sensitivity values they achieved were considerably lower than those of our current study (55% vs 58% for gadolinium-enhanced MR imaging and 32% vs 68% for SPIO-enhanced MR imaging). One explanation for the difference between the results of the study of Matsuo et al and those of our study is that more than half of the patients (31 of 53) examined by Matsuo et al had underlying cirrhosis and almost two-thirds of their lesions (53 of 87) were hepatocellular carcinomas. Tang et al (35) have also demonstrated the superiority of gadolinium-enhanced MR imaging for the detection of small hepatocellular carcinomas. The difference between the results of that study and ours may also reflect refinements in the T2-weighted GRE MR sequence used in our study. Technique optimization is an important aspect of comparative studies, and we attempted to achieve this optimization for all three protocols evaluated in this study.

Researchers in several studies (17,18,28,36,37) have demonstrated a substantial increase in lesion-to-liver contrast after administration of SPIO, and these findings are further endorsed by the results of our study. At high confidence scores, only four subcentimeter lesions in three patients were depicted on gadolinium-enhanced MR images or CT images and were missed on SPIO-enhanced MR images, and all were visible at retrospective review. Conversely, 21 lesions that were clearly demonstrated on SPIO-enhanced MR images were not depicted on either gadolinium-enhanced MR images or CT images (n = 8) or on CT images only (n = 13), and only five were visible retrospectively.

With CT, considerably more false-positive interpretations were produced than were with either MR imaging technique, and nearly half of these were caused by the incorrect interpretation of benign lesions as metastases. This is a major limitation of CT, since a confident diagnosis of benign lesions that are actually malignant may result in denial of potentially curative surgery or more extensive resection in patients so affected. In an evaluation of thin-section helical CT for detection of small metastases, Haider and colleagues (26) also misclassified a substantial proportion of the lesions in their study as benign, and there was no improvement in lesion characterization at a thinner collimation. Gadolinium-enhanced MR imaging is particularly useful in characterization of lesions on the basis of their perfusion patterns, but cysts, hemangiomas, and metastases may all be hyperintense and indistinguishable after administration of SPIO.

To reflect clinical practice, we reviewed both gadolinium-enhanced MR images and SPIO-enhanced MR images in combination with unenhanced T1- and T2-weighted MR images. The half-Fourier rapid acquisition with relaxation enhancement MR sequence used in this study was invaluable for demonstration of the high water content associated with most cysts and hemangiomas, and this finding facilitated the correct diagnosis in all cases. In-phase and out-of-phase T1-weighted GRE MR images (chemical shift imaging) are necessary because they provide findings that aid in a definitive diagnosis of focal or diffuse fatty infiltration. This is particularly important in patients who are being considered for hepatic resection, since extensive fatty change may compromise hepatic function postoperatively. Nineteen of the patients in our current study had varying degrees of diffuse fatty change, and this was diagnosed at MR imaging in all.

False-positive findings with both MR imaging techniques were rare. In previous studies, most false-positive interpretations with SPIO-enhanced MR images have been attributed to the high signal intensity of vascular structures, but this accounted for only one false-positive lesion in our current study. Improvements in image quality with high vessel-to-liver contrast and observers with extensive experience in the interpretation of SPIO-enhanced MR images are likely to have contributed to the reduced the number of false-positive findings.

Our study had several limitations. A potential criticism of this study was the inclusion of patients with a large number of lesions, and inclusion of them may have biased our results in that it caused a clustering effect. We believe that this inclusion was valid, since surgical candidates with multiple lesions reflect those who are increasingly seen in clinical practice. Also, in the alternating free-response receiver operating characteristic model, multiple lesions on an image are regarded as independent events, so any bias is likely to be minimal; while the uninvolved liver tissue has a similar signal intensity in all patients, the signal intensity of metastases in a single patient often is variable. Further, the average size of lesions in our patients with more than seven lesions was smaller than that in our patients with fewer lesions, so interpretation in this group of patients would be expected to produce more errors. All our false-negative and false-positive lesions, however, were randomly distributed across all patients and both observers.

Also, in our patient with 21 lesions, all 21 were depicted on SPIO-enhanced MR images, and 19 of 21 were depicted on gadolinium-enhanced MR images and CT images. Our results also may have been biased because we had no patients without metastases. Since the observers were aware that all patients were surgical candidates, they were also aware that all patients were likely to have at least one metastasis, but they were unaware of the number and distribution of lesions, and this varied considerably in individual patients.

