Published online before print April 15, 2008, 10.1148/radiol.2473070785
(Radiology 2008;247:738-746.)
© RSNA, 2008
Pixel Distribution Analysis: Can It be Used to Distinguish Clear Cell Carcinomas from Angiomyolipomas with Minimal Fat?1
Onofrio A. Catalano, MD 2,
Anthony E. Samir, MD,
Dushyant V. Sahani, MD, and
Peter F. Hahn, MD, PhD
1 From the Division of Abdominal Imaging and Intervention, Department of Radiology, Massachusetts General Hospital, 55 Fruit St, WHT 270, Boston, MA 02114. Received May 4, 2007; revision requested July 2; revision received September 7; accepted September 28; final version accepted November 13.
Address correspondence to P.F.H. (e-mail: phahn{at}partners.org).
 |
ABSTRACT
|
|---|
Purpose: To retrospectively determine if pixel histogram analysis of unenhanced computed tomographic (CT) images can be used to distinguish angiomyolipomas (AMLs) with minimal fat from clear cell renal cell carcinomas (CCRCCs).
Materials and Methods: The human studies committee approved this HIPAA-complaint study, with waiver of informed consent. Patients with pathologically proved AMLs lacking visible macroscopic fat at CT and patients with pathologically proved CCRCCs were included. Lesions were measured, and a histogram (number of pixels with each attenuation) was calculated electronically within a central region of interest. The percentage of pixels below the attenuation thresholds –20 HU and 10 HU was calculated in both cohorts. The unpaired Student t test was used to compare the average percentage of subthreshold pixels at each threshold. P < .05 indicated a significant difference. The number of lesions with more than the selected percentage of subthreshold pixels was calculated in both groups, and the 
2 test was used to test the significance of differences between cohorts. The area under the receiver operating characteristic (ROC) curve was used to determine if any percentage of subthreshold pixels could be used to differentiate between the two cohorts.
Results: There were 22 patients with pathologically proved AMLs lacking visible macroscopic fat on CT images. Tuberous sclerosis affected three of these patients. Mean maximal transverse lesion diameter was 20 mm (range, 11–38 mm). There were 28 patients in the CCRCC comparison group. Mean maximal transverse lesion diameter was 26 mm (range, 15–36 mm). Neither the Student t test (P > .2 for all thresholds <0 HU) nor the 2 test (P > .15 for all thresholds <0 HU) revealed a significant difference between cohorts. A lesion with more low-attenuation pixels was significantly more likely to be characterized as CCRCC than as AML with ROC curve analysis.
Conclusion: Once AMLs with visible fat on CT images are excluded, pixel histogram analysis cannot be used to distinguish between AMLs and CCRCCs.
© RSNA, 2008
|
INTRODUCTION
|
|---|
Angiomyolipoma (AML) is the most common benign solid renal tumor, with a prevalence of 0.3%–3.0% (1). Histopathologic analysis reveals that AMLs are composed of thick-walled blood vessels, adipose tissue, and smooth muscle in proportions that vary greatly among individual tumors (1). The presence of macroscopic fat permits one to confidently diagnose the majority of AMLs on computed tomographic (CT) and magnetic resonance (MR) images (2–5). Although the presence of fat is characteristic of AML, the amount of fat is variable, and, on occasion, AMLs have minimal fat that is not discernible on unenhanced CT images (3). However, with use of a monoclonal antibody, fat-containing AMLs and AMLs with minimal fat stain positively against the melanoma-associated antigen HMB-45. This feature helps to characterize AMLs, including AMLs with minimal fat, at histologic analysis (3).
On unenhanced CT images, the appearance of AMLs with minimal fat has been described as homogeneous and hyperattenuating compared with the surrounding renal parenchyma (6). However, the hyperattenuation is nonspecific, as other renal masses—including metastases, venous infarctions, leiomyomas, and 22% of clear cell renal cell carcinomas (CCRCCs)—share this feature on CT images (3,7). Histologic analysis shows that CCRCCs contain intracellular lipid and glycogen deposits but that they usually lack focal deposits of adipose tissue (8). On the other hand, it has been reported that AMLs with minimal fat possess at least 3% adipose tissue at histologic analysis (mean, 4%; range, 3%–10%), even when AMLs lack a homogeneous focus of adipose tissue larger than 5 mm at gross examination (3).
