| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Gastrointestinal Imaging |
1 From the Department of Clinical Radiology (J.W., J.A.G., D.W., P.A., P.J.R.), Hepatobiliary and Transplantation Unit (J.P.L., G.J.T.), and Department of Histopathology (J.I.W.), St James’s University Hospital, Beckett St, Leeds LS9 7TF, England. Received March 28, 2002; revision requested June 10; final revision received December 18; accepted January 23, 2003. Address correspondence to J.W. (e-mail: janice.ward@leedsth.nhs.uk).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Thirty-one patients with colorectal hepatic metastases underwent T1-weighted gradient-echo (GRE) and T2-weighted fast spin-echo (FSE) MR imaging before and after SPIO enhancement. Four sequences were optimized for lesion detection: T2-weighted FSE, multiecho data image combination (MEDIC), T2-weighted GRE with an 11-msec echo time (TE), and T2-weighted GRE with a 15-msec TE. Images were reviewed independently by three blinded observers. The accuracy of each sequence was measured by using alternative free-response receiver operating characteristic analysis. All results were correlated with findings at surgery, intraoperative ultrasonography, or histopathologic examination. Differences between the mean results of the three observers were measured by using the Student t test.
RESULTS: Postcontrast T2-weighted GRE sequences were the most accurate and were significantly superior to postcontrast T2-weighted FSE and unenhanced sequences alone (P < .05). For all lesions that were malignant or smaller than 1 cm, respectively, mean accuracies of postcontrast sequences were 0.082 and 0.64 for T2-weighted FSE, 0.90 and 0.78 for MEDIC, 0.92 and 0.80 for GRE with an 11-msec TE, 0.93 and 0.82 for GRE with a 15-msec TE, and 0.81 and 0.62 for unenhanced sequences.
CONCLUSION: Optimized SPIO-enhanced T2-weighted GRE combined with unenhanced T2-weighted FSE MR sequences were the most sensitive. Breath-hold FSE postcontrast sequences offer no improvement in sensitivity compared with unenhanced sequences alone.
© RSNA, 2003
Index terms: Iron • Liver neoplasms, metastases, 761.33 • Liver neoplasms, MR, 761.121411, 761.121412, 761. 121415, 761.121417, 761.12143 • Magnetic resonance (MR), comparative studies, 761.121411, 761.121412, 761. 121415, 761.121417, 761.12143 • Magnetic resonance (MR), contrast enhancement, 761.12143 • Magnetic resonance (MR), contrast media, 761.12143
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
MR images are best acquired during breath holding to eliminate motion artifact, and with high-performance gradient systems and phased-array body coils, acquisition of high-quality breath-hold T1- and T2-weighted MR images is now routine. However, most breath-hold T2-weighted MR imaging is based on FSE sequences that use long echo trains and short echo spacing, and it is known that increasing echo train length not only increases magnetization transfer effects but also further reduces the SI loss produced by SPIO. GRE sequences, which also allow the acquisition of T2-weighted MR images during breath holding, are particularly sensitive to the effects of SPIO because they lack a 180° refocusing pulse, so the effect of susceptibility from local field inhomogeneities is increased. However, GRE parameters must be optimized to minimize noise and motion-related artifact, which are more pronounced with the longer echo times (TEs) required for T2-weighted MR imaging. Moreover, susceptibility is also dependent on field strength, so nonoptimized sequence parameters may result in such a profound SI loss at high field strengths that small lesions may be obscured by a "blooming" effect from the diffuse SI loss in the adjacent parenchyma.
