| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, University of Wisconsin Medical School, 600 Highland
Ave., Madison, WI 53792.
2 Department of Radiology, Uniformed Services University of the Health Sciences,
Bethesda, MD.
Received May 3, 2005;
accepted after revision July 10, 2005.
The opinions and assertions contained herein are those of the authors and
should not be construed as official or as reflecting the opinions of the
Department of the Navy, the Department of the Army, or the Department of
Defense.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1. Methods with...References |
|---|
MATERIALS AND METHODS. The study group consisted of 30 polyps on supine and prone CTC data sets. Measurements were performed separately by two experienced radiologists using the same CTC system (Vitrea 2), resulting in 120 linear and volume measurements. Linear size was defined as the longest dimension among the 2D multiplanar reconstruction views. Semiautomated volume determination required manual tracing of polyp boundaries on 2D multiplanar reconstruction views. The relative error between reviewers 1 and 2 was defined as 100 x (|D1 - D2|)/Dave for linear size (D) and as 100 x (|V1 - V2)/Vave for volume (V), where ave is average.
RESULTS. The mean relative error for linear size and volume
measurement was 10.4% ± 10.7% and 16.9% ± 13.2%, respectively.
Median linear size and volume of polyps were 9.4 mm and 270 mm3,
respectively. CTC-derived volumes for medium-sized polyps closely approximated
hemispheric volume (V= (
/12) x Db,
where b =3.13, r2 = 0.90). Small incremental
changes in hemispheric size result in a 3:1 relative change in volume versus
diameter, such that a 1-mm diameter change in a medium-sized hemispheric polyp
results in a relative change in linear size and volume ranging from
approximately 11-18% and 31-53%, respectively.
CONCLUSION. Because changes in polyp volume are amplified compared with linear dimension, volume measurement rather than diameter measurement will better allow detection of small incremental changes in polyp size using CTC.
Keywords: colon • colonography • CT technique
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1. Methods with...References |
|---|
If noninvasive CTC surveillance for unresected medium-sized polyps becomes a standard clinical practice, reliable determination of interval change in polyp size will be critical. The current standard for reporting polyp size at CTC is to use the longest linear dimension from the 2D multiplanar reconstruction views [4]. Although widely accepted and analogous to endoscopic measurement, the reproducibility of linear polyp measurement is not well established. Ideally, reliable discernment of interval polyp growth or regression of approximately 1 mm might be desirable to allow appropriate clinical decision making. Because a relatively small change in polyp diameter corresponds to a much larger proportional change in polyp volume, the margin of error for volume measurement is somewhat relaxed compared to linear measurement. The purpose of this study was to compare the interobserver error associated with linear polyp measurement with that of polyp volume measurement at CTC. Correlation with a hemispheric model was used for medium-sized lesions. Implications with regard to CTC surveillance of unresected polyps were emphasized.
|
|
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1. Methods with...References |
|---|
The study group was drawn from CTC cases performed at our institution where the presence of a polyp measuring 6 mm or greater was identified on both the supine and prone data sets. Segmental location of the 30 lesions included the rectum (n = 6), sigmoid colon (n = 3), descending colon (n = 2), splenic flexure (n = 1), transverse colon (n = 3), hepatic flexure (n = 2), ascending colon (n = 8), and cecum (n = 5). Subjective assessment of polyp morphology at CTC yielded 19 sessile lesions, six pedunculated lesions, four flat lesions, and one annular mass. Histologic diagnoses included 17 tubular adenomas, four tubulovillous adenomas, one villous adenoma, one adenocarcinoma, two hyperplastic polyps, and one hamartoma. Four of the polyps have not been removed and are either awaiting optical colonoscopy or are undergoing CTC surveillance.
Polyp Measurement Techniques
All linear size and volume measurements were performed separately by two
experienced radiologists using the same CTC system (Vitrea 2, Vital Images).
Supine and prone measurements were performed in separate sessions spaced at
least 1 week apart. The window settings were held constant for all
measurements (width, 2,000 H; level, 0 H), and appropriate zoom magnification
was used in all cases. The stalks of pedunculated polyps were excluded from
both linear and volume measurements.
