| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, Box 3808, Duke University Medical Center, Erwin Rd.,
Durham, NC 27710.
2 Present address: Institute for Diagnostic, Interventional and Pediatric
Radiology, Inselspital Bern, University of Bern, Bern, Switzerland.
Received September 14, 2007;
accepted after revision January 31, 2008.
Address corrrespondence to L. M. Ho
(lisa.ho{at}duke.edu).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. One hundred thirty-two adrenal nodules were identified in 104 patients with lung cancer who underwent PET/CT. Sixty-five nodules were classified as lipid-rich adenomas if they had an unenhanced CT attenuation of less than or equal to 10 H. Thirty-one masses were classified as lipid-poor adenomas if they had an unenhanced CT attenuation greater than 10 H and stability for more than 1 year. Thirty-six masses were classified as lung cancer metastases if they showed rapid growth in 1 year (n = 27) or were biopsy-proven (n = 9). Histogram analysis was performed for all lesions to provide the mean attenuation value and percentage of negative pixels.
RESULTS. All lipid-rich adenomas had more than 10% negative pixels; 51.6% of lipid-poor adenomas had more than 10% negative pixels and would have been classified as indeterminate nodules on the basis of mean attenuation alone. None of the metastases had more than 10% negative pixels. Using an unenhanced CT mean attenuation threshold of less than 10 H yielded a sensitivity of 68% and specificity of 100% for the diagnosis of an adenoma. Using an unenhanced CT threshold of more than 10% negative pixels yielded a sensitivity of 84% and specificity of 100% for the diagnosis of an adenoma.
CONCLUSION. CT histogram analysis is superior to mean CT attenuation analysis for the evaluation of adrenal nodules and may help decrease referrals for additional imaging or biopsy.
Keywords: adrenal adenoma • CT • histogram analysis • lipid-poor
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Currently, three established methods are used to diagnose a lipid-poor adenoma. Calculation of an absolute percentage of washout between enhanced and delayed CT has been shown to have excellent diagnostic performance for the diagnosis of a lipid-poor adenoma [1, 5]. However, this test may require a return visit to radiology as well as additional radiation and contrast media exposure. Opposed-phase MRI has been shown to diagnose lipid-poor adenomas that measure between 10 and 30 H on unenhanced CT [6, 7]. The third method is percutaneous biopsy. Because of its invasive nature and inherent risk, biopsy is usually considered a last resort when noninvasive testing has been inconclusive.
Histogram analysis is another method for evaluation of adrenal masses that was first reported by Bae et al. [8]. With this technique, a region of interest (ROI) is placed on the unenhanced CT image of an adrenal mass and the number of pixels that measure less than 0 H are counted. Subsequent studies using histogram analysis have shown that the sensitivity and specificity of this technique can be improved if a threshold of more than 5–10% negative pixels is applied for diagnosis of an adenoma [4, 9].
The purpose of our study was to determine whether histogram analysis in conjunction with the application of a threshold of 5–10% negative pixels can be used to increase the sensitivity for diagnosis of adenomas compared with using a standard CT attenuation threshold of less than or equal to 10 H. We hypothesize that some adrenal masses that would have been classified as indeterminate on the basis of unenhanced CT attenuation of more than 10 H, can be characterized as lipid-poor adenomas using histogram analysis, thus avoiding the need for additional imaging or biopsy.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Imaging
Diagnostic-quality CT of the adrenal glands was performed on a 16-MDCT
scanner (Discovery ST, GE Healthcare) as part of a PET/CT exami nation.
Imaging parameters for CT were as follows: 140 kVp, automated tube current
modu lation, noise index of 15–20 (weight-dependent),0.5-second gantry
rotation, 3.75-mm slice thick ness, 1.375:1 pitch, table speed of 27.5 mm per
rotation, and 16 x 1.25 mm detector config uration. No IV contrast
material was administered for these CT examinations. Additional detailed in
formation regarding the PET/CT technique used in this study has been described
previously [10].