To our knowledge, there is only one previous study in which breath-hold T2-weighted MR sequences were evaluated after SPIO enhancement for the detection of hepatic metastases, and GRE sequences were not included in the analysis (11). The purpose of the current study was to compare the accuracy of four breath-hold MR imaging sequences by using optimized parameters to establish the most effective SPIO-enhanced sequence for the detection of colorectal hepatic metastases.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
The final study group comprised 31 patients (15 men, 16 women; mean age, 59 years; age range, 28–75 years) who underwent surgical exploration with intraoperative ultrasonography (US). Surgical resection was subsequently attempted in patients whose disease was confined to the liver and who were considered to have sufficient normal liver to maintain liver function postoperatively. Number, diameter, and distribution of lesions were not limiting factors. Left hepatectomy was performed in four patients, right hepatectomy in 10, right trisegmentectomy in six, and left trisegmentectomy in four. In eight of these 24 patients, additional small metastases were removed from the residual liver. Five patients underwent more limited resection: bilateral resection of segments II, III, V, and VI in one, resection of segments V and VI in one, resection of segments VI and VII in one, removal of multiple metastases in the right lobe in one, and removal of a solitary metastasis from the remnant liver in one who had undergone previous left hepatectomy. In these 29 patients, all lesions detected with intraoperative US were removed and confirmed at histopathologic examination. One additional lesion, which was identified by means of surgical palpation and confirmed histologically, was not detected with US. In addition, one patient underwent radiofrequency ablation, and in another patient, multiple peritoneal deposits not detected preoperatively precluded resection. In both cases, detailed evaluation of the liver with intraoperative US confirmed MR imaging findings.
Local ethics committee approval was granted, and informed written consent was obtained from each patient prior to entry into the study.
MR Imaging
All MR imaging was performed at a field strength of 1.5 T (Symphony; Siemens, Erlangen, Germany) by using a body phased-array coil. The unenhanced sequences performed prior to SPIO injection were T2-weighted FSE (repetition time msec/TE msec, 2,650/88) with an echo train length of 29, an echo spacing of 8 msec, one signal acquired, and a matrix size of 172 x 256; and T1-weighted GRE in-phase (161/4.7, 75° flip angle) and opposed-phase (144/2.4 msec; 75° flip angle) sequences with one signal acquired and a matrix size of 134 x 256.
After SPIO enhancement, the same T2-weighted FSE sequence was performed, followed by the multiecho data image combination (MEDIC) GRE sequence (188/27, 30° flip angle, one signal acquired, matrix size of 102 x 256) and two T2-weighted GRE sequences (165/11, 15; 30° flip angle; one signal acquired; matrix size of 115 x 256). For all MR sequences, images were acquired in the transverse plane with a section thickness of 8–10 mm and a 10% gap, and the field of view was 30–40 cm, depending on patient size; in each patient, these factors were the same for all sequences before and after SPIO administration.
All MR images were acquired during breath holding, and the following strategies were used to minimize motion-related artifact and produce optimum images. For the T2-weighted FSE sequence, spectral fat suppression and presaturation bands positioned above and below the imaging volume were applied to minimize motion-related artifact. The MEDIC sequence involves the acquisition of six full data sets of GRE images at different TEs, which are then combined to produce an effective TE that is calculated from the average raw data of the six. This approach results in increased signal-to-noise ratio and reduced sensitivity to flow artifacts; motion-related artifacts were further reduced by the application of flow compensation gradients.
For the T2-weighted GRE sequences, a reduced bandwidth of 65 Hz per pixel was used to minimize noise and increase signal-to-noise ratio. To minimize flow artifact, we used presaturation bands positioned above and below the imaging volume for the GRE sequence with a TE of 11 msec (TE too short for application of additional flow compensation gradients) and flow compensation for the GRE sequence with a TE of 15 msec. For the MEDIC and T2-weighted GRE sequences, a flip angle of 30° was chosen to maximize the SI from metastases on the basis of the Ernst angle of metastases at a field strength of 1.5 T and a T1 of 1,000 msec. For all sequences, a rectangular field of view was used to improve spatial resolution by maximizing the number of phase-encoding lines for a given acquisition time. In all cases, multiple acquisitions were required to encompass the entire liver; 11, three, six, and six sections were acquired during a breath-hold period of 19–21 seconds for the FSE, MEDIC, and two T2-weighted GRE sequences, respectively.
