| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Special Report |
1 From the Departments of Radiology (M.E.J.P., J.D., M.N.J.M.W., J.L.B.), Medical Statistics (H.C.v.H.), and Surgery (R.A.E.M.T.), Leiden University Medical Center, Albinusdreef 2, 2333 ZA Leiden, the Netherlands. Received February 3, 2003; revision requested April 22; final revision received November 9; accepted November 10. Address correspondence to M.E.J.P. (e-mail: m.e.j.pijl@lumc.nl).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: The SIs of 48 liver metastases (originating from colorectal cancer) in 20 patients, the surrounding liver parenchyma, and the background noise were measured on T2-weighted MR images. All 2000 and 2001 issues of the American Journal of Roentgenology, the Journal of Magnetic Resonance Imaging, Magnetic Resonance Imaging, and Radiology were searched for articles describing quantitative analyses. SDNRs were calculated by using formulas from these articles and various region-of-interest (ROI) locations to measure metastasis and background noise SIs. The Wilcoxon signed rank test was used to compare the various SDNR calculations.
RESULTS: In 34 articles in which quantitative analyses of focal masses are described, the reported SDNRs were calculated with four different formulas. The SDNRs for our study material calculated with the four formulas reported in the literature differed grossly in both number and unit. The SDNRs for ROIs encompassing the entire metastasis differed significantly (P = .034) from the SDNRs for ROIs in a homogeneous area of the metastasis margin. Differences in SDNRs between various noise ROI locations were significant (P < .022).
CONCLUSION: Slight changes in the variables of quantitative analysis of focal masses had marked effects on reported SDNRs. To overcome these effects, the use of a standardized method involving one formula, a lesion ROI in a homogeneous area at the metastasis margin, and a background noise ROI along the phase-encoding axis in the air (including systematic noise) is proposed for the quantitative analysis of findings on magnitude MR images.
© RSNA, 2004
Index terms: Data analysis • Liver neoplasms, 75.33, 76.32, 76.33 • Liver neoplasms, metastases, 75.33, 76.33 • Liver neoplasms, MR, 76.121411, 76.121416, 76.12146 • Magnetic resonance (MR), tissue characterization, 76.12146
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
The purpose of this study was to assess the effects of changing analytic method variables on SDNRs for the contrast between lesions and background organs depicted on MR images and to propose a standardized analytic method for the quantitative analysis of focal masses seen at MR imaging.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
For 13 patients, there was no standard of reference because only hemangiomas were diagnosed (in one patient) or because of advanced disease (extrahepatic tumor or end-stage liver involvement in 12 patients). Thus, 20 patients (15 men, five women; mean age, 59 years; age range, 38–74 years) were included in this study. The mean age of both the women (age range, 38–72 years) and the men (age range, 45–74 years) was 59 years.
The primary malignancy was colorectal adenocarcinoma in all patients, each of whom had at least one histologically proved metastasis to the liver. The diagnosis of liver metastases at the time of referral was established on the basis of an increased serum level of carcinoembryonic antigen in 11 patients and was supported by ultrasonographic (US) findings in eight and by computed tomographic (CT) findings in three patients. The diagnosis of liver metastasis was established on the basis of perioperative findings (synchronous liver metastases) at colorectal surgery in seven patients. In one patient each, the referral diagnosis was established with US only or CT only. In none of the patients was a relevant comorbidity of the liver, such as cirrhosis or steatosis, diagnosed. The median interval between preoperative MR imaging and surgery was 25 days (range, 3–124 days).
MR Imaging
For this study, we used only a non– fat-suppressed respiratory-triggered T2-weighted fast spin-echo MR imaging sequence, which was performed with a 1.5-T system (Gyroscan NT 15; Philips Medical Systems, Best, the Netherlands) by using a quadrature body coil for signal transmission and reception. The effective echo time, echo spacing, and echo train length were 120 msec, 14 msec, and 16, respectively. Respiratory triggering was used, so the repetition time depended on the respiratory frequency (median repetition time, 2,100 msec; range, 1,750–2,500 msec). A belt placed around the upper abdomen generated the signal that was used for respiratory triggering.
The patients were not given special breathing instructions for the MR imaging examinations. Image acquisition was started just before end expiration, allowing data acquisition during a period of reduced respiratory motion. Eighteen 10-mm-thick sections with a 1-mm intersection gap in the transverse orientation and anteroposterior phase-encoding direction were obtained with four acquired signals. The matrix size was 202 x 256. The field of view was 375 mm, with a rectangular field of view of 80%; these parameters were balanced to achieve an optimal combination of spatial resolution and acquisition time and to generate a proper amount of air in the field of view to place the background noise ROI.
