| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Section of Abdominal Imaging, Imaging Institute, Desk Hb6, Cleveland Clinic,
9500 Euclid Ave., Cleveland, OH 44195.
2 Present address: Radiologia y Imagen, Hospital Central Militar, Mexico City,
Mexico.
3 Section of Anatomic Pathology, Cleveland Clinic, Cleveland, OH.
4 Section of Quantitative Health Sciences, Cleveland Clinic, Cleveland,
OH.
Received May 7, 2008;
accepted after revision August 19, 2008.
Address correspondence to B. R. Herts
(hertsb{at}ccf.org).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Of 33 angiomyolipomas identified among 719 renal masses resected from 702 patients over 4 years, 15 masses in 15 patients were prospectively diagnosed on the basis of the presence of fat at MDCT. The 18 patients with minimal-fat angiomyolipoma and a matched (age, sex, tumor size) cohort of patients with renal cell carcinoma were included in this study. Three radiologists independently counted the number of pixels with attenuation less than –10, –20, and –30 HU. Receiver operating characteristic analysis of the number of pixels at each cutoff was used to calculate sensitivity, specificity, and positive predictive value with the following criteria: 1, more than 10 pixels less than –20 HU; 2, more than 20 pixels less than –20 HU; 3, more than 5 pixels less than –30 HU.
RESULTS. Using criterion 1, reader A identified six angiomyolipomas; reader B, five; and reader C, two. The combined sensitivity was 24%; specificity, 98%; and positive predictive value, 69%. Using criterion 2, reader A identified three angiomyolipomas; reader B, four; and reader C, two. The combined sensitivity was 17%; specificity, 100%; and positive predictive value, 100%. Using criterion 3, reader A identified four angiomyolipomas; reader B, four; and reader C, two. The combined sensitivity was 18%; specificity, 100%; and positive predictive value, 100%.
CONCLUSION. CT findings of more than 20 pixels with attenuation less than –20 HU and more than 5 pixels with attenuation less than –30 HU have a positive predictive value of 100% in detection of angiomyolipoma, but most angiomyolipomas with minimal fat cannot be reliably identified on the basis of an absolute pixel count.
Keywords: angiomyolipoma • CT • renal neoplasms
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The ability to reliably and noninvasively differentiate AML from RCC and other malignant tumors is vital to determining the appropriate management of renal masses. Small AML, when asymptomatic, is typically managed conservatively with observation [10]. RCC, depending on the size and location of the tumor, is typically managed with nephron-sparing surgery or nephrectomy. The purposes of this study were to determine whether counting pixels with subzero attenuation on MDCT scans can aid in the diagnosis of renal AML with minimal fat and to determine whether an association exists between number of pixels with subzero attenuation and pathologic assessment of fat content.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
|
|
|
|
All examinations were performed at 120 kVp and an average of approximately 200 mAs. Diagnostic images consisted of 3-mm slice thickness every 3 mm (Sensation 16) or 5-mm slice thickness every 2.5 mm (Volume Zoom, Somatom Plus 4). A slice collimation of 3.0 mm (Somatom Plus 4), 2.5 mm (Volume Zoom), or 0.75 mm (Sensation 16) was used.
Initial Review
The initial review was designed to mimic standard review of CT scans in the
clinical setting. Three radiologists (readers A, B, and C) blinded to the
diagnosis independently and randomly reviewed the preoperative scans of the 18
patients with AML and their matched cohort of 18 patients with RCC. The
radiologists were not blinded to the study design. First, each radiologist
visually evaluated the lesions on unenhanced scans and measured, using region
of interest (ROI) mean attenuation as has been standard practice
[3], areas of low attenuation
for the presence of fat within the lesion. The lesions were then objectively
evaluated by measurement of the mean attenuation of the tumor in the
unenhanced, corticomedullary, and nephrographic phases.
Assessment of Pixels with Negative Attenuation Values
Using a circular ROI, each radiologist independently counted the number of
pixels on the unenhanced CT scans at predetermined maximum thresholds of less
than –10 HU, less than –20 HU, and less than –30 HU. For
counting of pixels with our workstation (MagicView 1000, Siemens Medical
Solutions), a specific threshold must be set for the ROI. To find a potential
best sensitivity and specificity, pixel counts at three levels were chosen.