Recent studies have shown that the hidden fat content of solid masses can be identified with histogram analysis of individual pixel attenuation on unenhanced CT images. This strategy has been used to detect fat in adrenal lesions. Indeed, some benign adrenal adenomas with an average attenuation higher than 10 HU on unenhanced CT images may be distinguished from metastases to the adrenal glands at histogram analysis if more than 10% of the pixels have an attenuation lower than 0 HU (9). Thus, the purpose of our study was to retrospectively determine if pixel histogram analysis of unenhanced CT images can be used to distinguish AMLs with minimal fat from CCRCCs.
|
MATERIALS AND METHODS
|
|---|
This single-institution study was conducted in accordance with the Health Insurance Portability and Accountability Act and was approved by the human studies committee of our hospital, with waiver of informed consent.
Patients
An institutional electronic database of histopathologically characterized renal masses was searched for patients with renal lesions who met the following inclusion criteria: First, patients had to have histopathologically proved renal AML. Second, they had to have undergone unenhanced CT between September 1997 and February 2005, and images had to be available in standard Digital Imaging and Communications in Medicine format. Third, there had to be an absence of visually discernible fat in the corresponding renal lesion. Although AMLs with visible fat are rarely examined with biopsy or resection, one investigator (O.A.C., 10 years experience interpreting abdominal CT images) reviewed CT images to exclude any AMLs with visible fat on CT sections that covered the renal mass. This investigator also examined patients' medical records to determine whether they had a history of tuberous sclerosis.
Once the cohort of patients with fat-poor AMLs had been established, we sought a size-matched cohort of patients with CCRCC who presented to our institution during the same period. We restricted our study to compare only CCRCCs with AMLs with minimal fat because, on the basis of histologic findings, CCRCCs would be expected to provide the greatest potential for overlap with AMLs with minimal fat.
We subsequently searched our database for instances of CCRCC pathologically proved at biopsy, surgery, or both. We included all patients who presented between September 1997 and February 2005 and had CCRCCs that were similar in size to AMLs, provided the CCRCC was visible on an available unenhanced CT image. When the same lesion was seen on more than one unenhanced CT image, we examined the most recent CT image obtained prior to resection or biopsy. When this image was unavailable, we used the same CT image used to guide the biopsy.
CT Scanning
Different CT scanners and scanning protocols were used over the 8-year period. We used a single–detector row helical CT unit (Highlight Advantage; GE Medical Systems, Milwaukee, Wis), a four–detector row CT unit (Lightspeed QX/I; GE Medical Systems), an eight–detector row CT unit (Lightspeed Ultra; GE Medical Systems), a 16–detector row CT unit (Lightspeed 16, GE Medical Systems; Sensation 16, Siemens Medical Solutions, Malvern, Pa), and a 64–detector row CT unit (Sensation 64; Siemens Medical Solutions). We determined the number of patients examined with each scanner and technique.
Pixel Histogram Analysis
Unenhanced CT studies were analyzed with commercially available image analysis software (Amira 4.1; Mercury Computer Systems, Chelmsford, Mass). One investigator (O.A.C.) assessed and measured each lesion. The index lesions were measured on each axial CT image. The maximal lesion diameter was recorded. An oval region of interest (ROI) was placed on the CT image that showed the maximal lesion diameter. The ROI was placed to encompass at least three-fourths of the lesion. Care was taken to avoid the external margins of the lesion to minimize volume averaging with surrounding fat. Histograms (number of pixels with each attenuation [measured in Hounsfield units]) were automatically calculated within the ROI throughout the attenuation range from –400 HU to 350 HU. The histograms were saved as an Excel spreadsheet (Microsoft Office 2000; Microsoft, Bethel, Wash). To define a lesion as containing fat, we chose the following attenuation thresholds: –20 HU, –15 HU, –10 HU, –5 HU, 0 HU, 5 HU, and 10 HU. For each lesion, AML, or CCRCC, the percentage of pixels below that threshold was calculated. We referred to this percentage as the percentage of subthreshold pixels.