Standard of Reference
The standard-of-reference diagnosis for all patients was based on a combination of surgical, intraoperative US, and histologic findings. After inspection of the entire abdomen and complete mobilization of the liver, one of two surgeons (including R.A.E.M.T.) involved in the liver surgery palpated the liver bimanually. One of two available radiologists (M.E.J.P. or M.N.J.M.W., with 3 and 5 years of experience, respectively), with full knowledge of the preoperative data, performed intraoperative US by using an Aloka 2000 system (Aloka, Tokyo, Japan) with a 7.5-MHz transducer tailored for intraoperative US procedures.
We categorized lesions as malignant or benign on the basis of histologic findings, consistency at palpation, characteristic appearance, and compressibility at intraoperative US. Lesions were considered to be malignant if they were firm and uneven at palpation and had one or more of the following features at US (10): irregular shape, irregular margin, bull’s eye appearance, and/or amorphous calcifications. Lesions were considered to be benign if they were soft, compressible, and hyperechoic and had the geographic characteristics of a hemangioma or if they were firm, sharply defined, thin walled, and anechoic and had the postacoustic enhancement characteristics of a cyst. Lesions that did not meet these criteria were considered to be potentially malignant and were either resected or sampled at intraoperative fine-needle aspiration biopsy.
Thus, we identified 123 focal hepatic lesions: 95 were malignant, and 28 were benign (16 hemangiomas, 11 cysts, one granulomatous nodule). Twenty-one (22%) of the 95 metastases were confirmed at histologic analysis, and the remaining 74 (78%) were confirmed by using a combination of inspection, palpation, and intraoperative US findings. The median number of metastases per patient was four (range, 1–12). The location of each lesion was specified by using the Bismuth system (11).
The largest diameter of all metastases was assessed at intraoperative US. Thirty-four metastases were 10 mm in diameter or smaller, and 61 were larger than 10 mm in diameter. Their median diameter was 20 mm (range, 2–105 mm).
Literature Review
All issues of the American Journal of Roentgenology, the Journal of Magnetic Resonance Imaging, Magnetic Resonance Imaging, and Radiology published in 2000 and 2001 were searched article by article for studies in which SDNRs or contrast-to-noise ratios for focal lesions seen at MR imaging were reported.
The formulas used to calculate the SDNR or contrast-to-noise ratio and the size and location of ROIs in the lesions and background organs reported in each article were documented. In addition to the size and location of the noise ROIs, the direction of the noise in relation to phase encoding and the inclusion or exclusion of ghost artifacts were recorded. One author (M.E.J.P.) manually searched the journals, collected information on the quantitative analysis parameters, and subsequently calculated SDNRs for the liver metastases detected in the current series by using the formulas reported in these articles.
Quantitative Analysis
One observer (M.E.J.P., with 3 years of experience) measured all SIs at a workstation (Easyvision; Philips Medical Systems) by using standard software (Philips Medical Systems) and operator-defined ROIs. All ROI measurements were performed on magnitude MR images. For each ROI, the mean SI and the SD were determined.
All benign lesions were excluded from this analysis. To minimize the effects of partial volume averaging, only the 61 metastases with a diameter of greater than 10 mm were eligible for this analysis. Ten metastases that were identified by using the standard of reference could not be detected with MR imaging and therefore were not included in this quantitative analysis. Because of tumor compression of portal venous structures (12), a part of a right liver lobe that contained three metastases in one patient showed extensive edema. We excluded these three metastases from analysis because there was no normal liver parenchyma that could be used for reference. Thus, 48 metastases (median of two metastases per patient; range, one to eight metastases), in 20 patients, larger than 10 mm in diameter were ultimately included in our quantitative analysis.
The SIs of the metastases were measured in two locations. For the first measurement, the ROI was placed in a homogeneous representative area at the margin of the metastasis, and for the second measurement, the ROI encompassed the entire metastasis. The median size of the first metastasis ROI was 21 mm2 (range, 8–93 mm2), and that of the second ROI was 560 mm2 (range, 96–7,224 mm2). An oval ROI was used to measure the SI at the margin of the metastasis, whereas the ROI covering the entire metastasis was identical to the shape of that particular metastasis.
The ROI in the liver parenchyma was located in the right lobe in an optically homogeneous area. Ghost artifacts and vessels were avoided as much as possible. In addition, the ROI in the liver parenchyma was oval and drawn as large as possible, but at least 1 cm2.
