| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Cardiac Imaging |
1 From the Grace Ballas Research Unit of the Cardiac Rehabilitation Institute (J.S., M.M.) and Department of Diagnostic Imaging (S.A., J.R., Y.I.), Chaim Sheba Medical Center, Sackler School of Medicine, Tel-Aviv University, Tel-Hashomer 52621, Israel; Division of Epidemiology and Preventive Medicine, Tel-Aviv University, Tel-Hashomer, Israel (N.K.M.); and SOCAR Research Societe Anonyme, Nyon, Switzerland (B.A.K.). Received April 30, 2003; revision requested July 10; final revision received December 29; accepted January 30, 2004. Supported by a grant from Bayer, Leverkusen, Germany. Address correspondence to J.S. (e-mail: dshemesh@netvision.net.il).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSTATISTICAL CONSULTANT...REFERENCES |
|---|
MATERIALS AND METHODS: The study group consisted of 382 consecutive patients. All underwent baseline dual-sector spiral CT, and CT was repeated at 2 and 4 years later. Progression of CAC was assessed with measurement of the increase in total calcium score (TCS) and with repeated-measures analysis and multivariate linear regression models. Logistic regression model was used to predict incidence of new lesions.
RESULTS: Eighty-seven percent (333 of 382) of the study group were men, with mean age of 65 years ± 11, and 13% (49 of 382) were women, with mean age of 68 years ± 11. The average TCS increased after 4 years by sixfold from baseline in the 1st quartile, and by four-, two- and 1.5-fold in the 2nd, 3rd, and 4th quartiles of baseline TCS (P < .01), respectively. Multiple linear regression analysis included age; sex; natural logarithm of baseline TCS; history of hypertension, diabetes mellitus, current smoking, hypercholesterolemia, and lipid-lowering therapy with cholesterol synthesis enzyme inhibitor (statin); and family history of premature CAD. Results demonstrated that natural logarithm of baseline TCS and history of current smoking were independent predictors of the 4th-year natural logarithm of TCS levels (R2 = 0.85, P < .001). New lesions were diagnosed in 56 (15%) patients. History of statin therapy (odds ratio = 0.35; 95% confidence interval [CI]: 0.16, 0.77; P < .01), age with an increment of 5 years (odds ratio = 0.76; 95% CI: 0.64, 0.90; P = .01), and natural logarithm of baseline TCS (odds ratio = 0.73; 95% CI: 0.62, 0.86; P < .01) were independent predictors for new calcific lesions during 4 years.
CONCLUSION: Accelerated progression of CAC during 4 years was found in clinically stable patients with CAD.
© RSNA, 2004
Index terms: Computed tomography (CT), multi–detector row, 54.12119 • Coronary vessels, calcification, 54.81 • Coronary vessels, CT, 54.12119 • Coronary vessels, diseases, 54.76
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSTATISTICAL CONSULTANT...REFERENCES |
|---|
Serial scanning for assessment of the progression of CAC has been performed in different clinical groups, which include patients with CAD, individuals at high risk for CAD (17–20), patients with heart transplants (12,21), and patients with renal failure (22). In most of these studies, researchers followed up a limited number of patients for relatively short periods that ranged from 1 to 3 years. The purpose of our study, therefore, was to prospectively assess the 4-year progression rate of CAC in patients with clinically stable CAD with multi–detector row CT.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSTATISTICAL CONSULTANT...REFERENCES |
|---|
Clinical Data
Clinical history and conventional risk factors were recorded and included the following: age; sex; time of first diagnosis of CAD; history of hypertension (defined as known systolic blood pressure >140 mm Hg, diastolic blood pressure >90 mm Hg, or medically treated hypertension), hypercholesterolemia (defined as a history of at least one measurement of total cholesterol of >250 mg/dL [6.46 mmol/L]), diabetes mellitus (defined as current insulin or oral hypoglycemic agent treatment), or current smoking (defined as smoking at least 10 cigarettes per day); and family history of premature CAD (defined as CAD in a first-degree relative at younger than 60 years). The use of statin therapy in this cohort reflects a secondary prevention requirement, and such therapy was prescribed by the treating cardiologists for patients with a low-density-lipoprotein level of more than 100 mg/dL (2.59 mmol/L). Stable coronary disease was defined as no coronary events during 4 years of follow-up.