CT Analysis
One observer with 4 years of CT experience determined the mean attenuation
of each adrenal nodule. Analysis was performed on a PACS workstation
(Centricity 2.1, GE Healthcare). For every adrenal mass, an oval or round ROI
was placed in the center of the lesion to include as much of the
cross-sectional area as possible but avoiding the outermost edge (to prevent
partial volume artifact), areas of calcification, and areas of necrosis. The
mean attenuation and longest long-axis measurement of each left or right
adrenal mass were recorded.
CT Histogram Analysis
Two observers with 3 and 14 years of CT experience, respectively, performed
the CT histogram analysis. The analysis was performed on a postprocessing
workstation (Advantage Windows 4.2, GE Healthcare) equipped with commercially
available software capable of displaying a pixel map in the form of a
histogram chart (CT Perfusion 3, GE Healthcare). An ROI was placed in the
center of the adrenal mass to include as much of the cross-sectional area as
possible but to avoid the outermost edge, areas of calcification, and areas of
necrosis. Using the histogram option of the CT Perfusion software, a histogram
and the total number of pixels were obtained for the prescribed ROI. The
histogram displays the range of pixel attenuations in a prescribed ROI along
the x-axis and reports the frequency of each pixel attenuation along
the y-axis. The number of negative pixels (i.e., attenuation value
measuring < 0 H) in a given ROI is determined by mathematically summing all
negative pixels reported on the histogram graph. The number of negative pixels
was divided by the total number of pixels to obtain the percentage of negative
pixels for each adrenal nodule (Fig.
1A,
1B). One ROI placement and one
histogram analysis were performed for every eligible adrenal nodule.
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The mean attenuation of the lipid-rich adenomas was significantly less than that of the lipid-poor adenomas and metastases (Table 1). The mean percentage of negative pixels was significantly higher for lipid-rich adenomas than for lipid-poor adenomas and metastases (Table 1). However, the mean percentage of negative pixels was not significantly different between the lipid-poor adenomas and metastases. The size of the metastases was significantly greater than the size of the lipid-rich and lipid-poor adenomas (Table 1). We also noted that the average size of lipid-rich adenomas was greater than lipid-poor adenomas; this difference was statistically significant.
|
We found that 100% of lipid-rich adenomas, approximately 50% of lipid-poor adenomas, and 0% of metastases showed more than 10% negative pixels. By comparison, 100% of lipid-rich adenomas, nearly 70% of lipid-poor adenomas, and 6% of metastases had more than 5% negative pixels. If the detection threshold is set to the lowest value of at least one negative pixel, then 100% of lipid-rich adenomas, 84% of lipid-poor adenomas, and 69% of metastases would meet this condition. Figure 2 illustrates our finding that the mean CT attenuation of an adrenal nodule correlates inversely with the percentage of negative pixels (r = –0.91). Our data indicate that using a threshold of more than 10% negative pixels yields an overall higher sensitivity for the diagnosis of a benign adenoma than using a mean CT attenuation threshold of less than 10 H, while maintaining 100% specificity (Table 2). Using a negative pixel threshold of 5% results in a slightly higher sensitivity but also a modest decrease in specificity compared with using either a 10% negative pixel threshold or a mean threshold CT attenuation of 10 H (Table 2).
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
In another study, Remer et al. [9] showed that application of a threshold of 10% negative pixels on unenhanced CT for the diagnosis of an adenoma resulted in a sensitivity of 70–72% and specificity of 91–97% when the study population was composed of only metastases and adenomas. Comparatively, we showed an overall higher sensitivity of 84% and specificity of 100% when a threshold of 10% negative pixels was applied. The reason for our greater number of true-positives and subsequent higher sensitivity is that 68% of our adenomas were lipid-rich and all of our lipid-rich adenomas had more than 10% negative pixels. By contrast, only 30% of the adenomas studied by Remer et al. were lipid-rich, which would likely decrease their overall percentage of true-positives and contribute to a lowered sensitivity. We believe that our study population composed of nearly 70% lipid-rich adenomas more closely reflects the expected proportions of lipid-rich and lipid-poor adenomas in the general population [1–4]. The reason for the lower specificity in the study by Remer et al. is unknown, but their results fall close to our 95% CI limits of 92–100% for specificity.