Each radiologist recorded the ROI size and total number of pixels. Each
radiologist determined the measurement sites and ROI size individually. Each
radiologist performed measurements at as many sites as desired, but values
from only one site were used. This method was specifically designed to allow
detection of even small amounts of fat anywhere within a lesion. It expands on
the standard practice of visually inspecting a lesion for low-attenuation
areas and then measuring the mean attenuation, as originally described by
Bosniak et al. [3]. The use of
absolute pixel counts instead of percentage of pixels made the analysis
independent of ROI size.
The mean attenuation of AML in the unenhanced, corticomedullary, and nephrographic phases was compared with the mean attenuation of RCC in the unenhanced, corticomedullary, and nephrographic phases. In addition, the enhancement in each of the corticomedullary and nephrographic phases (enhanced minus unenhanced atten uation) and the ratios of enhancement in the corticomedullary phase to enhancement in the nephrographic phase were compared between AML and RCC. These comparisons were performed for each reader (Student's t test) and for all of the readers combined (Fisher's exact test).
Fourteen of the 18 AML specimens were retrospectively reviewed by two pathologists who in consensus estimated the fat content as greater or less than 10%. The four specimens not evaluated were unavailable for review. The RCC specimens were not reviewed for fat content.
Statistical Methods
With the constraint that the numbers should be simple criteria, receiver
operating characteristic analysis was used to identify the pixel counts with
the best sensitivity and positive predictive value (PPV). On the basis of the
results of the receiver operating characteristic analysis, the following three
criteria were established: 1, more than 10 pixels less than –20 HU; 2,
more than 20 pixels less than –20 HU; 3, more than 5 pixels less than
–30 HU. Sensitivity, specificity, and PPV were calculated for each
reader for both the initial qualitative diagnosis and using the three
negative-attenuation pixel count criteria. According to Bayes' theorem, the
PPVs were necessarily adjusted for a 4.6% prevalence of AML in the study
population. Results for each reader also were averaged (not pooled) and
reported. The sensitivity, specificity, and PPV of number of pixels with
attenuation less than –10 HU were too low to be considered useful for
the purposes of this study, and this value was not studied further.
Chi-square tests were used to assess the statistical significance of the associations between pathologically estimated fat content less than or greater than 10% (14 AMLs with available specimens) and the number of pixels with negative attenuation on unenhanced CT scans. For each reader, at each phase, we used two-sample Student's t tests to compare the mean attenuation of AML with that of RCC.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Assessment of Negative-Attenuation Pixels
Using criterion 1 (more than 10 pixels less than –20 HU), reader A
identified six AMLs (sensitivity, 33%; specificity, 100%; PPV, 100%); reader B
identified five AMLs (sensitivity, 28%; specificity, 100%; PPV, 100%); and
reader C identified two AMLs (sensitivity, 11%; specificity, 94%; adjusted
PPV, 8%). The combined sensitivity, specificity and PPV for the three readers
were 24%, 98%, and 69% (Table
2).
|
Using criterion 2 (more than 20 pixels less than –20 HU), reader A identified three AMLs (sensitivity, 17%; specificity, 100%; PPV, 100%), reader B identified four AMLs (sensitivity, 22%; specificity, 100%; PPV, 100%); and reader C identified two AMLs (sensitivity, 11%; specificity, 100%; PPV, 100%). The combined sensitivity, specificity, and PPV for the three readers were 17%, 100%, and 100%.
Using criterion 3 (more than 5 pixels less than –30 HU), reader A identified four AMLs (sensitivity, 22%; specificity, 100%; PPV, 100%); reader B identified four AMLs (sensitivity, 22%; specificity, 100%; PPV, 100%); and reader C identified two AMLs (sensitivity, 11%; specificity, 100%; PPV, 100%). The combined sensitivity, specificity, and PPV for the three readers were 18%, 100%, and 100%.