In all patients, the SPIO agent ferumoxides (Endorem; Guerbet, Roissy, France) was administered at a dose of 7.5 µmol of iron per kilogram of body weight and was diluted in 100 mL of 5% glucose solution and infused over 30 minutes. The dose of ferumoxides was based on that in a previous study in which it was shown that a dose of 7.5 µmol/kg produced lesion-to-liver contrast-to-noise ratio and image quality equivalent to that of the higher 15 µmol/kg dose that is recommended by the supplier (12).
Qualitative Analysis
Five sets of images were analyzed for each patient: unenhanced images (T1-weighted in-phase and opposed-phase GRE and T2-weighted FSE) and SPIO-enhanced T2-weighted FSE, MEDIC, and T2-weighted GRE images with TEs of 11 and 15 msec. The hard-copy images obtained with each sequence were printed with patient details and sequence parameters removed to minimize observer bias. The images were then ordered randomly so that different images obtained in the same patient were not presented consecutively, and the images were viewed independently by three observers who were unaware of the results of the other imaging sequences, the results of the other observers, and the findings at surgery and histopathologic examination. Because it may be difficult to distinguish benign and malignant nonhepatocellular lesions after SPIO administration, the unenhanced FSE images were reviewed with SPIO-enhanced images to aid in lesion characterization. All three observers had more than 5 years of experience in MR imaging of the liver.
For each sequence, each observer recorded the presence of all lesions on the basis of a four-point confidence scale: 1 = probably not a lesion, 2 = possible lesion, 3 = probable lesion, and 4 = definite lesion. At the time of scoring, each observer was also asked to make a subjective assessment of image quality for each sequence on the basis of lesion-to-liver contrast and level of artifact. For each sequence, quality scores were assigned: 1 = poor, 2 = fair, 3 = good, and 4 = excellent. Poor lesion-to-liver contrast was defined as an uncertain lesion with SI similar to that of background liver; fair, lesion seen but only slightly hyperintense relative to background liver; good, lesion moderately hyperintense relative to background liver; and excellent, lesion markedly hyperintense relative to background liver. Level of artifact was defined as poor, lesion visualization severely impaired by marked ghosting and/or blurring; fair, lesion visible but partly obscured by ghosting and/or blurring; good, lesion visualization not affected by minimal ghosting and/or blurring; and excellent, lesion sharply defined with no ghosting and/or blurring.
To facilitate lesion identification and accurate correlation between the scored lesions and those confirmed at surgery and histopathologic examination, a grid reference was used at the time of scoring. The image number and segmental location and size of each lesion were recorded. Lesions were also categorized as benign, malignant, or indeterminate on the basis of relative SI on unenhanced FSE images, with benign indicating extremely high SI, similar to that of cerebrospinal fluid; indeterminate, moderate to high SI but less than that of cerebrospinal fluid; and malignant, slight to moderately high SI relative to background liver or nonvisualization of the lesion. At the time of surgery, all of the lesions identified by the surgeon and those identified with intraoperative US were carefully correlated with MR images by one author (P.A.); to avoid bias, this was a different author from those who performed the blinded reading. Further comparison of histopathologic findings and MR images was performed after dissection of the resected specimen, which was sectioned in the transverse plane at 10-mm intervals by an experienced pathologist (J.I.W.). All surgery was performed by one of two hepatobiliary surgeons who were aware of the findings of preoperative imaging, and intraoperative US was performed by one of three experienced sonographers in conjunction with the surgeons. All MR examinations had been performed between 1 and 8 weeks (mean, 4.6 weeks) before surgery. To find small lesions that may have been missed at preoperative imaging, surgical inspection, and intraoperative US, follow-up CT or MR images obtained at least 6 months after surgery were also reviewed by one author (J.W.), who then reevaluated the preoperative images to see if any of the new lesions were visible retrospectively on the initial images.
Once the location of all true-positive lesions was established, an assessment of all false-positive findings with confidence levels of 3 or 4 was undertaken by the same author (J.W.) by means of review of the combined MR images together with all other imaging and histopathologic findings.