CT Protocol
We used the previously described protocol for multi–detector row CT, with the modified Agatston scoring method (4).
Image acquisition.—Imaging was performed by using a commercially available dual-sector spiral CT scanner (Twin; Philips, Cleveland, Ohio) and spiral scanning mode (without injection of contrast material or electrocardiographic triggering). Scanning time was 1 second for two continuous 2.5-mm sections and 15–22 seconds for the entire zone of interest that encompassed the whole heart. Examination was performed during a single, unforced, withheld inspiration. During helical scanning, with the tube rotating at one revolution per second and the table moving at 5 mm/sec with a 1:1 scanning pitch, images were obtained with an effective section thickness of 3.2 mm (a nominal section width of 2.5 mm) and a reconstruction increment of 1.5 mm (overlapping section method). Scanning was performed with 120 kVp and 210 mAs, standard resolution, and a 43-cm field of view. The total duration of the procedure was 10 minutes.
Determination of coronary calcification.— A calcific lesion was defined as an area within a coronary artery that had a CT attenuation above a threshold of 90 HU and that covered an area of at least 0.5 mm2. Regions of interest around all lesions were placed by the same experienced reader (J.S., with 10 years of experience) in all instances and were automatically analyzed by the workstation software. This reader evaluated all initial 2- and 4-year scans. Intraobserver variation of this reader was evaluated with a random sample of 30 consecutive measurements in which the reader was blinded to previous results. A modification of the Agatston scoring method was applied with a threshold of 90 HU instead of 130 HU, and the attenuation factor for each lesion was determined as follows: factor 1, 90–199 HU; factor 2, 200–299 HU; factor 3, 300–399 HU; and factor 4, >400 HU. A score for each region of interest was calculated automatically by multiplying the attenuation factor by the area. The total calcium score (TCS) was the combined sum of the lesion scores for all sections. All these sections were imaged and included circles of all regions of interest. These traces enable the reader to fix the starting level at the same level as in the baseline scan, as well as to accurately identify all baseline calcium lesions. The reader was blinded to the score of the previous scans at the time of follow-up reading.
Presence of CAC was defined as TCS greater than 0. Progression of CAC was defined as any increase in TCS and was calculated as follow-up TCS minus baseline TCS (score units). New lesions were defined as the presence of a new calcification at least 1 cm from a region of calcification on a previous scan.
Statistical Analysis
Data were analyzed with software (SPSS, version 11.0; SPSS, Chicago, Ill). Baseline characteristics were compared between men and women by using the independent t test for continuous variables and the 
2 test for categoric variables. Progression of calcification was measured with means and standard deviations and with medians and interquartile ranges in each quartile of baseline TCS in the 4-year follow-up. Repeated-measures analyses were conducted for progression measured with means, and nonparametric Friedman tests were conducted for progression measured with medians. Multivariate analysis for prediction of TCS after 4 years was performed with multivariate linear regression models. Because the distribution of TCS was skewed, a natural logarithm of TCS was introduced into the model. Small positive value (value = 1) was added to TCS = 0 to define the log-transformed values. The F test was conducted to determine the significance of the model, and the R2 of the model was calculated. Scatterplots with the fitted regression lines show the linear correlation between natural logarithm of TCS at baseline and natural logarithm of TCS after 2 and 4 years. Logistic regression model analysis was conducted to estimate the incidence of new lesions with the independent variables.