Statistical Analysis
Mean lesion size (± standard deviation) was calculated, and the range of lesion sizes was recorded. The mean value and standard deviation of the percentage of subthreshold pixels was computed at each attenuation threshold for AMLs and CCRCCs. Then, the unpaired Student t test was used to test the significance of any difference between the average percentage of subthreshold pixels in the AML cohort and that in the CCRCC cohort. We calculated the number of lesions with more than 4%, 5%, 6%, 7%, 8%, 9%, and 10% of pixels below each attenuation threshold in both groups. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated at each of the selected attenuation thresholds for each percentage of pixels, first assuming that AMLs contain more fat than do CCRCCs and then assuming that CCRCCs contain more fat than do AMLs. The fact that each pair consisted of an attenuation threshold and a corresponding percentage of subthreshold pixels gave rise to a 2 x 2 contingency table that consisted of the number of lesions (AMLs or CCRCCs) with and without sufficient pixels below the threshold. We performed 2 analysis with the Fisher exact test to determine if differences between the groups might have arisen by chance, and two-tailed P values were calculated. To determine if different CT techniques used with the different imagers could have affected the results, we performed a 2 test to compare the distribution of patients with AMLs and those with CCRCCs between scanners with a single detector row or four detector rows and those with more than four detector rows. For the same reason, we recalculated the Fisher exact test for each subgroup of patients who underwent CT with the same technique. The t and 2 tests were performed with Stata software (version 10; Stata, College Station, Tex).
A priori, a larger or smaller percentage of subthreshold pixels might be used to distinguish AMLs with minimal fat. We used receiver operating characteristic (ROC) curve analysis to determine whether the percentage of subthreshold pixels could be used to distinguish AMLs with minimal fat from CCRCCs at any threshold, comparing the notion that large percentage of subthreshold pixels imply AML with the notion that small percentage of subthreshold pixels imply AML at each threshold. Continuous variable ROC analysis was performed with the CLABROC program (Charles E. Metz, University of Chicago, Chicago, Ill), which was running on a Macintosh computer (system 9.3; Apple Computer, Cupertino, Calif), to compare the area under estimated binormal ROC curves. In all statistical comparisons, P < .05 was considered to indicate a significant difference that was unlikely to have arisen by chance.
|
RESULTS
|
|---|
Between September 1997 and February 2005, 43 consecutive patients had pathologically proved renal AMLs after they underwent CT that did not lead to a diagnosis of AML as the cause of the renal lesion. Seven patients were excluded because visible fat in the lesion was seen at CT during study enrollment. The remaining 36 patients were considered to have AMLs with minimal fat. Eleven patients were excluded because no unenhanced CT image was available. Three patients were excluded because they had undergone unenhanced CT elsewhere and CT images were available only as film hard copies. A retrievable Digital Imaging and Communications in Medicine–format unenhanced CT study that did not show visible fat in the lesion was available for 22 patients. These 22 patients comprised our AML study group (13 women, nine men; mean age, 55 years; age range, 21–85 years). Tuberous sclerosis affected three patients (two men, one woman). Eleven AMLs were pathologically proved at biopsy, 10 were proved at surgery, and one was proved at biopsy and surgery. Mean maximal transverse lesion diameter was 20 mm ± 7 (range, 11–38 mm) (Fig 1).
View larger version (151K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1a: AML with minimal fat in the left kidney. (a) Axial unenhanced CT image shows a homogeneous oval cortical mass (arrows) without visible fat. (b) Corresponding pixel histogram shows that 40 (7%) of 564 pixels have an attenuation lower than 0 HU.
|
|
View larger version (12K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1b: AML with minimal fat in the left kidney. (a) Axial unenhanced CT image shows a homogeneous oval cortical mass (arrows) without visible fat. (b) Corresponding pixel histogram shows that 40 (7%) of 564 pixels have an attenuation lower than 0 HU.
|
|
Between September 1997 and February 2005, 48 consecutive patients had pathologically proved CCRCC after they underwent CT and AML was ruled out as the cause of the renal lesion. We excluded 19 patients with CCRCC because their maximal transverse lesion diameter was larger than 38 mm, which was the diameter of the largest lesion in the AML group. Thus, the study cohort included 29 consecutive patients with a biopsy or surgical diagnosis of CCRCC who had undergone unenhanced CT and whose renal lesions were the same size as those in the AML group. One patient was subsequently excluded because unenhanced CT had been performed elsewhere and only film hard copies were available. The remaining 28 patients comprised our CCRCC comparison group (eight women, 20 men; mean age, 58 years; age range, 37–88 years). Mean maximal transverse lesion diameter was 26 mm ± 8 (range, 15–36 mm). Twenty-two CCRCCs were pathologically proved only at biopsy; three, only at surgery; and three, at biopsy and surgery (Fig 2). Of these 28 patients with CCRCCs, 20 underwent radiofrequency ablation, six underwent surgery, and two underwent treatment elsewhere.