The SI of the background noise was measured at four locations in the air outside the patient, as illustrated in Figure 1: (a) in an oval area with the long axis along the phase-encoding direction, anterolateral to the right side of the abdomen; (b) in an oval area with the long axis perpendicular to the phase-encoding direction, in front of the right liver lobe; (c) in a rectangular area with the long axis perpendicular to the phase-encoding direction, in front of the entire liver; and (d) in a rectangular area with the long axis perpendicular to the phase-encoding direction, in front of the entire abdomen.
|
|
|
|
Since malignant tumors are often inhomogeneous, we applied a subjective optical ranking system. One observer (M.N.J.M.W.) who was neither involved in performing the ROI measurements nor aware of the measurement results ranked the extent of areas with low SI (fibrosis and/or calcification) and high SI (necrosis) in each metastasis. Areas of low and high SI were separately categorized by using a three-point scale: 0 meant no areas with high or low SI apparent; 1, small or punctate areas; and 2, large areas. For each metastasis, we combined the scores for these two categories (ie, low- and high-SI areas) into a total optical score by subtracting the low-SI area score from the high-SI area score to classify metastasis inhomogeneity on a five-point scale: –2 indicated a metastasis with large areas of low SI; –1, a metastasis with some areas of low SI or with large areas of low SI and some areas of high SI; 0, a homogeneous metastasis or a metastasis with similar amounts of high- and low-SI areas; 1, a metastasis with some areas of high SI or with large areas of high SI and some areas of low SI; and 2, a metastasis with large areas of high SI.
To compare the formulas used to calculate the SDNR, we used fixed locations of metastasis and background noise ROIs: a homogeneous area at the margin of the metastasis and an oval noise ROI in front of the right liver lobe, respectively. We chose these ROI locations because we believed that they were the most appropriate.
Statistical Analyses
The Wilcoxon signed rank test was used to compare the various SDNR calculations. Since the formulas used to calculate the lesion–background organ SDNR that we found in the literature were quite different—both in scale and in unit—we believed that it was inappropriate to perform a statistical analysis of their ranking.
To quantify the relationships between the SDNRs calculated with the various formulas, the metastasis ROI locations, and the background noise ROI locations, Spearman correlation coefficients were used. All statistical comparisons were performed by using computer software (SPSS, version 10.0.5; SPSS, Chicago, Ill). P < .05 was considered to indicate significance.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Four different formulas were used to calculate lesion–background organ SDNRs or contrast-to-noise ratios in these 34 articles:
1. (SIL – SIBO)/SDN was used 26 times.
2. (SIL – SIBO)/SIN was used five times.
3. (SIL – SIBO)/(SIBO – SDN) was used once.
4. (SIL – SIBO)/(SIBM · SDN) was used once.
5. No formula was described in one article.
In these formulas, SIL is the mean SI of the lesion; SIBO, the mean SI of the background organ; SDN, the SD of the background noise; SIN, the mean SI of the background noise; and SIBM, the mean SI of back muscles. The formula used to calculate the lesion–background organ SDNR was not described in one article (41).
The information about the ROIs in the focal lesions, background organs, and noise that is reported in the 34 articles is summarized in Table 1. Ghost artifacts were excluded from the noise ROI in six studies and were included in four. Only one article (1), similar to our report, included a description of all ROI data. In four (12%) of the 34 articles (6,23,32,38), detailed information about the ROIs was not provided.
|
|
|
= 0.97), as demonstrated in Figure 2. On average, the SDNR calculated with formula 1 was 1.78 times the SDNR calculated with formula 2. The correlation coefficients for the relationship between SDNRs calculated with formulas 1 and 2 and SDNRs calculated with formulas 3 and 4 ranged from 0.60 to 0.75. The correlation coefficient for the relationship between SDNRs calculated with formulas 3 and 4 was 0.48.
|
was 0.89 for the correlation between margin ROI and entire-lesion ROI SDNR values.
The total optical scores calculated for the presence of high- and low-SI areas were –2 for six, –1 for 19, 0 for 14, 1 for nine, and 2 for zero metastases. The mean SDNRs and total optical scores for the ROIs at the metastasis margins and for those encompassing the entire metastases are reported in Table 4. The presented rank data indicate significant differences in SDNRs between the "margin" and "entire" metastasis ROIs with calculated optical scores of –2, –1, and 1, with P values of .028, less than .001, and .008, respectively. We observed no significant difference (P = .92) in SDNRs between the two ROI groups among those with a calculated optical score of 0. Spearman values were 0.43, 0.93, 0.93, and 0.97 for the metastasis ROIs with calculated optical scores of –2, –1, 0, and 1, respectively.