Linear polyp size was defined by the longest dimension among the three orthogonal 2D multiplanar reconstruction views (axial, coronal, and sagittal). Linear measurements were obtained with electronic calipers. The orthogonal plane from which the longest diameter was found was recorded. For the purposes of this analysis, "medium-sized" polyps are defined as lesions with a linear size measuring 6.0 mm or more and less than 10.0 mm.
Polyp volume was obtained by these steps: The polyp boundaries were manually traced using a mouse-driven stylus on all 2D axial images in which the polyp was visualized, and then the outlined regions of interest were combined by automated interpolation performed by the CTC system, which then derived a volume (Fig. 1A, 1B, 1C and 1D). In some cases, the polyp boundaries were traced on either the 2D coronal or sagittal images because of improved conspicuity compared with the axial plane. In our experience, this semiautomated approach for volume measurement is currently more reliable than the existing methods for automated polyp segmentation, which often result in inaccurate border determination.
Statistical Analysis
For this study, the "relative error" between reviewers 1 and 2
for linear size measurement (D) represents the relative percentage
difference, which was defined as 100 x (|D1 -
D2|)/Dave, where
ave refers to average. Similarly, the relative error for polyp volume
measurement (V) was defined as 100 x
(|V1 -
V2|)/Vave. The
absolute values of the measurement differences were used to avoid cancellation
of error values that would overestimate precision. The denominator represents
the average of the two measurements because neither one represents the initial
value. Relative error was used as an indicator of reproducibility in lieu of
other interobserver measures (such as the kappa statistic) because the results
allow more direct assessment of clinical impact. The relative errors were
reported as a percentage; mean values with SD (± SD) were calculated
for supine, prone, and combined assessment. To assess whether relative error
(either diameter or volume) depended on diameter, we plotted these two
parameters against diameter and then evaluated two measures of correlation,
the Pearson's (gaussian) r2 and the Spearman's
(nonparametric) r.
For practical comparison, the results from polyp measurement at CTC were related to the changes associated with varying the size of a hemisphere, particularly within the range of a medium-sized polyp. For uniformity, the relative percentage change for both hemispheric diameter and volume were derived using the same equations for relative error measurements just defined. This comparison allows findings of measurement precision to be placed in a more understandable context because the relationship between polyp diameter and volume is not linear. The CTC-derived volume and diameter were plotted for different size ranges; an empirical power curve was compared with the volume versus diameter curve for an idealized hemisphere for two size ranges (diameter, < 10 mm and < 35 mm).
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1. Methods with...References |
|---|
No significant correlation was found between polyp size and the relative
error for diameter and volume for polyps less than 35 mm. For diameter versus
relative error of the diameter, the Pearson's (gaussian)
r2 =0.022 (p = 0.45) and the Spearman's
(nonparametric) r = -0.05888 (p = 0.76). For diameter versus
relative error of the volume, the Pearson's (gaussian) r2
<0.001 (p > 0.99), and the Spearman's (nonparametric)
r = 0.046 (p = 0.81). A good empirical model of the averaged
polyp volume data for polyps less than 35 mm was obtained by modeling
V to D with a power curve, yielding V(D) =
(/12) x Db, where b = 2.83
(r2 = 0.86); a similar curve was generated for the
individual (nonaveraged) CTC-derived volumes: V(D) = (/12)
x Db, where b = 2.81
(r2 = 0.80) (Fig.
2A and
2B). For larger polyps
(3 10 mm), CTC-derived volume was usually less than hemispheric
volume.
|
|
/12) x Db, where
b =3.13 (r2 = 0.90). A similar fitted curve was
derived using the individual (non-averaged) CTC-derived volumes: V(D)
= (/12) x Db, where b =3.13
(r2 = 0.69) (Figs.