View larger version (120K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2a: CCRCC of the right kidney. (a) Axial unenhanced CT image shows a homogeneous oval cortical mass (arrows). (b) Corresponding pixel histogram shows that 261 (15%) of 1700 pixels have an attenuation lower than 0 HU.
|
|
View larger version (14K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2b: CCRCC of the right kidney. (a) Axial unenhanced CT image shows a homogeneous oval cortical mass (arrows). (b) Corresponding pixel histogram shows that 261 (15%) of 1700 pixels have an attenuation lower than 0 HU.
|
|
CT Scanners
We performed 12 CT examinations (AML, n = 8; CCRCC, n = 4) on a single–detector row CT scanner (120–140 kVp, 170–230 mA, 5-mm section thickness, 5.0–7.5 mm/sec table speed, 1.0-second rotation speed). We performed 26 CT examinations (AML, n = 10; CCRCC, n = 16) with a four–detector row CT scanner (120–140 kVp, 170–280 mA, 2.5–5.0-mm section thickness, 11.25–18.75 mm/sec table speed, 0.8–1.0-second rotation speed). We performed two CT examinations (AML, n = 2) with an eight–detector row CT scanner (140 kVp, 230 mA, 2.5-mm section thickness, 13.5 mm/sec table speed, 0.5-second rotation speed). We performed nine CT examinations (AML, n = 2; CCRCC, n = 7) with a 16–detector row CT scanner (120–140 kVp, 130–240 mA, 2.5–5.0-mm section thickness, 5.0–18.75 mm/sec table speed, 0.8–1.0-second rotation speed). We performed one CT examination (CCRCC, n = 1) with a 64–detector row CT scanner (120 kVp, 200 mA, 5.0-mm section thickness, 5.0 mm/sec table speed, 1.0-second rotation speed).
Pixel Analysis
The percentage of pixels below the selected attenuation threshold revealed there was a higher percentage of subthreshold pixel content in the CCRCC cohort than in the AML cohort (Table 1). Sensitivity, specificity, PPV, and NPV were calculated at each of the selected attenuation thresholds for each percentage of pixels, first assuming that AMLs contain more fat than do CCRCCs (Table 2) and then assuming that CCRCCs contain more fat than do AMLs (Table 3).
The unpaired Student t test (Table 4) did not reveal any significant difference in the mean percentage of pixels below the selected threshold between the AML and CCRCC cohorts at any of the negative attenuation thresholds (<–20 HU to <0 HU). The P values of the mean percentage of pixels below each threshold were as follows: <–20 HU, .84; <–15 HU, .68; <–10 HU, .49; <–5 HU, .31; <0 HU, .21; <5 HU, .07; and <10 HU, .03. There was a significant difference at the 10-HU subthreshold, with the mean percentage of subthreshold pixels being slightly higher for CCRCCs than for AMLs.
For attenuation values ranging from –20 HU to 10 HU and for pixel thresholds ranging from more than 4% of pixels to more than 10% of pixels, two-tailed significance was calculated with the Fisher exact test. For an attenuation of less than –20 HU, P values varied from .682 at the more than 4% pixel threshold to .439 at the more than 10% pixel threshold. For an attenuation of less than –15 HU, P values varied from .439 to more than .999 for the same range of pixel thresholds (more than 4% to more than 10%). For an attenuation of less than –10 HU, P values ranged from .153 to more than .999. For an attenuation of less than –5 HU, P values ranged from .061 to .444. For an attenuation of less than 0 HU, P values ranged from .159 to .154. For an attenuation of less than 5 HU, P values ranged from .045 to .011. For an attenuation of less than 10 HU, P values ranged from .021 to .024. Thus, significant differences between the cohorts appeared for only positive thresholds, and CCRCCs tended to have more subthreshold pixels than did AMLs at these high thresholds.