Statistical Evaluation of Qualitative Analysis
Alternative free-response receiver operating characteristic (ROC) analysis of all lesions was performed for each sequence and each observer. By using the McNeil and Hanley method (13), ROC curves were calculated by plotting the true-positive fraction against the likelihood of obtaining a false-positive finding (ie, one or more false-positive lesions) at each confidence level. The area under each curve was used to compare the overall performance (accuracy) of each technique and each observer. The sensitivities for each observer and each technique were also calculated by using only those lesions awarded confidence scores of 3 or 4, and the McNemar test was used to assess statistical significance. The Student t test was used to assess the statistical significance of any differences between the mean results of the three observers for each sequence, mean sensitivity, and areas under the ROC curves. Separate analyses of all malignant lesions, malignant lesions 1 cm and larger, and malignant lesions smaller than 1 cm were undertaken. Lesions were categorized by size on the basis of the final standard of reference (ie, intraoperative US or histopathologic findings). Analysis of all false-negative and false-positive (confidence levels of 3 or 4) observations was also undertaken.
statistics were used to assess interobserver variability for lesion detection with each sequence. A value of 0.40 or lower was considered to indicate poor correlation; 0.41–0.75, good; and 0.76 and higher, excellent.
Statistical Evaluation of Quantitative Analysis
User-defined 1–2-cm regions of interest were positioned in the lesions, and adjacent liver parenchyma and background noise (anterior to the lesion in the phase-encoding direction) SI measurements were obtained for each sequence. All regions of interest were placed by the same observer. Except for slight differences in position owing to variations in respiration or patient movement, identical regions of interest were used for each sequence. The regions were placed to encompass as much of the lesion as possible, while avoiding areas of necrosis or scarring within larger lesions. Lesions smaller than 1 cm were excluded from quantitative analysis because it was difficult to avoid the inclusion of adjacent liver within the region of interest.
For each postcontrast image, lesion-to-liver contrast-to-noise ratios were calculated according to the following equation: SIlesion - SIliver/SInoise. On the basis of SI measurements on T2-weighted MR images obtained before and after SPIO administration, the percentage SI change in normal liver was calculated according to the following equation: postcontrast SI - precontrast SI/precontrast SI x 100.
Data were expressed as the mean ± SD. The significance of differences between sequences was assessed by using the paired Student t test.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Accuracy
For each MR sequence, the areas under the ROC curves for each observer and the mean of the three observers and 95% CIs for the differences in the mean ROC values are shown in Figures 1 and 2. The results for all malignant lesions and for malignant lesions smaller than 1 cm are shown in Figures 1 and 2, respectively. Of 101 lesions detected, 63 were observed and scored by all three observers with all pulse sequences. In 12 patients, all lesions were found by all observers with all pulse sequences, but in the other 19 patients, at least one lesion was missed by at least one observer with at least one sequence.
|
|
Sensitivity
For each MR technique, the sensitivity for each observer, the mean sensitivity of the three observers, and 95% CIs for the difference in the mean sensitivities are shown in Figures 3 and 4. The results for all malignant lesions are shown in Figure 3, and the results for malignant lesions smaller than 1 cm are shown in Figure 4.
|
|
|
|
|
|
|
|
On the basis of the MR imaging findings, the surgical approach was changed in three other patients. In one patient with a 7-week delay between MR imaging and surgery, intraoperative US showed a more extensive tumor involving the right hepatic vein than had been demonstrated on MR images, and a planned segmental excision was extended to a right hepatectomy. In one patient with a fatty liver, a false-positive finding with intraoperative US resulted in removal of tissue from the left lobe in addition to the planned right hepatectomy, and in one additional patient, peritoneal and lymph node deposits precluded resection. Intraoperative US and surgical palpation were used to confirm the MR imaging findings of the liver in this patient.
In 25 patients, there was no evidence of additional hepatic metastases on CT or MR images obtained at least 6 months after surgery. One patient developed bone metastases 6 weeks after surgery and died 3 months later without further follow-up. In the other five patients, hepatic lesions were found on CT scans obtained 4–6 months after resection, so it is likely that they were present at the time of surgery and missed at imaging and at surgical inspection with intraoperative US. None of the additional lesions were visible at retrospective review of the initial MR images.