The Spearman correlation test was used to estimate the intraobserver variation.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSTATISTICAL CONSULTANT...REFERENCES |
|---|
|
Calcium Progression
After 4 years, 91% (348 of 382) of the patients had measurable progression of CAC. A lower score than that at baseline was found in 16% (60 of 382) of the patients after 2 years and in only 5% (19 of 382) of them after 4 years. An unchanged score (including those with TCS = 0 at baseline) was found in 6% (23 of 382) of the patients after 2 years and in 4% (15 of 382) of them after 4 years. The TCS levels at each point of measurement are presented in Table 2, according to quartiles of the baseline TCS. Consistent significant progression during the follow-up was found in all baseline TCS quartiles (P < .01). The mean and standard deviation of TCS was 464 ± 716 at baseline and increased to 621 ± 856 and 841 ± 1064 after 2 and 4 years, respectively (P < .001). The median levels of TCS at each point of measurement according to quartiles of baseline TCS are presented in Figure 1. After 4 years, the average TCS increased sixfold from baseline in the 1st quartile, and it increased four-, two-, and 1.5-fold in the 2nd, 3rd, and 4th quartiles, respectively. The median TCS progression for men and women are presented in Figure 2. No significant differences were found between the sexes for CAC progression. Multiple linear regression analysis (Table 3) included the following variables: age; sex; natural logarithm of baseline TCS; history of hypertension, diabetes mellitus, current smoking, hypercholesterolemia, and statin therapy; and family history of premature CAD. Results of this analysis demonstrated that natural logarithm of baseline TCS and history of current smoking were independent predictors of the 4th-year natural logarithm of TCS levels (R2 = 0.85, P < .001). Continuous progression was found during 4 years of follow-up; the slope of the predicted line of natural logarithm of TCS after 2 years based on the natural logarithm of baseline TCS (b = .91, P < .001) (Fig 3a) was similar to the slope of the natural logarithm of TCS after 4 years based on the natural logarithm of TCS after 2 years (b = .89, P < .001) (Fig 3b). A similar slope was found in men and women.
|
|
|
|
|
|
|
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSTATISTICAL CONSULTANT...REFERENCES |
|---|
An accelerated progression of CAC has been shown in patients with clinically known CAD compared with asymptomatic subjects without CAD (11,15,16). Janowitz et al (15) used electron-beam CT to quantify the progression of CAC in 25 subjects. After 1 year, the subjects with proved CAD had a 48% increase in calcific plaque volume compared with the asymptomatic subjects, who had a 22% increase in this value. In another work (11), the 1-year progression of CAC was compared between 26 asymptomatic subjects and 30 patients with known CAD by using multi–detector row CT. The percentage of increase in TCS from baseline was 15% in the asymptomatic patients compared with 49% in the patients with CAD (P < .01). The annual increase in TCS was not substantial among the asymptomatic subjects (from 54 ± 118 to 62 ± 104), compared with the marked increase from 324 ± 476 to 482 ± 590 in the patients with CAD. Similar accelerated progression of TCS was reported by Mohlenkamp et al (16), who used electron-beam CT to compare the progression of CAC between 17 patients with angiographically demonstrated obstructive CAD and 16 subjects without known CAD. The TCS increased after a mean period of 16 months from 138 ± 112 at baseline to 184 ± 112 in the patients with CAD and from 58 ± 90 to 66 ± 94 in the non-CAD group (33% vs 14%, respectively). An annual progression of CAC of 24% on average was found in 82 unselected healthy adults by Maher et al (23).
Rapid CAC progression, measured with multi–detector row CT and a technique identical to the one used in the present study, was found in high-risk hypertensive adults (mean age, 67 years) (10). In that cohort, the annual progression rate was high for patients with CAD (39%) and for those without CAD (41%). Moreover, in that study the annual increase of TCS (60%) was even higher in those in whom coronary events occurred during a 3-year follow-up period (14). In the present study, we found that the rapid progression of CAC was consistent during 4 years in patients with CAD even if they had not had any coronary events. After 4 years, the average TCS increased 1.5- to sixfold from baseline. Of particular interest is the finding that this high progression rate was similar in users and nonusers of statins.