The 2 test did not reveal any significant difference regarding the distribution of AMLs and CCRCCs between CT scanners with a single detector row or four detector rows and those with more than four detector rows. Moreover, the Fisher exact test, which was repeated for each subgroup of patients who underwent CT with the same technique, did not reveal a difference from the same test run considering all the patients together, independent of the CT technique used.
ROC Analysis
For every threshold from –20 to 10 HU, the ROC curve for the assumption that a small percentage of subthreshold pixels implied AML dominated the ROC curve for the assumption that a large percentage of subthreshold pixels implied AML; areas under the ROC curves ranged from 0.67 to 0.71 for the former and from 0.29 to 0.32 for the latter. P values for comparisons at the seven thresholds were as follows: .026 for less than –20 HU, .014 for less than –15 HU, .018 for less than –10 HU, .008 for less than –5 HU, .019 for less than 0 HU, .004 for less than 5 HU, and .005 for less than 10 HU. The more low-attenuation pixels a lesion had, the more likely it was to be a CCRCC than an AML with minimal fat. This characteristic was independent of attenuation threshold within the range of –20 HU to 10 HU (Fig 3).
|
DISCUSSION
|
|---|
About a third of AMLs associated with tuberous sclerosis are of the minimal fat variety, although some controversy exists regarding the AML subtypes to which this term should be applied (6,10). Some authors believe AML with minimal fat should be reserved to describe renal masses devoid of macroscopic fat, which appears homogeneously hyperdense on unenhanced CT images, enhancing on contrast material–enhanced CT images, hypointense on T2-weighted MR images, and isoechoic on ultrasonographic images (10). Others use the term more liberally to indicate AMLs that do not contain macroscopically visible fat, independent of their signal intensity, attenuation, and echotexture (10). In our study, we did not evaluate the attenuation, contrast enhancement, or signal intensity of AMLs; instead, we relied solely on the absence of macroscopic fat on unenhanced CT images to classify pathologically proved AMLs as AMLs with minimal fat.
Macroscopic fat-containing renal cancers are extremely rare, and—to our knowledge—they have been discussed only in case reports (11–13). Macroscopic fat content in patients with renal cell carcinoma has been attributed to bone metaplasia, tumor lipid necrosis, and engulfment of perirenal or sinus fat (12). Both AMLs and renal cancer lesions enhance after contrast medium administration (6). AMLs tend to enhance in a homogeneous fashion, but this characteristic may not provide the reassurance necessary to avoid surgery (3). Follow-up imaging may be used in patients suspected of having AMLs, although it was proved that 53% of AMLs smaller than 4 cm grow during 4-year follow-up (14). The growth rate per year varies among different types of AMLs, with solitary AMLs exhibiting a 5% growth rate per year; multiple AMLs not associated with tuberous sclerosis, a 22% growth rate per year; and multiple AMLs associated with tuberous sclerosis, an 18% growth rate per year (15).
Different strategies have been explored to discern AMLs with minimal fat from renal cancers. These have included pixel mapping, analysis of contrast enhancement pattern at contrast-enhanced CT, and chemical shift MR imaging (1,7,16).
In a study by Kim et al (7), homogeneous enhancement and a prolonged enhancement pattern were valuable CT findings that enabled the authors to differentiate AML with minimal fat from CCRCC, with a PPV of 91% and an NPV of 87%. Unfortunately, 47% of AMLs with minimal fat did not exhibit these enhancement characteristics (7). Although some authors have reported that MR imaging is unable to depict adipose tissue in AMLs with minimal fat (3), others have found different results. Kim et al (16) used double-echo gradient-recalled-echo chemical shift MR imaging to calculate the signal intensity index and the tumor-to-spleen ratio to differentiate AMLs from other renal neoplasms. They found that the quantification of intratumoral fat was useful for differentiating AMLs from non-AML renal neoplasms. Specifically, a signal intensity index of 25% yielded a sensitivity of 96% and a specificity of 93%, and a tumor-to-spleen ratio of –32% yielded a sensitivity of 88% and a specificity of 97%. Kim et al (16) found the intratumoral fat content was significantly higher in the AML cohort than in the non-AML cohort.