False-Positive Findings
At confidence levels of 3 or 4, 1,117 true-positive interpretations and 37 false-positive interpretations (nine at unenhanced MR imaging, 13 at SPIO-enhanced FSE imaging, five at MEDIC imaging, six at GRE imaging with a TE of 11 msec, and four at GRE imaging with a TE of 15 msec) were recorded to give an overall false-positive rate of 3.3%. SPIO-enhanced FSE imaging produced the highest number of false-positive findings (6.5%), and GRE with a TE of 15 msec produced the lowest (1.6%). None of the false-positive "lesions" was identified by more than one observer with more than one sequence. At retrospective review, 32 of the 37 false-positive lesions (86%) were smaller than 1 cm; 26 of 37 (70%) were attributed to vessels. The remaining 30% were attributed to partial volume effects from the bowel, stomach, and gallbladder (seven of 37); perfusion effects (one of 37); and extracapsular fat (three of 37).
Image Quality
The level of artifact on SPIO-enhanced FSE images was judged to be higher than that on any other type of SPIO-enhanced image. Level of artifact was rated as poor or fair in 15 of 31 cases on FSE images, whereas the artifact level was considered good or excellent in 28, 27, and 26 cases on GRE images with a TE of 15 msec, MEDIC images, and GRE images with a TE of 11 msec, respectively. The most severe artifacts on FSE images resulted from motion-induced ghosting from the stomach, aorta, and subcutaneous fat due to inadequate suppression of fat immediately adjacent to the phased-array coil. With the other three sequences, images were suboptimal in three patients who had difficulty suspending respiration, and flow artifact was prominent in the remainder.
Liver-to-lesion contrast was judged best on GRE images with a TE of 15 msec in 18 of 31 patients, on MEDIC images in three of 31, and on GRE images with a TE of 11 msec in one of 31. The contrast level was judged to be equally high on MEDIC and GRE images with a TE of 15 msec in four patients, on both types of GRE images in three patients, and on both types of GRE images and MEDIC images in two patients. Liver-to-lesion contrast was never judged to be highest on SPIO-enhanced FSE images.
Interobserver Variability
The values for lesion detection for each observer with each technique for all malignant lesions and malignant lesions smaller than 1 cm are shown in the Table. For all malignant lesions, good to excellent agreement was obtained for all sequences. When only lesions smaller than 1 cm were analyzed, although all observers achieved similar sensitivities with each sequence, the actual lesions recorded by each observer varied considerably. The best agreement was achieved with both GRE sequences, however, which was good to excellent for all observers.
|
|
|
|
|
Mean liver-to-lesion contrast-to-noise ratios were 4.6 ± 2.5, 11.6 ± 6.1, 22.9 ± 10.7, and 25.4 ± 10.2 for FSE, MEDIC, GRE with a TE of 11 msec, and GRE with a TE of 15 msec sequences, respectively. Each improvement in contrast-to-noise ratio achieved statistical significance (P < .01 for all).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
In accordance with other studies, all but one of the lesions not detected by any observer on any image in the present study were smaller than 1 cm. Although one 1.2-cm lesion was not scored by any observer on any image, it was appreciated retrospectively but only on the GRE images. The choice of pulse sequence has a major influence on the diagnostic effectiveness of SPIO, and this was reflected in our current results. Even for lesions 1 cm and larger, mean sensitivities ranged from 83.0% on unenhanced and SPIO-enhanced FSE images to 95.0% on GRE images with a TE of 15 msec.
For FSE images, we used fat suppression and parallel saturation bands to minimize motion-related artifact, but there are no strategies that effectively reduce magnetization transfer or increase susceptibility effects in multishot FSE sequences with long echo train lengths. Consequently, our breath-hold FSE sequence was always the least sensitive and accurate of our SPIO-enhanced sequences, and it was not significantly better than unenhanced images. Previous investigators who reported favorable results with FSE sequences after SPIO enhancement have used non–breath-hold sequences with relatively short echo train lengths (none exceeded an echo train length of nine) (5,9,11,14). When Schwartz et al (15) evaluated an FSE sequence with an echo train length of 16, they found no improvement in lesion-to-liver contrast-to-noise ratio after SPIO enhancement.