Determinants of CAC Progression
Although some investigators suggested an influence of age and sex on CAC progression (24–26), others showed that these factors did not substantially affect CAC progression (18,19). Yoon et al (18) studied the progression of CAC with electron-beam CT in 217 asymptomatic women and 114 men (mean age, 57 years) who underwent scanning at least twice as part of a screening program. The mean interval between the time the first and the last scan was obtained was 25 months. The mean number of risk factors was 1.4. The amount of CAC present at the initial study was the most important determinant of calcium progression, whereas neither sex nor age was a substantial predictor of CAC progression. Yoon et al (18) concluded therefore that "calcium begets calcium." We confirmed this observation in the present study in patients with established CAD. Eusebio et al (27) studied the 5-year changes in CAC with electron-beam CT in 125 asymptomatic subjects and also found that increases in CAC correlated with the initial score (r = 0.84, P < .01). Similar findings were reported by Schmermund (28), who followed up 102 symptomatic subjects for 18 months and found that the baseline plaque burden determined the rate of CAC progression, whereas none of the conventional risk factors predicted progression. In the present study, we found that only the natural logarithm of baseline TCS and history of current smoking were independent predictors of the 4th-year natural logarithm of TCS levels (R2 = 0.85, P < .001) in a multiple linear regression analysis (Table 3), which included age; sex; natural logarithm of baseline TCS; history of hypertension, diabetes mellitus, current smoking, hypercholesterolemia, and statin therapy; and family history of premature CAD.
New Lesions
The ability to track CAC progression and to detect new lesions by using multi–detector row CT was shown in the patients in the present study as well as in other groups of patients. By using multi–detector row CT with our protocol, new lesions were identified in the coronary arteries of heart transplant survivors (12). Twenty-four consecutive patients with heart transplants had been serially scanned by using multi–detector row CT at baseline and 2 years later. Very minor new calcific lesions were identified in seven of the 20 survivors with a mean TCS of 6.7 ± 4.0. Similar results were reported by Barbir et al (21) by using electron-beam CT. In another study (13) with our protocol, 26 (27%) of the 96 hypertensive adults without known CAD and with no CAC at baseline developed new calcific lesions during a 3-year follow-up.
In the present study, we found that in our selected population new lesions tended to develop in patients with mild CAC, which characteristically is found in younger patients. Those who developed new lesions had significantly lower TCS at each scanning period than those who did not (P < .001). Furthermore, a significantly lower incidence of new lesions was found among those who received statin therapy compared with those who did not (12% vs 20%, P = .022). Stary (29) demonstrated that drastic reduction of blood cholesterol levels in rhesus monkeys during 3.5 years resulted in the disappearance of macrophages, macrophage foam cells, and lymphocytes and in the reduction of extracellular lipid from advanced lesions, whereas calcium deposits remained in the arterial wall and were not changed. In accordance with the pathologic findings of Stary, we found that CAC in patients with clinically stable CAD progresses rapidly despite statin therapy. The absence of a major effect of statins on CAC progression in our study contradicts the findings of other investigators (6,19) who found regression of CAC after statin therapy. In a recently published study (30), it was shown that 47% (171 of 366) of the lesions in patients who developed acute coronary events during 3 years of follow-up had only mild calcification. In the present study, we found that calcification progresses mainly with an increase in the volume of existing calcific lesions and confluence of adjacent lesions rather than with development of new lesions. This type of topographic change was first described in a pilot study by Janowitz and colleagues (15), who found that "comparison of serial studies showed that smaller calcific deposits often coalesced into single larger calcific deposits."