Mean CT attenuation measurements rely on averaging different tissue densities within the selected ROI. Thus, CT attenuation measurement may result in inadequate information when small areas of tissue heterogeneity have been averaged. CT histogram analysis accurately displays the various CT attenuations and the corresponding number of pixels within the selected ROI. This technique has been shown to be more sensitive for use in lipid detection, represented as pixels with negative attenuation values in lipid-poor adrenal adenomas, particularly in those with a mean CT attenuation of more than 10 HU. Indeed, some adrenal cortical adenomas with high average attenuation may nevertheless be distinguished from metastases to the adrenal glands if more than 10% of pixels have an attenuation less than 0 HU (9,17).
Prompted by the results of pixel histogram analysis studies on adrenal adenomas, we investigated whether a similar approach—but not necessarily a threshold based at 0 HU—could be used to distinguish AMLs with minimal fat from CCRCCs. We restricted our study to compare minimal-fat AMLs with CCRCCs because, on the basis of histologic findings, CCRCCs would be expected to provide the greatest potential for overlap.
In our series, CCRCCs had a higher fat content than did size-matched AMLs. Thus, our results differ from the results of others who have focused attention on AMLs with minimal fat (1,16). In a study by Simpson and Patel (1), pixel mapping analysis with a line of four consecutive pixels with an attenuation of less than or equal to –10 HU or a square of four contiguous pixels with an attenuation of less than or equal to –10 HU had a sensitivity of 86% and a specificity of 97% for diagnosis of AML. In their study, the control group comprised patients with renal cell cancers, regardless of histologic subtype; those with normal renal parenchyma; and those with simple renal cysts. Moreover, it appears that the AML group did not include AMLs with minimal fat only or AMLs that were devoid of any area of lucency. During ROI and pixel mapping analysis, the most lucent area on inspection was chosen for analysis. In our study, we compared AMLs with minimal fat that were devoid of any lucent area at visual inspection with CCRCCs only. We analyzed the largest axial section of the lesion instead of the most lucent one because patients in our AML cohort did not have an area of visible fat.
Moreover, we did not perform pixel mapping, which may yield information on the spatial relationships between pixels. Instead, we evaluated the pixel histograms of the lesions, which provide information about the percentage attenuation composition of the pixels independent of their location. Kim et al (16) used double-echo gradient-recalled-echo chemical shift MR imaging to differentiate AMLs with minimal fat from other renal neoplasms. The main difference between their study and ours appears in the renal neoplasm cohort, which was restricted to include only CCRCCs in our study. Perhaps this factor contributed to the differing results. In our study, when AMLs with obvious macroscopic fat have been excluded, additional pixelogram analysis has not allowed us to differentiate AMLs from CCRCCs on the basis of their fat composition.
Our observations are not discordant with the findings of a recent study on CT pixel analysis of pathologically evaluated adrenal masses. Remer et al (18) found that metastases, pheochromocytomas, and adrenocortical carcinomas contained negative pixels, with a resultant decrease in the specificity of CT histogram analysis in the diagnosis of adrenal adenomas from 38.2% to 28.9% when one negative pixel was used as a diagnostic threshold.
Moreover, in a recently published article, Silverman et al (19) stated that focal or diffuse signal intensity suppression on opposed-phase gradient-recalled-echo MR images of a renal mass does not enable differentiation of CCRCCs from AMLs. Our finding may have important implications for patient care when a small solid renal lesion containing minimal fat is detected at CT. Our results favor a more aggressive biopsy-based approach, as opposed to one based on calculation of composition characteristics on CT images.
Our study had several limitations. It was a retrospective study; thus, various scanning parameters were used over the study period. The thin-section feature of advanced multi–detector row CT scanners was not used in our study. All of the images had a 2.5–5.0-mm section thickness. Thus, one would not expect the advanced technology of 16– and 64–detector row CT scanners to make a difference in the detection of small foci of fat. However, we did not have a sufficient number of patients to study all of the detector configurations; therefore, this was a limitation. In a future study, researchers might use 16– or 64–detector row scanners to reconstruct pixels from isotropic voxels to determine whether such thin sections might be more sensitive to fat within renal lesions.