To our knowledge, only one published study (11) involved the evaluation of a breath-hold FSE sequence after SPIO enhancement in patients with hepatic metastases. Abe et al (11) performed a quantitative evaluation of conventional spin echo; non–breath-hold FSE; single-shot breath-hold FSE; half-Fourier rapid acquisition with relaxation enhancement, or RARE; and breath-hold echo-planar MR sequences. Although their echo-planar MR sequences achieved the greatest decrease in liver SI after SPIO enhancement, they were not considered suitable for clinical imaging because they were severely degraded by image distortion and chemical shift artifact. All lesions were more conspicuous after SPIO enhancement on all images, but the decrease in liver SI and lesion-to-liver contrast-to-noise ratio after SPIO enhancement was least for half-Fourier RARE. Nevertheless, the authors recommend half-Fourier RARE as a useful alternative to non–breath-hold imaging after SPIO enhancement because the sequence was virtually unaffected by motion artifact. Because the half-Fourier RARE sequence is a single-section technique, it is relatively unaffected by magnetization transfer contrast, so it may be expected that lesion-to-liver contrast would be higher on half-Fourier RARE images than on the multishot FSE images obtained in our study. In our experience, however, the multishot FSE sequence performs better after SPIO enhancement because it has a shorter echo train length and fewer 180° refocusing pulses with longer echo spacing, so there is more SPIO enhancement.
We attempted to avoid a compromise between SI loss and reduced motion artifact by using optimized GRE sequences to maximize the effect of SPIO enhancement while reducing motion artifact and increasing signal-to-noise ratio. In our study, not only was the GRE sequence with a TE of 15 msec the most accurate and sensitive, it also achieved the highest mean percentage SI change and the highest lesion-to-liver contrast-to-noise ratio. In most patients, it was also judged to be the best sequence in terms of level of artifact and lesion-to-liver contrast.
Although T2-weighted GRE sequences have been shown to be more sensitive than FSE or conventional spin-echo MR sequences after SPIO enhancement, sequence parameters must be optimized to minimize noise and flow-related artifact. The application of flow compensation or parallel saturation bands effectively reduced flow artifact on our GRE images, and a reduced sampling bandwidth of 65 Hz per pixel resulted in a substantially improved signal-to-noise ratio. At a TE of 15 msec, the SI of normal liver is virtually eliminated after SPIO enhancement, so we were able to optimize lesion-to-liver contrast by using a flip angle that maximized the SI for metastases. The maximum SI of any tissue occurs at a specific flip angle, known as the Ernst angle (16). Our flip angle of 30° was based on the Ernst angle for hepatic metastases at a field strength of 1.5 T and a T1 of 1,000 msec (17). We also evaluated the same GRE sequence at the shorter TE of 11 msec. None of the lesions missed on GRE images with a TE of 15 msec were seen on GRE images with a TE of 11 msec, and GRE imaging with a TE of 11 msec was marginally less sensitive for the detection of small lesions.
The optimum dose of ferumoxides is unknown, and different doses have been approved for use in the United States (10 µmol/kg) and Europe (15 µmol/kg). In the only published study, to our knowledge, on the evaluation of the effect of dose on lesion detection (by using a rat tumor model), greater SI loss in normal liver was observed with increasing doses of SPIO (18). We previously compared the effectiveness of the recommended dose of 15 µmol/kg and a reduced dose of 7.5 µmol/kg in patients with colorectal hepatic metastases (12,19). We found no significant difference in image quality or in the number of detected lesions when identical imaging parameters were used. We also evaluated lesion-to-liver contrast-to-noise ratio at the lower dose of SPIO by using the same GRE sequence at different TEs. All lesion-to-liver contrast-to-noise ratios increased as TE was lengthened, and no lesion was obscured by blooming. As TEs were increased beyond 15 msec, however, there was a decrease in the apparent size of small lesions, which is why we did not include a GRE sequence with a longer TE in the present study. By using a dose of 7.5 µmol/kg, we found a GRE sequence with a TE of 15 msec to be optimal. At a higher dose there is greater liver SI loss, however, which may lead to more pronounced blooming, so it would be reasonable to expect a GRE sequence with a shorter TE to be more effective at the higher dose of 10 or 15 µmol/kg.