The validity of our findings would have been strengthened with the addition of more observers, but such an addition was precluded by the detailed and laborious nature of our image analysis. Given good to excellent interobserver agreement for all three techniques, however, it would be reasonable to expect similar results to be achieved by other experienced observers. We also used fixed delay times for each imaging phase after gadolinium-based contrast agent administration, and this may have resulted in a suboptimal arterial phase acquisition in some patients. This aspect of our gadolinium-enhanced MR imaging protocol has since been changed, and we now routinely perform a timing injection to determine the appropriate interval between the start of contrast agent injection and the start of image acquisition. A further limitation was the exclusion of dynamic T1-weighted MR images obtained after administration of SPIO from our analysis. This occurred because our study protocol was submitted for peer review and ethics approval at a time when we were unsure of the role of SPIO-enhanced T1-weighted MR imaging for lesion detection. We now regard T1- and T2-weighted MR sequences performed after administration of SPIO as complementary, and we believe that this combination may increase the accuracy of SPIO-enhanced MR imaging further. In the first few minutes after injection, the thinner effective section thickness with our T1-weighted 3D MR sequence is particularly helpful in the depiction of small lesions and reliably helps to distinguish them from adjacent vessels.

In conclusion, the detection of lesions that are 1.0 cm or smaller appears to be improved substantially with SPIO-enhanced MR imaging, which also is associated with an extremely low false-positive rate when it is combined with unenhanced T1- and T2-weighted MR sequences. On the basis of the results of this study, in patients who are eligible for liver resection, the standard protocol for liver imaging at our institution is now unenhanced in-phase and out-of-phase T1-weighted GRE and half-Fourier rapid acquisition with relaxation enhancement MR sequences followed by SPIO-enhanced T1-weighted fat-suppressed 3D GRE and T2-weighted GRE MR sequences. If a lesion has benign characteristics on half-Fourier rapid acquisition with relaxation enhancement images and a location that may alter the surgical approach, we obtain sequential gadolinium-enhanced T1-weighted fat-suppressed 3D GRE MR images (optimized from a timing injection) immediately after we obtain SPIO-enhanced MR images for more reliable lesion characterization.

	ACKNOWLEDGMENTS

 
The authors thank Sheila Boyes for assistance with the preparation of the manuscript and statistical results; Paul Arnold, BSc, for his assistance at the early stages of this project; Jane Bates, MPhil, Sarah Riley, MSc, and Stephen Wolstenholme, MSc, for intraoperative US comparison; and the staff of the MR imaging unit.

	FOOTNOTES

 

Abbreviations: CI = confidence interval • GRE = gradient echo • SPIO = superparamagnetic iron oxide • 3D = three-dimensional

Authors stated no financial relationship to disclose.

Author contributions: Guarantor of integrity of entire study, J.W.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; literature research, J.W.; clinical studies, J.W., P.J.R., J.A.G., S.D., J.P.A.L., G.J.T., J.I.W.; statistical analysis, J.W., D.W.; and manuscript editing, J.W., P.J.R., J.A.G., S.D., D.W., J.P.A.L., K.R.P., G.J.T.

	References

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION References

 

1. Kuszyk BS, Bluemke DA, Urban BA, et al. Portal-phase contrast-enhanced helical CT for the detection of malignant hepatic tumors: sensitivity based on comparison with intra-operative and pathologic findings. AJR Am J Roentgenol 1996;166:91–95.[Abstract/Free Full Text]
2. Valls C, Lopez E, Gumà A, et al. Helical CT versus CT arterial portography in the detection of hepatic metastasis of colorectal carcinoma. AJR Am J Roentgenol 1998;170:1341–1347.[Abstract/Free Full Text]
3. Ward J, Naik KS, Guthrie JA, Wilson D, Robinson PJ. Hepatic lesion detection: comparison of MR imaging after the administration of superparamagnetic iron oxide with dual-phase CT by using alternative–free response receiver operating characteristic analysis. Radiology 1999;210:459–466.[Abstract/Free Full Text]
4. Scott DJ, Guthrie JA, Arnold P, et al. Dual phase helical CT versus portal venous phase CT for the detection of colorectal liver metastases: correlation with intra-operative sonography, surgical and pathological findings. Clin Radiol 2001;56:235–242.[CrossRef][Medline]
5. Valls C, Andía E, Sánchez A, et al. Hepatic metastases from colorectal cancer: preoperative detection and assessment of resectability with helical CT. Radiology 2001;218:55–60.[Abstract/Free Full Text]
6. Furuhata T, Okita K, Tsuruma T, et al. Efficacy of SPIO-MR imaging in the diagnosis of liver metastases from colorectal carcinomas. Dig Surg 2003;20:321–325.[CrossRef][Medline]
7. Oi H, Murakami T, Kim T, et al. Dynamic MR imaging and early-phase helical CT for detecting small intrahepatic metastases of hepatocellular carcinoma. AJR Am J Roentgenol 1996;166:369–374.[Abstract/Free F