|
|
values ranged from the largest value of 0.98 for the correlation between the SDNR for the oval noise ROI in front of the right liver lobe and the SDNR for the rectangular noise ROI in front of the entire liver (Fig 3) to the smallest value of 0.92 for the correlation between the SDNR for the oval noise ROI anterolateral to the abdomen and the SDNR for the rectangular noise ROI in front of the entire abdomen.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
In the literature search that we performed, the information concerning the quantitative analysis parameters, the locations of the ROIs, and the formulas used to calculate the lesion–background organ SDNR was often incomplete. The various formulas used to calculate lesion–background organ SDNRs that we found in the literature were not equal in unit and had numeric differences over a factor of 200 when we used them for our study material. The metastasis–liver parenchyma SDNRs for the various locations of metastasis (P = .034) and background noise (maximal P = .022) ROIs were significantly different.
Not surprisingly, the metastasis–liver parenchyma SDNR was larger when the formula in which the background noise SD was the denominator (formula 1) was used, as compared with the SDNR calculated with the background noise SI as the denominator (formula 2), because the background noise SI is always the larger number. The SDNRs calculated with formulas 3 and 4 differed substantially from those calculated with formulas 1 and 2—not only in numeric value but also in unit—and thus in our opinion should not be compared with data calculated with formulas 1 and 2. Furthermore, no theoretic background information can be obtained for the rarely used formulas 3 and 4 (8,13,14).
The choice of whether to use formula 1 or formula 2 is more difficult, although formula 2, as explained by Kaufman et al (13), seems the preferable one when SDNRs are calculated from magnitude MR images. However, this choice might be less important because a ratio of the average noise level to the SD of the average noise level exists on magnitude MR images (42). Henkelman (42) reported a theoretic average noise level–to–SD of average noise level ratio of 1.91 and a tested ratio of 1.92 in areas without motion artifacts. In our study, with the presence of motion artifacts, this ratio was 1.78. The strong mathematic relationship between SDNRs calculated with formulas 1 and 2 is reflected by the high correlation coefficient (Spearman = 0.97) obtained when we compared the results obtained with these two formulas. If real rather than magnitude MR images are used, with a resulting background noise SI of around 0, then the noise data must be manipulated, and, consequently, the SD of the manipulated background noise SI should be used, as explained by Kaufman et al (13).
As far as we know, no theoretic background information on the placement of lesion ROIs has been reported. It seems common sense to place the ROI in viable tumor tissue and to avoid areas of the lesion with extreme SI, both high and low. When these areas with high or low SI are large enough relative to the lesion size, they might make the lesion more conspicuous, but they artificially increase or decrease the calculated SDNR. Thus, we think that the margin of a neoplasm, which is virtually always the viable growing part of the lesion, is the region that truly generates contrast with the surrounding tissue. This seems especially important when homogeneous or small lesions, which typically lack areas with extreme SI values, are considered because the viable peripheral component of neoplasms is likely to exhibit the same SI characteristics as that of homogeneous or small lesions consisting entirely of viable tumor.
Our study results demonstrate that significant (P = .034) differences in SDNRs were seen when the ROI at the margin of the metastasis was compared with the ROI encompassing the entire metastasis. When the SDNRs were stratified according to lesion inhomogeneity by using a qualitative factor—specifically, the total optical score—these differences increased for inhomogeneous lesions (P < .028) but almost disappeared for both lesions containing equal amounts of high- and low-SI areas and homogeneous lesions (P = .92).
In general, noise in an image can obliterate diagnostic information and consists of statistical and systematic components (14). Statistical noise spreads almost evenly over an image, whereas systematic noise, depending on its source, can be structured or have some sort of distribution pattern. In general, as stated by Kaufman et al (13), noise should be measured in the air outside of the object being imaged, along the frequency-encoding axis, and away from the filter cutoff region. In the Kaufman et al article, the authors report that for comparison purposes, if one is interested in overall image quality, noise can also be measured along the phase-encoding axis. However, the trade-off for measuring noise in this direction is that the existence of systematic noise, such as that generated by patient motion—for example, breathing—may lead to an overestimation of noise and subsequently a decrease in the SDNR.
Because good overall image quality is crucial for focal lesion detection, we believe that in studies of focal lesion detection, noise should be measured along the phase-encoding axis. Second, since ghost artifacts clearly influence overall image quality, they should not be avoided when placing noise ROIs. Therefore, the noise ROI should not be placed anterolateral to the right side of the abdomen because in this location, systematic noise is almost completely avoided, and, thus, the SDNR will be artificially elevated. Also, the noise ROI should not be placed in front of the entire abdomen because noise generated to the left of the liver will be included in the ROI, and this will cause the SDNR to be artificially decreased.