The MEDIC sequence was evaluated because it had the potential to offer an acquisition with a high signal-to-noise ratio, minimum motion artifact, and strong T2* weighting. However, the increase in signal-to-noise ratio associated with multiple echoes was partly offset by the increased bandwidth of 195 Hz per pixel, which was necessary to obtain an acceptable number of images in a breath-hold period. Despite the high bandwidth, we were able to obtain only three sections per breath hold compared with six per breath hold for the other GRE sequences.
Although several studies performed with MR imaging strategies that now would not be considered state-of-the-art have yielded sensitivities and accuracies similar to those achieved in our study, most have been limited by lack of complete correlation (6,7,9), small sample size (particularly of lesions smaller than 1 cm) (7,14,20), inhomogeneous patient population (7,9,20), or exclusion of small lesions (4). In contrast, we studied a homogeneous patient population with a substantial number of small lesions (27%), and by using alternative free-response ROC methodology, which allows an observer response for all of the lesions present, we were able to analyze every lesion in every patient. In allowing a plurality of position-dependent decision analysis variables per image, it is assumed in the ROC model that multiple lesions on an image are independent events. We believe that this assumption is a valid one to make for hepatic metastases because the uninvolved liver tissue shows similar SI in all patients, but the conspicuity of metastases is variable, even when considering lesions of the same size in a single patient.
In the current study, we observed no pattern or clustering of false-positive or false-negative lesions for any technique or any patient. Our six false-negative lesions occurred in six patients, and our false-positive lesions were randomly distributed across the patients and observers. We also attempted to verify every lesion by performing a meticulous correlation between MR imaging findings and surgical findings with intraoperative US and histopathologic results. Despite our rigorous methodology, however, it is likely that our results still underestimate the problem of metastases in the 1-mm size range, since five of our 31 patients had evidence of previously undetected lesions at follow-up imaging performed within 4–6 months of surgery.
Fat suppression is known to improve the conspicuity of surface lesions at MR imaging, but it was not applied to the GRE sequences in the present study because it would have resulted in a substantial reduction in the number of sections available for a single breath hold. Given the number of sequences being evaluated, we considered that this would have resulted in a prohibited number of breath holds per patient. On the basis of the results of this study, it is now our practice to perform only the GRE sequence with a TE of 15 msec after SPIO enhancement. In an attempt to improve the detection of surface lesions and lesions smaller than 1 cm, we have now further refined the sequence by reducing the section thickness from 10 to 6 mm and applying fat suppression.
False-positive findings were highest on SPIO-enhanced FSE images (6.5%) and lowest on GRE images with a TE of 15 msec (1.6%). In accordance with findings in other studies, most of our false-positive findings were attributed to the high SI of vascular structures relative to the reduced SI of liver parenchyma (10,21). However, only four false-positive lesions were recorded on GRE images with a TE of 15 msec, they were all recorded by single observers, and only one was attributed to a vessel. Most vessels were easily identified as tubular structures because of high vessel-to-liver contrast, which was highest on GRE images with a TE of 15 msec.