Mean Attenuation and Enhancement of AML Versus RCC
Two of the three readers found a statistically significant difference
between the mean attenuation of AML and that of RCC on unenhanced scans. The
attenuation of AML was approximately 6 HU higher (37 HU) than that of RCC (31
HU). There was no significant difference between the attenuation of AML and
that of RCC in either the corticomedullary or the nephrographic phase for any
reader or the readers combined. There was no significant difference between
the enhancement of AML and that of RCC in the corticomedullary phase for any
of the readers or the readers combined (p > 0.14). Enhancement in
the nephrographic phase was significantly higher for RCC (76.9 HU) than for
AML (65.0 HU) for all readers combined (p = 0.0002). The ratio of
corticomedullary to nephrographic enhancement was significantly lower for RCC
(1.19) than for AML (1.35) for all readers combined (p = 0.04).
Correlation with Pathologic Assessment
Retrospective review of the pathologic specimens showed that six of 14 AMLs
had more than 10% fat. The other eight AMLs with available specimens had less
than 10% fat. Among the six AMLs with more than 10% fat, reader A identified
more than 10 pixels less than –20 HU in four and more than 5 pixels less
than –30 HU in three AMLs. Reader B identified more than 10 pixels less
than –20 HU in three of six and more than 5 pixels less than –30
HU in three of six AMLs. Reader C identified more than 10 pixels less than
–20 HU in one of six and more than 5 pixels less than –30 HU in
one of six AMLs. In the eight AMLs with less than 10% fat, none of the three
readers identified any pixels less than –20 HU and therefore no pixels
less than –30 HU.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
In previous studies of AML with minimal fat, distinguishing features of AML have been found to be homogeneous high attenuation on unenhanced CT scans [6, 8, 19], homogeneous enhancement [6, 7], and prolonged enhancement [7]. In our study, both AML and RCC had greater enhancement in the corticomedullary phase than in the nephrographic phase (ratios > 1), limiting the utility of early and late enhancement for differentiating the two types of tumor. Our results support those of previous studies in that AML had higher attenuation on unenhanced CT. Two of the three readers found a statistically significant difference in mean attenuation values, but the difference was not statistically different for the readers combined. It is unlikely, however, that the higher attenuation on unenhanced CT scans would be sufficient for confident diagnosis of AML with minimal fat. More likely, the finding may lead to percutaneous biopsy in selected patients.
As was ours, a few studies have been conducted to examine measurement of pixel attenuation values. In two studies with only six patients each, investigators attempted to identify fat within small AMLs using single voxel measurement less than –30 HU [20] and pixel mapping to identify three to six contiguous pixels with mean attenuation less than –10 HU [21]. Both sets of investigators used multiple slice thicknesses and found that thinner collimation increased sensitivity. However, neither of these studies had pathologic correlation or a matched control group of patients with RCC. Simpson and Patel [22] studied mean attenuation within a small ROI and pixel mapping. Unlike the other two sets of investigators, Simpson and Patel had a group of patients with histologically proven AML as well as a control group. They found that an ROI –10 HU or less has a specificity of 100% and that use of four adjacent pixels of –10 HU or less would increase the sensitivity of identification of AML and keep a high specificity of 97%. The aim of their study was to find the optimal threshold ROI value; they were not specifically focused on identifying AML with minimal fat. Only six of the 22 AMLs (27%) in that study were minimal-fat AMLs, a marked difference from the masses in our study.
Kim et al. [23] investigated the use of CT histogram analysis to differentiate AML with minimal fat from RCC. As did we, Kim et al. counted pixels with attenuation at less than an arbitrarily chosen threshold within a given ROI. Also as we did, Kim et al. found a statistically significant larger number of pixels with negative attenuation in AMLs with minimal fat than they did in RCCs. That study differed from ours in that 13 of the 34 AMLs evaluated were not histologically proven. The lesions were presumed to be AML on the basis of 24 months of stability and a signal intensity index of 40% or greater on doubleecho gradient-echo chemical shift MR images. In addition, the ROI in the study by Kim et al. was chosen by a single radiologist, who attempted to include the entire tumor and used proprietary software for analysis. In our study, a routine imaging workstation was used, and three radiologists independently chose the ROI site and size to visually search for areas of low attenuation. Despite differences in method, both Kim et al. and we found a similarly low sensitivity of pixel counts for AML with minimal fat.