2A and
2B). This closely correlates
with the volume-to-diameter relationship of a hemisphere (where b =3)
and supports the use of a hemispheric model for assessing small incremental
changes in polyp size. Unlike larger polyps, note that the CTC-derived volumes
for medium-sized polyps were generally greater than hemispheric volume for the
same diameter. Subjectively, many medium-sized sessile polyps encountered at
CTC have a roughly hemispheric morphology
(Fig. 3).
|
Therefore, to detect a 1-mm change in polyp diameter reliably (for a medium-sized lesion), the CTC measurement error (e.g., mean error + SD) for linear size should be less than 11-18%, and the measurement error for volume should be less than 31-53%. According to our results for all polyps, the mean error + SD for linear measurement was 21.1% (10.4 + 10.7) and the value for volume measurement was 30.1% (16.9 + 13.2). If we restrict comparison with results for medium-sized lesions, the mean error + SD for linear measurement was 14.4% (8.6 + 5.8) and for volume measurement was 30.5% (16.9 + 13.6). At both ends of the medium-sized polyp diameter spectrum, only the volume measurement error would be less than the actual volume change, whereas the diameter measurement error would almost always be greater than the actual diameter change.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1. Methods with...References |
|---|
CTC offers the potential for noninvasive surveillance of subcentimeter lesions [5]. Because volume is a much more sensitive indicator of change compared with linear dimension, it would seem well suited for this task. Before being applied to either initial polyp evaluation or surveillance, however, polyp volume assessment at CTC must first be validated as a reasonably reproducible measure. Although our findings indicate that the mean relative error associated with polyp volume measurement is slightly greater than that for linear size measurement (16.9% vs 10.4%), this actually translates into a more reliable indicator of change because the resultant volume greatly amplifies any given change in linear size.
The incremental change in the volume of a hemisphere is essentially three times greater than that of its diameter (Appendix 1). As shown in the results, the use of a hemispheric model is supported by our findings, particularly for medium-sized polyps. This is further supported by the subjective observation that most medium-sized polyps are sessile and show an approximately hemispheric appearance at CTC. Because larger polyps often preferentially elongate in one dimension, it is not surprising that their estimated volumes were typically less than that of a hemisphere.
The relative change in volume and linear size of a medium-sized hemispheric polyp that increases or decreases by 1 mm is at least 31% and 11%, respectively. To detect such an interval change at CTC reliably, the overall measurement error (e.g., mean error + SD) should be less than these values. Our results indicate that only the volume measurement at CTC shows the requisite precision to meet these demands. Considerable variation in linear 2D measurement, which only affords a narrow margin of error, has also been shown with irregular extracolonic lesions at standard CT [12].
Prior studies evaluating volume measurement at CT have generally used a method described in 1982 [13] that involved multiplying the sum of individual axial cross-sectional areas by the reconstruction interval, rather than by automated interpolation and volume determination [14-19]. However, accuracy of CT volume measurement has been shown in vitro with constructed phantoms [15, 16, 20]. As with our study, previous in vivo investigations of CT volume measurement for tumors in humans have dealt more with reproducibility or interval change because an absolute reference standard was not available [14, 17, 18, 21]. In general, all of these prior studies evaluated much larger lesions than the polyps in our study.
Compared with existing guidelines for assessing tumor response to neoadjuvant treatments that use 2D CT measurements [22, 23], volume assessment offers the potential for improved evaluation of tumor bulk. A study by Luccichenti et al. [21] used a volumetric CT approach very similar to our own to assess for change in rectal cancers after radiation therapy. However, this group did not assess the error associated with volume measurement and reported reduction in tumor volume after treatment that was less than 10%. According to our findings, this result may lie within the "noise" of the measurement, which underscores the importance of validation before implementation. Volumetric analysis also offers the potential for evaluating a wide range of chemoprevention strategies.
Parallel and somewhat analogous research for CT surveillance of small pulmonary nodules has included investigation of volume assessment [24, 25]. Studies have shown that linear 2D measurement was not reliable for small indeterminate pulmonary nodules [26], whereas volumetric analysis was more reproducible [24]. Reproducibility of polyp volume measurement in our study compared favorably with a study by Kostis et al. [25] that evaluated stable pulmonary nodules. The SD for relative error in that study ranged from 17.1% to 19.3% for pulmonary nodules measuring 5-10 mm in diameter, which is greater than that seen with similar-sized colorectal polyps in our study. Unfortunately, the mean relative errors between our studies cannot be compared because Kostis et al. did not use absolute values, resulting in significant underestimation of this value. In general, colorectal polyp measurement at CT is more challenging than pulmonary nodule measurement because the latter is typically surrounded by air without the confounding factors of adjacent soft tissue and fluid associated with the former.