Moreover, we did not include all renal masses, or even all AMLs, encountered during the study period. Numerous additional AMLs were encountered, recognized by their fat content, anddismissed by experienced fellowship-trained subspecialty abdominal radiologists. Had CT interpretations been provided by readers with less experience, lesions with more fat might have been resected or sampled for biopsy, and this might have influenced our results. We tried to control for this by re-evaluating all of the CT images that showed resected or biopsied AMLs and excluded those AMLs that we believed had visible macroscopic fat. Our practice also differs from that of others in that we make extensive use of percutaneous biopsy of focal renal masses (20,21). In other practice settings, renal masses suspected to be fat-poor AMLs might not come to histologic sampling as readily as in our study.
Furthermore, although a histologic diagnosis was made in every patient, the percentage of fat content in the resected lesions was not determined routinely. Thus, we relied on historic controls for the asserted differences in macroscopic fat content between AMLs with minimal fat and CCRCCs.
In summary, we performed pixel-by-pixel histogram analysis of unenhanced CT images of histologically proved AMLs with minimal fat and CCRCC. Once lesions with macroscopic fat have been excluded, pixel attenuation histogram analysis cannot be used to distinguish AMLs from CCRCCs; on the contrary, lower CT attenuation is more strongly associated with CCRCCs than with AMLs.
|
ADVANCE IN KNOWLEDGE
|
|---|
- Once lesions with macroscopic fat have been excluded, high fat content seen at histogram analysis cannot be used to distinguish angiomyolipoma from clear cell renal cell carcinoma (CCRCC); on the contrary, high fat con-tent actually increases the chances that a renal mass is a CCRCC.
|
IMPLICATION FOR PATIENT CARE
|
|---|
- When a small noncystic non–fat-containing renal lesion is detected at CT, our results favor a more aggressive biopsy-based approach over one based on calculation of composition characteristics on CT images.
|
ACKNOWLEDGMENTS
|
|---|
We acknowledge statistical advice provided by Elkan F. Halpern, PhD, during the study and the preparation of this manuscript.
|
FOOTNOTES
|
|---|
Abbreviations: AML = angiomyolipoma • CCRCC = clear cell renal cell carcinoma • NPV = negative predictive value • PPV = positive predictive value • ROC = receiver operating characteristic • ROI = region of interest
2 Current address: Department of Radiology, AO G Rummo, Benevento, Italy. 
Author contributions: Guarantors of integrity of entire study, O.A.C., 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; manuscript final version approval, all authors; literature research, O.A.C.; clinical studies, O.A.C.; statistical analysis, O.A.C., A.S., P.F.H.; and manuscript editing, all authors
Authors stated no financial relationship to disclose.
|
References
|
|---|
- Simpson E, Patel U. Diagnosis of angiomyolipoma using computed tomography-region of interest < or =–10 HU or 4 adjacent pixels < or =–10 HU are recommended as the diagnostic thresholds. Clin Radiol 2006;61:410–416.[CrossRef][Medline]
- Kutikov A, Fossett LK, Ramchandani P, et al. Incidence of benign pathologic findings at partial nephrectomy for solitary renal mass presumed to be renal cell carcinoma on preoperative imaging. Urology 2006;68:737–740.[CrossRef][Medline]
- Hafron J, Fogarty JD, Hoenig DM, Li M, Berkenblit R, Ghavamian R. Imaging characteristics of minimal fat renal angiomyolipoma with histologic correlations. Urology 2005;66:1155–1159.[CrossRef][Medline]
- Pereira JM, Sirlin CB, Pinto PS, Casola G. CT and MR imaging of extrahepatic fatty masses of the abdomen and pelvis: techniques, diagnosis, differential diagnosis, and pitfalls. RadioGraphics 2005;25:69–85.[Abstract/Free Full Text]