Imaging Atherosclerosis
Angiography is the currently used method for definition of coronary anatomy and for tracking the atherosclerotic process in patients with CAD. However, many deficiencies are inherent in this method (31–34), which involves acquisition of a simple two-dimensional projection of the vessel lumen. Measurement of the percentage of lumen stenosis represents the traditional method for characterization of angiographic lesion severity. In some circumstances, diffuse, concentric, and symmetric coronary atheromatosis affects the entire vessel and results in the angiographic appearance of a small but normal artery. The most salient example of the dissociation between quantitative angiography and clinical outcomes is provided by the studies about regression of atherosclerosis. A multitude of randomized trials with lipid-lowering medication employed both angiographic and clinical assessment and showed a negligible improvement of luminal caliber, typically an absolute difference of only 1%–3% (35). These same studies yielded a large reduction in death and/or acute coronary ischemic events. Thus, by tracking plaque progression with angiography, a significant amount of information is lost because only the protruding portion of the plaque is demonstrated, which could be a cardinal reason why minor or no changes were found in those studies. Intravascular ultrasonography (US) can overcome this limitation and can reveal atherosclerosis at coronary sites when no apparent disease is found at angiography (36–38). Although intravascular US offers the opportunity for important new insights into the disease process, it is invasive and expensive and it demands highly trained operators. Therefore, it cannot be used for a large asymptomatic population. Evaluation of the progression of the atherosclerotic process by using multi–detector row CT provides a promising noninvasive alternative.
Technical Aspects
With the 90-HU threshold instead of the traditional 130-HU threshold, a better sensitivity with equal specificity was yielded, compared with those of angiographic methods for investigation of obstructive CAD (4,5). Therefore, we believe that for this nongated technique, 90 HU is more adequate. When we started the present study, the new spiral devices capable of electrocardiographic gating were not available. Tracking the progression of CAC with a nontriggering technique could be achieved because of the high progression rate, which exceeds the intertest variability. In this study, reproducibility was not measured directly to avoid further radiation to participants. With the same method and algorithm, however, it has been demonstrated (39) that the mean intertest variability, expressed as a percentage, was 32% and ranged from 23% for a TCS greater than 100 to 54% for a TCS of 1–10. In the present study, the mean TCS increased sixfold in the 1st baseline quartile range and 1.5-fold in the 4th quartile.
Study Limitations
Several potential sources of bias were considered in our study. First, a common limitation to many observational studies is the absence of information concerning potential spontaneous or therapy-induced changes in cholesterol level and its fractions during the follow-up period. Second, in this study, we did not measure the serum lipid levels and their subtypes but rather used the reported information in regard to the use of statins and known high levels of cholesterol. Therefore, the differentiation between patients who reached the goal of a low-density-lipoprotein level of less than 100 mg/100 mL and those who did not could not be accomplished. Third, as we included only clinically stable patients, we could not confirm a previous observation that patients with CAD who experienced coronary events have a higher progression rate than their stable counterparts (14). In this study, we included a selected group of patients with stable CAD. Further studies are needed about how CAC progression is related to cardiac outcome and the mechanisms involved. Last, another limitation of our study was that dual-sector spiral CT scanners are being replaced by other units, and presumably reproducibility and accuracy may be somewhat higher with thinner sections and the use of 16-section CT.
In summary, findings of the present study provide further evidence that multi– detector row CT can be used as a noninvasive technique to measure the progression of the coronary atherosclerotic process.
|
STATISTICAL CONSULTANT COMMENTARY |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSTATISTICAL CONSULTANT...REFERENCES |
|---|
Usually, the intention of analysis of variance is to justify the conclusion that groups have different means. It is not enough, however, to simply describe how far apart the group means are. What matters is how far apart they are relative to the variability of individual observations within each group. Yes, it is odd that this statistical procedure is called analysis of variance when what is of interest is the analysis of means, but the statistical language is appropriate because with this procedure two components of variation are compared.
David Moore (The Basic Practice of Statistics, 2nd ed, New York, NY: Freeman, 2000; 506–518) summarizes the details as follows: In analysis of variance, one wants to know the following: Do the groups have a different mean? The question can be rephrased by stating two hypotheses: (H0) There is no relationship. All k populations have the same mean. (Ha) There is a hypothesized relationship. Not all of the k population means are the same.
One decides between these two hypotheses by comparing two measures of variation, each called a mean square, in the numerator and the denominator of the F ratio. A mean square is a more general form of a sample variance, which is the average of the squared deviations of observations from their mean. Thus, the name mean square is used.