Our study has several limitations. It is important to reiterate that no reliable reference exists by which to compare in vivo CTC volume measurement in absolute terms to a true gold standard. However, this study is more concerned with precision and reproducibility because detection of incremental change is of more immediate clinical concern than ground-truth accuracy. Linear 2D polyp size will clearly remain the primary clinical touchstone until volume criteria are established. Note that the window setting used for both linear and volume measurements (assuming border determination is manual) will greatly affect accuracy but not precision, if held constant [19]. Furthermore, what actually constitutes a "significant" change in size at CTC surveillance (whether linear or volume) needs to be established. Paradoxically, it appears that the necessary information on the natural history of medium-sized polyps will ultimately come from data that will be derived from subsequent CTC surveillance of such polyps.
Another limitation related to CTC volume measurement is polyp boundary determination, which likely represents the single largest source of error. Precise polyp segmentation, whether manual or automated, is complicated by several factors. For one, distinguishing the polyp border from the colonic wall, which is of soft-tissue attenuation similar to polyps, can be difficult. The composition of the luminal contrast material surrounding the polyp, either gas or opacified fluid, will also affect delineation of the border. Finally, the degree of luminal distention may affect assessment.
A final limitation involves interpretation time. Although the extra time required for volume measurement at CTC is relatively minor, it is clearly greater than that needed for simple linear measurement. This added time will likely not be prohibitive if restricted to follow-up surveillance studies and not used for initial routine evaluation. Once fully automated techniques for volume measurement are capable of reproducible results, it may be feasible to incorporate it into routine evaluation.
In conclusion, compared with linear measurement, the relative error associated with polyp volume measurement at CTC will better allow detection of small incremental changes in polyp size because changes in volume are amplified compared with changes in linear dimension. Polyp volume measurement at CTC could prove useful if noninvasive surveillance of subcentimeter polyps becomes an accepted practice, if volume thresholds for diagnostic algorithms are eventually established, or if specific chemoprevention strategies are found to be effective.
APPENDIX 1. Methods with Which to Measure Incremental Change in the Size of a Hemisphere
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1. Methods with...References |
|---|
Infinitesimal Method
Assuming a Euclidian geometry, the volume of a hemisphere of diameter
D is ( / 12) x D3. The infinitesimal
change in volume (dV) is thus ( / 4) x
D3 x dD. Following the convention in this
article where the "relative change" for diameter is dD /
D and the "relative change" for volume is dV /
V, then
 |
In other words, for very small changes in diameter, the resulting relative change in volume is 3.
Macroscopic Method
Another way to approach this model is to calculate relative errors for
macroscopic, rather than infinitesimal, changes in diameter. Toward this end,
what is the relative change in volume when the diameter increases by 1 mm?
Assume as above an initial hemisphere of diameter D. The volume of
the new hemisphere (VN) with diameter one unit larger is
VN =( / 12) x (D + 1)3. Then
 |
Solving, the relative change (error) for an increment of one unit in
diameter is thus
 |
In other words, for "large" D and an increment of 1
unit in diameter, the ratio approaches 3, the same result we obtained using
the infinitesimal method for any D and "tiny"
D.
As shown in Figures 2A and
2B, polyp volume is not exactly
modeled by the geometry of a hemisphere. However, one can obtain a good
empiric model of the observed polyp volume data by fitting V to
D with a power curve, yielding V= ( / 12) x
Db. One then obtains
 |
In the case of polyps smaller than 10 mm, b = 3.13 (see Results),
a result very similar to that obtained using an exact Euclidian hemisphere
(b = 3). If one prefers the macroscopic method just detailed,
assuming D =1, then
 |
Using the first term of the Taylor expansion of the exponential expression
and assuming "large" D, one obtains
 |
a result identical to the differential model used above.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1. Methods with...References |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  P. J. Pickhardt Screening CT Colonography: How I Do It Am. J. Roentgenol., August 1, 2007; 189(2): 290 - 298. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. C. Yeshwant, R. M. Summers, J. Yao, D. S. Brickman, J. R. Choi, and P. J. Pickhardt Polyps: Linear and Volumetric Measurement at CT Colonography Radiology, December 1, 2006; 241(3): 802 - 811. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