Catalano et al. [24] measured the percentage of pixels at different thresholds to determine whether 22 AMLs could be differentiated from 28 size-matched RCCs. In that study, the two tumors were not reliably differentiated. The authors used a single ROI oval on the slice at the level of the maximum diameter of the lesion. Although it is reproducible, the technique is not effective in identifying eccentric fat content within the lesion; it is based on the assumption that in minimal-fat AML the fat is evenly distributed throughout the lesion. With our technique, the reader can objectively assess any potential region for pixels with low attenuation values. Using three readers compensates for the potential limitation that our measurement technique may not be reproducible. In addition, as described in Materials and Methods, we chose an absolute pixel count because percentages are highly dependent on ROI and lesion size.
The strength of our study compared with previous studies was the use of an age-, sex-, and size-matched cohort of patients with RCC. This aspect of the study allowed us to evaluate whether AML can be reliably differentiated from RCC according to the criteria we established, which would be the situation in clinical practice. Furthermore, because we started with pathologically proven specimens, we were able to calculate the incidence of AML with minimal fat in our population and were able to statistically correct the sensitivity, specificity, and PPV, a step not performed in other studies.
A subset of AMLs with minimal fat in our study met each of the three criteria established: AMLs with minimal fat that had pixels with negative attenuation less than –20 HU and –30 HU. Although the sensitivity of the number of pixels with negative attenuation was poor (17–24%), with criterion 2 (more than 20 pixels less than –20 HU) and criterion 3 (more than 5 pixels less than –30 HU), the specificity and PPV for AML both were 100%. Therefore, the reader can be confident that when these criteria are met, although an infrequent occurrence, the lesion is AML.
In the comparison of the sensitivity, specificity, and PPV of the pixel count criteria with the findings at initial visual inspection by the three readers, the use of pixel counts did not add to the initial interpretation. Only one reader using criterion 1 found more AMLs using negative-attenuation pixel count. However, negative-attenuation pixel count is an objective rather than a subjective criterion. Furthermore, none of these AMLs was truly identified prospectively.
Our results were supported by the pathologic assessment of AMLs with minimal fat. We found that the AMLs with subjectively less than 10% fat at pathologic examination also had no CT criteria that would have suggested the presence of fat: No pixels with attenuation less than –20 HU were identified in these lesions by any of the three readers. All AMLs in the subset that met criteria 2 and 3 were found at pathologic examination to contain more than 10% fat.
There were limitations to our study. As previous studies have been, our study was limit ed by a relatively small number of lesions, but pathologically proven AML with minimal fat is uncommon and represents approximately 4% of all AMLs [5, 6]. In fact, the number of renal lesions included to identify these patients was quite large, more than 700. In our patient group, AML was found in only 33 of 719 renal mass resections (4.6%), only 18 of 719 lesions (2.5%) being AML with minimal fat. Another limitation was that our retrospective study included three CT scanners and variation in scanning techniques because of the 4-year time period. Although use of several scanners reflects most clinical practices, use of consistent scanning parameters and techniques would have yielded more reliable and reproducible data.
We conclude that a small subset of AMLs with minimal fat can be reliably identified on the basis of the presence of more than 20 pixels with less than –20 HU attenuation or more than 5 pixels with less than –30 HU attenuation. Although they afforded no advantage over subjective evaluation in the study setting, pixel counts are objective criteria for the positive diagnosis of AML with minimal fat. Otherwise, most AMLs with minimal fat, specifically those containing less than 10% fat at pathologic examination, cannot be reliably identified on the basis of pixel attenuation values.
Acknowledgments
The authors thank Cristina Magi-Galluci for assistance with reviewing the
pathologic findings and with manuscript editing.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
–10 HU or 4 adjacent pixels of
–10 HU are recommended as the diagnostic thresholds. Clin
Radiol 2006; 61:410
–416[CrossRef][Medline]
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