We consider unenhanced images to be an essential component of the SPIO-enhanced MR examination. We routinely obtain in-phase and opposed-phase T1-weighted MR images because they provide an absolute diagnosis of fatty infiltration within the liver. Focal or diffuse fatty change is a feature of some patients with hepatic metastases, particularly following chemotherapy, and since extensive change may compromise hepatic function postoperatively, this information is valuable for surgical planning. Unenhanced FSE images are necessary for tumor characterization. Coexistent benign and malignant lesions are common, and while failure to detect small benign lesions does not alter patient care, distinguishing benign lesions from metastases is as important as lesion detection in determining a patient’s suitability for surgery. The ability of SPIO enhancement to help distinguish benign nonhepatocellular lesions from metastases is extremely limited because cysts, hemangiomas, and metastases may all be highly hyperintense against the reduced SI of background liver after SPIO enhancement. However, most lesions can be correctly characterized by a comparison of pre- and postcontrast images. In this study, all three observers correctly characterized all the malignant lesions identified with SPIO enhancement, and at least one observer correctly classified all 17 benign lesions by means of combined review of postcontrast images and unenhanced FSE images. Indeterminate lesions can be characterized at laparotomy or at dynamic gadolinium-enhanced MR imaging, which can be performed immediately after SPIO-enhanced MR imaging (22–24).
In conclusion, breath-hold SPIO-enhanced MR imaging combined with unenhanced T2-weighted FSE MR imaging is significantly more sensitive than is unenhanced MR imaging. When compared with results in previous studies, the sensitivity for detection of metastases smaller than 1 cm appears to be improved substantially with SPIO-enhanced MR imaging by using high-performance gradients and phased-array technology, but choice of pulse sequence is of critical importance. At 1.5 T, an optimized GRE MR sequence with a TE of 15 msec was the most sensitive breath-hold sequence with SPIO enhancement at a dose of 7.5 µmol/kg. Enhanced breath-hold FSE sequences are not recommended because they offer no improvement in sensitivity, and they are associated with a higher false-positive rate and more pronounced artifact.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Author contributions: Guarantor of integrity of entire study, J.W.; study concepts and design, J.W., P.J.R., J.A.G., D.W.; literature research, J.W.; clinical studies, J.W., P.J.R., J.A.G., J.P.L., G.J.T., J.I.W.; data acquisition, J.W., J.A.G., P.A., P.J.R.; data analysis/interpretation, J.W.; statistical analysis, J.W., D.W.; manuscript preparation and definition of intellectual content, J.W.; manuscript editing, P.J.R., J.A.G., D.W., J.P.L., G.J.T., J.I.W.; manuscript revision/review and final version approval, J.W.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  J. Ward, J. A. Guthrie, M. B. Sheridan, S. Boyes, J. T. Smith, D. Wilson, J. I. Wyatt, D. Treanor, and P. J. Robinson Sinusoidal Obstructive Syndrome Diagnosed With Superparamagnetic Iron Oxide-Enhanced Magnetic Resonance Imaging in Patients With Chemotherapy-Treated Colorectal Liver Metastases J. Clin. Oncol., September 10, 2008; 26(26): 4304 - 4310. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. Charnsangavej, B. Clary, Y. Fong, A. Grothey, T. M. Pawlik, and M. A. Choti Selection of Patients for Resection of Hepatic Colorectal Metastases: Expert Consensus Statement Ann. Surg. Oncol., October 1, 2006; 13(10): 1261 - 1268. [Full Text] [PDF]
|
|
|
|
 |
|
 M.-J. Kim, J. H. Kim, J. Y. Choi, S. H. Park, J.-J. Chung, K. W. Kim, and D. G. Mitchell Optimal TE for SPIO-enhanced gradient-recalled echo MRI for the detection of focal hepatic lesions. Am. J. Roentgenol., September 1, 2006; 187(3): W255 - W266. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Ward, P. J. Robinson, J. A. Guthrie, S. Downing, D. Wilson, J. P. A. Lodge, K. R. Prasad, G. J. Toogood, and J. I. Wyatt Liver Metastases in Candidates for Hepatic Resection: Comparison of Helical CT and Gadolinium- and SPIO-enhanced MR Imaging Radiology, October 1, 2005; 237(1): 170 - 180. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Bipat, M. S. van Leeuwen, E. F. I. Comans, M. E. J. Pijl, P. M. M. Bossuyt, A. H. Zwinderman, and J. Stoker Colorectal Li |
