| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Genitourinary Imaging |
1 From the Department of Radiology–White 270, Massachusetts General Hospital and Harvard Medical School, 55 Fruit St, Boston, MA 02114. From the 2004 RSNA Annual Meeting. Received December 14, 2004; revision requested February 7, 2005; revision received July 7; accepted August 2; final version accepted October 25. Address correspondence to P.F.H. (e-mail: phahn{at}partners.org).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Materials and Methods: The study was approved by the local ethical committee, which waived patient consent, and was conducted in compliance with HIPAA. Electronic searching revealed patients with adrenal nodules scanned with both a Siemens 16–detector row scanner and one of eight GE Medical Systems multi–detector row scanners between January 2000 and September 2004 without the use of intravenous contrast material. Lesions were characterized by using histologic findings, fat content, or size stability. Size stability for 6 months was required unless both scans were obtained within 21 days of each other. Two radiologists independently measured lesion attenuation for regression analysis. Lesions considered benign (
10 HU) on one scan and indeterminate (>10 HU) on the other were separately analyzed, and technical parameters of scanning were compared.
Results: There were 47 patients (27 men, 20 women; age range, 40–86 years; mean age, 64 years) with four metastases, 42 adenomas, and one myelolipoma (long-axis length, 10–85 mm; mean, 24 mm). GE scans were obtained with four–detector row scanners (n = 32), an eight–detector row scanner (n = 2), and 16–detector row scanners (n = 13). Correlation between the two readers was 0.99 for both Siemens and GE scan attenuation measurements. The slope of regression lines for both readers plotting GE attenuation (y-axis) against Siemens attenuation (x-axis) was less than 1, which indicated a slight but statistically significant tendency for GE scans to have lower attenuations than do Siemens scans. For both readers, there were more lesions indeterminate (>10 HU) on Siemens scans but benign (10 HU) on GE scans than the reverse (McNemar test, P < .02 for reader 1, not significant for reader 2). Average Siemens and GE scan technical parameters were similar.
Conclusion: There are only slight differences in attenuation of adrenal nodules measured on scans obtained with different scanners.
© RSNA, 2006
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
One of the features of benign adrenal lesions that can distinguish them from malignant lesions is fat content. Many adrenal adenomas contain sufficient intracellular fat to influence their imaging characteristics in a diagnostically detectable way. In unenhanced CT scans, the presence of fat reduces the x-ray attenuation (3), often to a level lower than that encountered in metastases.
For adrenal lesion attenuation to constitute a practical test for differentiation of benign from malignant lesions, a threshold must be established. Although numerous studies have been performed, to our knowledge a threshold that indicates perfect discrimination has not been found. Indeed, metastases have uncommonly been encountered that measure less than 10 HU, which is barely above the attenuation of water (4). Because the risk to the patient from misdiagnosis of a malignant lesion exceeds the risk from misdiagnosis of a benign one, the attenuation threshold should be set so that almost all lesions below it are benign. Lesions above the threshold would be either benign or malignant; that is, they would be indeterminate. Thresholds between 0 and 20 HU have been considered in the literature (5). An analysis published in 1998 of data pooled from 10 prior studies recommended 10 HU as a reasonable choice for many practices (5). Local prevalence of disease, radiologist experience, and clinical history influence how the threshold is applied to individual cases.
Results of a recent ex vivo imaging study (6) indicated that characteristics of individual CT scanners influence adrenal lesion x-ray attenuation. Excised adrenal lesions in an anthropomorphic phantom showed varying results when scanned with three different scanners from two different manufacturers.
At our institution, we have performed on average for the past 5 years about 100 000 CT examinations each year with scanners made by two different manufacturers. Many patients have undergone more than one noncontrast examination that included the adrenal glands. Thus, the purpose of our study was to retrospectively compare noncontrast scans of the same adrenal lesion obtained with two different manufacturers' multidetector scanners to assess whether there are differences in attenuation measurements.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
During the period of the study, our hospital and its coordinated imaging centers used eight multi–detector row CT scanners, most of which were Lightspeed Advantage scanners (GE Medical Systems, Milwaukee, Wis). Some scanners were upgraded from earlier models after January 2000. For most of the study duration, there were four 16–detector row, one eight–detector row, and either two or three four–detector row GE scanners. In December 2002, one of the three four–detector row GE scanners was replaced with a Siemens 16–detector row scanner (Sensation 16; Siemens Medical Solutions, Malvern, Pa). All of the scanners are maintained and calibrated according to the manufacturers' specifications by the manufacturers' service personnel.