- Zagoria RJ. Imaging of small renal masses: a medical success story. AJR Am J Roentgenol 2000;175:945–955.[Free Full Text]
- Choyke PL, Glenn GM, Walther MM, Zbar B, Linehan WM. Hereditary renal cancers. Radiology 2003;226:33–46.[Abstract/Free Full Text]
- Kim JK, Park SY, Shon JH, Cho KS. Angiomyolipoma with minimal fat: differentiation from renal cell carcinoma at biphasic helical CT. Radiology 2004;230:677–684.[Abstract/Free Full Text]
- Outwater EK, Bhatia M, Siegelman ES, Burke MA, Mitchell DG. Lipid in renal clear cell carcinoma: detection on opposed-phase gradient-echo MR images. Radiology 1997;205:103–107.[Abstract/Free Full Text]
- Bae KT, Fuangtharnthip P, Prasad SR, Joe BN, Heiken JP. Adrenal masses: CT characterization with histogram analysis method. Radiology 2003;228:735–742.[Abstract/Free Full Text]
- Jinzaki M, Silverman SG, Tanimoto A, Shinmoto H, Kuribayashi S. Angiomyolipomas that do not contain fat attenuation at unenhanced CT. Radiology 2005;234: 311.[Free Full Text]
- Schuster TG, Ferguson MR, Baker DE, Schaldenbrand JD, Solomon MH. Papillary renal cell carcinoma containing fat without calcification mimicking angiomyolipoma on CT. AJR Am J Roentgenol 2004;183:1402–1404.[Free Full Text]
- Prando A, Prando D, Prando P. Renal cell carcinoma: unusual imaging manifestations. RadioGraphics 2006;26:233–244.[Abstract/Free Full Text]
- Helenon O, Chretien Y, Paraf F, Melki P, Denys A, Moreau JF. Renal cell carcinoma containing fat: demonstration with CT. Radiology 1993;188:429–430.[Abstract/Free Full Text]
- Yamakado K, Tanaka N, Nakagawa T, Kobayashi S, Yanagawa M, Takeda K. Renal angiomyolipoma: relationships between tumor size, aneurysm formation, and rupture. Radiology 2002;225:78–82.[Abstract/Free Full Text]
- Lemaitre L, Robert Y, Dubrulle F, et al. Renal angiomyolipoma: growth followed up with CT and/or US. Radiology 1995;197:598–602.[Abstract/Free Full Text]
- Kim JK, Kim HS, Jang YJ, et al. Renal angiomyolipoma with minimal fat: differentiation from other neoplasms at double-echo chemical shift FLASH MR imaging. Radiology 2006;239:174–180.[Abstract/Free Full Text]
- Jhaveri KS, Wong F, Ghai S, Haider MA. Comparison of CT histogram analysis and chemical shift MRI in the characterization of indeterminate adrenal nodules. AJR Am J Roentgenol 2006;187:1303–1308.[Abstract/Free Full Text]
- Remer EM, Motta-Ramirez GA, Shepardson LB, Hamrahian AH, Herts BR. CT histogram analysis in pathologically proven adrenal masses. AJR Am J Roentgenol 2006;187:191–196.[Abstract/Free Full Text]
- Silverman SG, Mortele KJ, Tuncali K, Jinzaki M, Cibas ES. Hyperattenuating renal masses: etiologies, pathogenesis, and imaging evaluation. RadioGraphics 2007;27:1131–1143.[Abstract/Free Full Text]
- Hunter S, Samir A, Eisner B, et al. Diagnosis of renal lymphoma by percutaneous image guided biopsy: experience with 11 cases. J Urol 2006;176:1952–1956.[CrossRef][Medline]
- Shah RB, Bakshi N, Hafez KS, Wood DP Jr, Kunju LP. Image-guided biopsy in the evaluation of renal mass lesions in contemporary urological practice: indications, adequacy, clinical impact, and limitations of the pathological diagnosis. Hum Pathol 2005;36:1309–1315.[Medline]
This article has been cited by other articles:
|
|
|
|
 
C. Simpfendorfer, B. R. Herts, G. A. Motta-Ramirez, D. S. Lockwood, M. Zhou, M. Leiber, and E. M. Remer
Angiomyolipoma with Minimal Fat on MDCT: Can Counts of Negative-Attenuation Pixels Aid Diagnosis?
Am. J. Roentgenol.,
February 1, 2009;
192(2):
438 - 443.
[Abstract]
[Full Text]
[PDF]
|
|
|
TOP
ABSTRACT
INTRODUCTION
) was plotted by using the assumption that a larger percentage of subthreshold pixels favored a diagnosis of CCRCC. This curve dominates the lower curve (
), which was determined by using the opposite assumption, that a larger percentage of subthreshold pixels favored a diagnosis of AML. TPF = tune-positive fraction.