The choice of whether to place an oval noise ROI in front of the right liver lobe or a rectangular noise ROI in front of the entire liver is less obvious and, in view of the high correlation (Spearman = 0.98) of SDNRs measured in these two areas, less important. In the end, we preferred the oval ROI in front of the right liver lobe rather than the rectangular ROI for measuring noise because the oval region has a larger diameter in the phase-encoding direction, and, thus, selecting it will ensure the inclusion of ghost artifacts generated from breathing, if they are present. Second, the geometry of an oval ROI allows a larger surface area than does that of a rectangular ROI because of the geometry of air within the field of view. In general, if one wants to perform a quantitative analysis, both the field of view and the rectangular field of view should be chosen in a manner such that enough space (ie, air) is available for the correct placement of the background noise ROI.
Although our goal was to demonstrate the effects of changing variables in the analytic method itself, and we believe that this goal is minimally influenced by the MR imaging technique used, a few limitations apply to the MR imaging technique that we used. First, a single pulse sequence was performed by using a machine from only one vendor. Second, the MR images were acquired by using a body coil instead of a phased-array coil, and this resulted in smaller SDNRs. The use of a phased-array coil, though, requires paying extra attention to the placement of the background noise ROI owing to the less uniform SI profile of this coil. To minimize these effects, when a phased-array coil is used, one should describe its exact position and measure the background noise at a fixed distance from the coil. Third, the use of a 10-mm section thickness and the exclusion of metastases 10 mm or smaller left the possibility of partial volume effects. When drawing the ROIs, we paid special attention not to include any normal tissue by keeping a safe distance from the lesion border; however, complete avoidance of tissue averaging can never be guaranteed. Fourth, since for some patients more than one metastasis was included in the analysis, SI interdependence could not be excluded and might have influenced outcomes. Finally, since the anteroposterior diameter of the rectangular noise ROI in front of the entire liver is relatively small compared with the diameter of the oval noise ROI in front of the right liver lobe, there is the chance that ghost artifacts will be missed and consequently that the SDNR will be overestimated.
Since apparently slight changes in the analytic method can result in important numeric changes in calculated metastasis–liver parenchyma SDNRs, we propose the use of a standardized method for the quantitative analysis of focal masses depicted on MR images that includes an accurate description of the ROIs. In our opinion, the SDNR on magnitude MR images should be calculated by using the formula (SIL – SIBO)/SIN, where SIL is the mean SI of the lesion; SIBO, the mean SI of the background organ; and SIN, the mean SI of the background noise. The lesion ROI should be placed in a homogeneous representative area at the margin of the lesion, and the background noise ROI should be placed along the phase-encoding axis in the air and include systematic noise, such as that generated from motion artifacts, since they influence overall image quality. In addition, the background organ ROI must be in a homogeneous representative part of the organ.
|
FOOTNOTES |
|---|
Author contributions: Guarantors of integrity of entire study, M.E.J.P., J.L.B.; study concepts and design, M.E.J.P.; literature research, M.E.J.P.; clinical studies, M.E.J.P., M.N.J.M.W., R.A.E.M.T.; data acquisition, M.E.J.P., M.N.J.M.W., R.A.E.M.T.; data analysis/interpretation, M.E.J.P., J.D., M.N.J.M.W., H.C.v.H.; statistical analysis, M.E.J.P., H.C.v.H., J.L.B.; manuscript preparation, M.E.J.P.; manuscript definition of intellectual content, M.E.J.P., J.D., M.N.J.M.W.; manuscript editing, M.E.J.P., M.N.J.M.W., J.L.B.; manuscript revision/review and final version approval, all authors
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  A. B. Siegel, E. I. Cohen, A. Ocean, D. Lehrer, A. Goldenberg, J. J. Knox, H. Chen, S. Clark-Garvey, A. Weinberg, J. Mandeli, et al. Phase II Trial Evaluating the Clinical and Biologic Effects of Bevacizumab in Unresectable Hepatocellular Carcinoma J. Clin. Oncol., June 20, 2008; 26(18): 2992 - 2998. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D P O'Regan, J Fitzgerald, J Allsop, D Gibson, D J Larkman, D Cokkinos, J V Hajnal, and S A Schmitz A comparison of MR cholangiopancreatography at 1.5 and 3.0 Tesla Br. J. Radiol., October 1, 2005; 78(934): 894 - 898. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. M. Merkle, S. G. Nour, and J. S. Lewin MR Imaging Follow-up after Percutaneous Radiofrequency Ablation of Renal Cell Carcinoma: Findings in 18 Patients during First 6 Months Radiology, June 1, 2005; 235(3): 1065 - 1071. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