The cohort under investigation included adult patients with focal adrenal enlargement who between January 2000 and September 2004 had undergone noncontrast CT that included the enlarged adrenal gland with both a Siemens and a GE scanner. An additional requirement concerned lesion stability. Unless the Siemens and GE scans had been obtained within 3 weeks of each other, the lesion had to exhibit size stability (which implied attenuation stability) for at least 6 months as determined with serial CT.
Patient Identification
We located patients for inclusion in the study with a two-step process. The first step involved preparation of two electronic lists of hospital medical record numbers. A database (Folio Views 4.2; Fien Group, Encino, Calif) of all radiologic reports was searched for any report that mentioned an adrenal nodule, adenoma, mass, tumor, metastasis, lesion, or enlargement. No restriction was placed on the type of study, so the adrenal abnormality might have been noted with some other modality, such as sonography or magnetic resonance imaging, rather than CT. A second list of medical record numbers was prepared directly from the radiology information system (IDXrad; IDX Systems, Burlington, Vt) to include all abdominal, chest, and vascular CT scans obtained with the Siemens scanner. These two lists were processed entirely electronically, which preserved patient anonymity even from the investigators. There were 397 medical record numbers common to both lists.
For the second step, one investigator (P.F.H.) reviewed all 397 cases in the radiology picture archiving and communication system (Impax 4.1; Agfa HealthCare, Ridgefield Park, NJ). The investigator determined by inspection of the CT images whether the patient met the inclusion criteria. For those who did, he recorded demographic data (age, sex, adrenal histologic findings if available, and any clinical or biochemical evidence of medullary hyperfunction) about the patient and technical data (radiation dose [milliampere-seconds], tube current [kilovolt peak], field of view, and section thickness) about the individual noncontrast Siemens and GE scans. When there was more than one study obtained with either Siemens or GE equipment, the most recent Siemens study was selected, and then the temporally closest GE study either before or after the Siemens study was selected. The investigator recorded the image number in which the lesion was best seen. If there were multiple adrenal lesions, the investigator identified and selected the largest, which was used for the analysis.
Lesions were characterized according to histologic findings when available. Water attenuation and soft-tissue–attenuation lesions not characterized by using histologic findings were considered adenomas if stable for more than 6 months and there was no evidence of medullary hyperfunction. Any fat attenuation lesion was classified as a myelolipoma. Any lesion otherwise fulfilling the inclusion criteria but not belonging to one of these categories would have been included in the analysis but would have been considered uncharacterized. However, no lesion remained uncharacterized.
Image Analysis
Image analysis was performed by two radiologists (P.F.H., M.A.B.) with special interest in adrenal imaging and 19 and 7 years of experience, respectively, interpreting abdominal CT scans. They worked independently at picture archiving and communication system workstations to measure the size and x-ray attenuation (mean ± standard deviation) of each lesion; one lesion per patient was measured. Each radiologist chose the size and position of the oval region-of-interest (ROI) cursor, but the image number and site (left or right adrenal gland) to be measured was specified beforehand. The radiologists were instructed to use the largest region that would fit entirely within the lesion being measured. Each lesion was measured once by each radiologist. To ensure that the ROIs were placed consistently, the readers viewed both the Siemens and the GE scans together when placing their ROIs. The attenuation data collected consisted of pairs of numbers, and attenuation of each lesion on Siemens scans and on GE scans were recorded in Hounsfield units. There was one pair of numbers for each patient from each of the two radiologists.