With respect to the mean signal, the numerator of F is a mean square that is used to measure variation among the k means. That is, independent simple random samples are obtained from each of the k groups. The kth sample has size nk, sample mean 
k, and sample standard deviation sk. Since the basis of the hypothesized model is that the groups may have different means, the mean square for the model (MSm) is used to compare the sample means. The mean square for the model is thus used to compare each sample mean with the overall mean (the mean of all N observations together), called . To keep it simple, one can assume k = 3 groups. Thus, the mean square for the model is as follows:
|
In regard to mean noise, the mean square in the denominator of F is used to measure variation among individuals within each group. Within each of the example groups, the sample variances (
) already are estimates of this variation. However, a single estimate is needed, so one uses an average of these individual variances, and the average is called the mean square error or MSe, which is calculated as follows:
|
Thus, MSe is a weighted average in which the value for each
is weighted by one fewer than the number of observations it represents. That is, each value for
is weighted by its degrees of freedom, nk – 1. With MSe, the sample variances are pooled into a single average, which is why the square root of MSe is called the pooled standard deviation. This is where the assumption of equal group variability comes into play. That is, if all the values for
are not approximately equal, then averaging them into a single pooled estimate of error (noise) may be problematic.
The decision about whether there is or is not a relationship is made by using the F ratio, which is calculated thus:
|
A large F ratio reflects a model with a signal that is large when the signal is compared with noise. With certain assumptions, this ratio has the F distribution with degrees of freedom of k – 1 and N – k. We use this distribution to calculate P values.
The P value that results from the F ratio answers the question, do the data support the presumption that there is no relationship? In the analysis of variance example, "no relationship" implies that the means are equal, and so, informally, the P value indicates the probability that there actually was no relationship. If a P value is small, typically less than .05, then the signal-to-noise ratio is statistically significant; one concludes that there is a relationship. In analysis of variance, this observation translates into the conclusion that there is a mean difference, or in regression analysis, this observation indicates that there is a trend. If the P value is large, then the observed differences or trends are within what can be expected with the chance variation that can occur, even when there is no relationship. In this latter case, the conclusion is that there is no evidence for a relationship.
The defensibility of these P values depends on the correct specification of the model and the satisfaction of the assumptions. Specifically, these assumptions are independent observations, common variances, and normal errors. One example in which these assumptions are not met is that in which observations are obtained from repeated measures in the same subject. These observations are not independent. In this case, however, there are methods whereby the F distribution may be correctly applied. In addition, there are methods to identify nonconstant variance or nonnormal distribution and to remedy these difficulties. Unless there are outliers, the F ratio is not very sensitive to departures from normality.
Although formulated by G. W. Snedecor, the F distribution is named for Sir R. A. Fisher, the father of modern statistics. In 1922, he obtained the distribution of this statistic and applied it to the analysis of variance. The F distribution is a generalized form of the Student t distribution.
|
FOOTNOTES |
|---|
Authors stated no financial relationship to disclose.
Author contributions: Guarantor of integrity of entire study, J.S.; study concepts, J.S., M.M.; study design, J.S.; literature research, S.A., J.R.; clinical studies, J.S.; data acquisition and analysis/interpretation, N.K.M., J.S.; statistical analysis, N.K.M., J.S.; manuscript preparation, J.S., S.A.; manuscript definition of intellectual content, J.S., B.A.K., M.M.; manuscript editing, M.M., Y.I.; manuscript revision/review, M.M.; manuscript final version approval, J.S., N.K.M.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSTATISTICAL CONSULTANT...REFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  A. E. Cassidy-Bushrow, L. F. Bielak, P. F. Sheedy II, S. T. Turner, I. J. Kullo, X. Lin, and P. A. Peyser Coronary Artery Calcification Progression Is Heritable Circulation, July 3, 2007; 116(1): 25 - 31. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Shemesh, R. Evron, N. Koren-Morag, S. Apter, J. Rozenman, D. Shaham, Y. Itzchak, and M. Motro Coronary Artery Calcium Measurement with Multi-Detector Row CT and Low Radiation Dose: Comparison between 55 and 165 mAs Radiology, September 1, 2005; 236(3): 810 - 814. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