Most recently, Jhaveri et al. [4] found that 46% of lipid-poor adenomas with CT attenuation greater than 10 H contained more than 10% negative pixels when histogram analysis was applied. Also in that study, none of the nonadenomas showed more than 10% negative pixels, for a specificity of 100% for the diagnosis of an adenoma. The results of our study support the findings of Jhaveri et al. Like those authors, we found that approximately 50% of lipid-poor adenomas and none of the metastases showed more than 10% negative pixels on histogram analysis.
By applying the threshold of 10% negative pixels, we increased our sensitivity for the diagnosis of adenoma from 68%, when a standard CT mean attenuation threshold of less than or equal to 10 H was used, to 84%. Therefore, we determined that our stated hypothesis is true. That is, some of the adrenal masses that would have been classified as indeterminate on the basis of unenhanced CT attenuation of more than 10 H, can be confidently characterized as lipid-poor adenomas using CT histogram analysis.
The application of a threshold of 5% negative pixels slightly increases sensitivity for the diagnosis of an adenoma compared with using a threshold of 10% negative pixels; however, the advantage of this higher sensitivity is compromised by decreased specificity. In an oncologic population, decreasing specificity means that more metastases may be misclassified as benign adenomas, which could lead to understaging of the primary disease. Thus, we suggest that application of a threshold of 10% negative pixels, with its high specificity and acceptable sensitivity, is preferable to one of 5% negative pixels, particularly in the setting of cancer patients.
The relationship between lipid content of adrenal adenomas and CT attenuation has been well established in the radiology literature. Korobkin et al. [12] compared the quantity of lipid in resected adrenal adenoma specimens with in vivo CT attenuation and found an inverse linear relationship between the percentage of lipid-rich cells and the unenhanced mean CT attenuation. They also reported that although most adenomas were composed of a homogeneous population of lipid-laden cells or lipid-poor cells, some adenomas showed an interwoven admixture of clusters of lipid-rich and lipid-poor cells. More recently, Gabriel et al. [13] reported a subset of adrenal adenomas that showed heterogeneous suppression during out-of-phase gradient-echo MRI. Histologic analysis of these cases revealed an interwoven admixture of clusters of vacuolated (lipid-laden) cells and compact (lipid-poor) cells similar to the heterogeneous adenomas described previously by Korobkin et al. [12]. This subset of adrenal masses with heterogeneous suppression constituted 14% (34/173) of their study population.
Although we have few histologic specimens of our adenomas available for evaluation, we speculate that the subset of adrenal adenomas with heterogeneous composition described Korobkin et al. [12] and Gabriel et al. [13] may correspond to a subgroup of adenomas that have both a mean CT attenuation of more than 10 H and more than 10% negative pixels on CT histogram analysis. In fact, 17% of all our adenomas (lipid-rich and lipid-poor) met both of these criteria, which is similar to the percentage of adenomas (14%) with heterogeneous composition described by Gabriel et al. CT histogram analysis, by way of its pixel-by-pixel analysis, is particularly suited to provide a more sensitive characterization of tissues with varying attenuations in a small cross-sectional area such as an adrenal adenoma with heterogeneous composition.
In keeping with the results of prior studies [2, 14], the malignant adrenal nodules in our study were significantly larger than benign ones. An unexpected finding was that lipid-rich adenomas were larger than lipid-poor adenomas, and this difference was also statistically significant. In a prior study of lipid-rich and lipid-poor adenomas, no significant difference in size was seen [5]. The reason our lipid-poor adenomas were smaller than the lipid-rich adenomas is unknown. Possibly the smaller nodules were more sensitive to partial volume effects, resulting in an overall higher mean CT attenuation, which subsequently led to the more frequent categorization of the smaller adrenal nodules as lipid-poor adenomas.
Although our study shows that CT histogram analysis with a threshold of 10% negative pixels is more sensitive for the diagnosis of an adenoma than using a mean attenuation threshold of 10 H on unenhanced CT, Jhaveri et al. [4] recently reported that a 20% signal drop threshold on opposed-phase MRI is more sensitive than CT histogram analysis. Their data suggest that if a patient is required to return for another imaging test to work up an adrenal mass, it would be better to perform dedicated adrenal MRI with opposed-phase imaging than CT histogram analysis.