Statistical Analysis
The pairs of attenuation values as measured by each radiologist were subjected to correlation and linear regression analysis. Perfect (identical) data would lie on the diagonal line y = x. Deviation (tipping) of the line below the diagonal (slope less than 1) would indicate a tendency for attenuation on Siemens scans to be higher than on GE scans. We rejected that such a difference would have arisen by chance if P < .05. The same analysis was performed on Siemens data from the two radiologists and on GE data from the two radiologists in order to determine to what extent observer differences might generate systematic discrepancies.
The Siemens-GE regression line for each reader represents an estimate of the GE attenuation measurement based on the Siemens attenuation measurement for a typical lesion at various levels of the attenuation scale. For each reader, we predicted the attenuation on a GE scan for a lesion that measured 10 HU and a lesion that measured 20 HU on a Siemens scan, and we computed the percentage deviation.
Lesions with attenuation of 10 HU or less on one scan but greater than 10 HU on the other manufacturer's scan were termed transitional lesions. Transitional lesions are important when a threshold of 10 HU is used, because such lesions would be considered benign or indeterminate, depending on which scan was measured. These lesions were considered separately by using the McNemar test to determine whether there were more transitional lesions that were benign on Siemens scans and indeterminate on GE scans or that were indeterminate on Siemens scans and benign on GE scans.
Finally, the technical features of the two manufacturers' examinations, including radiation dose (milliampere-seconds), tube current (kilovolt peak), field of view, and section thickness, were compared with a t test to look for systematic differences. Where significant technical differences were found, these differences were correlated with Siemens-GE ROI attenuation differences for each reader. In this way, we sought to estimate the contribution these differences might have made to the differences in measured adrenal lesion attenuation. ROI standard deviation on the paired scans was also compared by using a t test and correlated with section thickness.
Correlation, regression, and t tests were performed with statistical software (Statview II v1.03, 1988; Abacus Concepts, Berkeley, Calif) on a personal computer (Macintosh IIci; Apple, Cupertino, Calif). To test for deviation from the diagonal of the Siemens-GE regression line, the linear regression model was applied to the difference between attenuation on GE scans and attenuation on Siemens scans plotted against the Siemens scan attenuation. Graphs of the Siemens-GE regression lines were prepared (Excel v9.0.5526; Office 2000; Microsoft, Redmond, Wash). The McNemar test was performed with software (MedCalc for Windows v7.4.1.0, 2004; MedCalc Software, Mariakerke, Belgium). In all statistical comparisons, differences were considered significant if P < .05.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Twenty-eight of the lesions were in the left adrenal gland, and 19 were in the right adrenal gland. All of the Siemens examinations were performed with our one 16–detector row Siemens scanner. Among the examinations performed with GE equipment, 32 were performed with four–detector row scanners, two with the eight–detector row scanner, and 13 with the 16–detector row scanners.
Attenuation Measurements
For Siemens scans, correlation between the two readers' measured attenuation of adrenal lesions was r = 0.99 and that for GE scans was also r = 0.99. The slopes of the regression lines were 0.97 for Siemens scans and 0.96 for GE scans. For the reader 1, the correlation of adrenal lesion attenuation measured on Siemens scans and that on GE scans was r = 0.90, and the regression line was y = 0.866x + 0.622 (P < .05) (Fig 1a). For reader 2, correlation was r = 0.90, and the regression line was y = 0.854x + 0.787 (P < .03) (Fig 1b). Slopes of 0.866 and 0.854, which are less than 1, means that for both readers, the regression lines of measurement pairs (x, y) = (Siemens, GE) were distinct from the diagonal line y = x; Siemens scan-based measurements tended to be higher than GE scan-based measurements. These regression lines predict that a lesion measuring 10 HU on the Siemens scan would measure 9.3 HU on the GE scan as measured by either reader, which is a deviation of 0.7/10 (7%). A lesion measuring 20 HU on a Siemens scan would measure 17.9 HU on a GE scan as measured by either reader, which is a deviation of 2.1/20 (10.5%). Reader 1 measured the attenuation of 27 of 47 lesions to be greater on Siemens scans than on GE scans; reader 2 measured the attenuation of 25 lesions to be greater on Siemens scans than on GE scans.