Our data also show considerable overlap in the negative percentage of pixels of lipid-poor adenomas and metastases, which had a mean CT attenuation between 20 and 40 H. This finding would indicate that CT histogram analysis would be less useful for evaluating adrenal nodules that measure more than 20 H.
Limitations
One limitation of this study is the potential for sample bias. Because
these patients were retrospectively identified from a pool of patients with a
history of lung cancer, it is possible that the results of this study may not
directly apply to patients with other tumors or to the general population. A
prospective study of incidentally discovered adrenal masses found in a
population of patients scanned for a variety of indications is needed to
validate our findings.
Another potential limitation of this study is the lack of long-term follow-up or pathologic proof for many of the adrenal masses having a mean attenuation of less than 10 H, which we categorized as lipid-rich adenomas. Because this criterion has been validated by several previous investigations [2, 3, 5] and has been used for years in many imaging departments, imaging follow-up or biopsy of lesions measuring less than 10 H on unenhanced CT was not obtained by our clinical colleagues. Further, it is possible that some of the adrenal lesions were not adrenal adenomas but rather adrenal cysts, which are uncommon but also benign. We also cannot entirely exclude the possibility that an adrenal lesion with an attenuation similar to that of water could represent a necrotic adrenal metastasis. To our knowledge, no collision tumors of adrenal adenomas and adrenal metastases were present in our study population.
In addition to study limitations, there is also a limitation of the CT histogram technique. The presence of negative pixels in nonadenomatous tissues, reported by both Remer et al. [9] and Jhaveri et al. [4], was also shown in our study. However, the proportional percentage of negative pixels in our adrenal metastases was small, measuring an average of approximately 2% compared with an average of 43% for lipid-rich adenomas and 12% for lipid-poor adenomas. Furthermore, the application of a threshold of 10% negative pixels reduces the influence of false-positive negative pixels. The explanation for the presence of negative pixels in metastases is unknown. Perhaps the inadvertent inclusion of normal adrenal tissue or a collision tumor in the ROI sample may be a source of negative pixels for some metastases. Another accepted explanation is that CT scanner noise may spuriously produce negative pixels [4, 9]. It has been reported that the percentage of negative pixels correlates highly with an increase in the SD of the mean attenuation values, which is also a measure of CT scanner noise [15].
Thus, measures to optimize image quality and diminish CT scanner noise have been recommended to reduce the likelihood of false-positive CT histogram results [4, 8, 9]. One strategy suggested by Bae et al. [8] is to use images reconstructed with a smooth soft-tissue kernel typically used for abdominal CT rather than a sharp lung or bone algorithm, which is usually noisier. Because image noise is inversely related to the square root of the tube current [16], CT images should be acquired with standard milliampere-second and peak kilovoltage values used for the abdomen rather than with a low-dose technique [8, 16].
Another limitation of this technique is the potential for interobserver
variability when performing CT histogram analysis. In a prior study by Remer
et al. [9], two observers
performing histogram analysis of adrenal nodules had moderate agreement
(
= 0.45; percentage of agreement, 76%) for unenhanced nonadenomas and
substantial agreement ( = 0.63; percentage of agreement, 87%) for
unenhanced adenomas. In our study, two observers performed the histogram
analysis together, yielding one measurement for each adrenal nodule. However,
because of the potential for interobserver variability with CT histogram
analysis, it is possible that two independent observers would find different
or conflicting results using our data. Nevertheless, it is reassuring that the
results of our study are consistent with and supportive of previously
published reports regarding the use of CT histogram analysis for adrenal
nodules [4,
9]. Although an analysis of our
data for interobserver variability is beyond the scope of this study, a
focused evaluation of interobserver variability with histogram analysis is a
subject that may be worthy of study in future research projects.
Conclusion
In summary, our study shows that CT histogram analysis using a threshold of
10% negative pixels can be used to characterize some adenomas that would
otherwise be considered indeterminate by mean CT attenuation alone. As a
result, CT histogram analysis may help decrease referrals for additional
imaging or biopsy of adrenal nodules, particularly in patients with a known
primary malignancy.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