|
|
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
The regression lines relating attenuation on Siemens scans to attenuation on GE scans as measured by the two readers provide an estimate of the deviation of the GE measurements from the Siemens measurements for lesions of various attenuation. For a lesion measuring 10 HU on a Siemens scan, the regression lines predict attenuation of 9.3 HU on a GE scan. If attenuation is 20 HU on a Siemens scan, the lesion attenuation would be 17.9 HU on a GE scan. We noted that the actual variability, both above and below the regression lines, is much greater. Consequently, patient factors and scan-to-scan variability independent of which scanner produced the images are likely to be more important than the manufacturer of the scanner. Such factors might include differences in section location, patient position, and body habitus, as well as scanning technique. In our study, we did not measure scan-to-scan variability—that is, differences in attenuation of the same lesion measured on two different scans obtained with the same CT scanner. The methods in this study could, however, be extended to address this question.
Although we compared scans obtained with a single Siemens scanner with scans obtained with many different GE scanners, similar underlying multidetector technology was employed by all of the scanners. By using similar multidetector technology, as well as having an in vivo retrospective design, our study differs from the previous study (6) of adrenal explants. In that study, investigators imaged explants with three different scanners, all with helical technology but only one with multidetector technology. For explants, the differences between measured attenuation values were on the same order of magnitude as the differences encountered in our study. Neither our study nor the previous ex vivo study compared the scans obtained with helical scanners with CT scans obtained with the older axial scanners from which most of the threshold determinations were originally made (5). More recent reports have used in part (7–9) or in their entirety (10,11), however, helical CT data to support the use of attenuation values for differentiation of benign from malignant adrenal lesions.
Our study has important limitations. It is a single-center study; characteristics of our scanners may differ from the same models used elsewhere. The preponderance of benign adrenal lesions is a characteristic of our cohort imposed because of the study design and may not reflect the ratio of benign to malignant lesions encountered in practice. The noncontrast CT scans that provided the in vivo adrenal measurements were obtained for many different reasons, so there was no uniformity of technique. Average section thickness of the Siemens scans was slightly greater than that of the GE scans. If this difference had any effect on measured attenuation, it would be to lower the Siemens measurements from z-axis partial volume averaging of retroperitoneal fat cephalic or caudal to the adrenal gland. Thus, it is possible that the tendency for Siemens measurements to exceed GE measurements might have been underestimated slightly in our study.
The two readers were not blinded to the manufacturer of each CT scanner. We presented the images in pairs so that the readers could place their ROIs consistently. If the readers harbored a bias as to the performance of either scanner, this bias could have been reflected in their measurements. The largely concordant performance of the two readers suggests that bias was not a factor.
More than a year had elapsed between many of the pairs of examinations. Although we excluded adrenal lesions that changed in size, patient body habitus and perhaps even the attenuation values of some of the adrenal lesions could have changed over time. More than one GE CT scanner was used, possibly obscuring greater differences that might exist between individual scanners.
Our research compared CT scans obtained with equipment provided by only two manufacturers. These two manufacturers' scanners are among the most widely used. Nevertheless, we cannot extrapolate our results to all scanners in use today.
Finally, our study does not address reliability of attenuation measurements made on postcontrast washout CT scans, which have been shown to provide additional information important for adrenal lesion characterization (8,9,12). Use of the postcontrast washout technique can help to characterize adrenal lesions considered to be indeterminate on an unenhanced CT scan.
We consider our results reassuring. Differences in scanners probably do produce small differences in adrenal attenuation measurements but so do other factors, which are at least as important. As any other test result, adrenal attenuation measurements should not be interpreted in a vacuum. Clinical data, together with the relative costs of either a false-negative or a false-positive diagnosis, must be considered in individual cases.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Abbreviations: ROI = region of interest
Authors stated no financial relationship to disclose.
Author contributions: Guarantor of integrity of entire study, P.F.H.; 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, P.F.H., G.W.L.B.; clinical studies, P.F.H., M.A.B.; statistical analysis, P.F.H.; and manuscript editing, all authors
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
