| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Breast Imaging |
1 From the Department of Radiologic Sciences, Hahnemann University Hospital, 246 N Broad St, MS 206, Philadelphia, PA 19102 (L.W.N.); and Department of Radiology, University of Pennsylvania Medical Center, Philadelphia (M.D.S., S.G.O.). From the 1999 RSNA scientific assembly. Received January 25, 2000; revision requested March 3; final revision received August 16; accepted September 19. L.W.N. supported in part by the Susan G. Komen Breast Cancer Foundation. M.D.S. supported in part by National Institutes of Health grants RO1-CA58358 and P41-RR02305 and General Electric Corporation. Address correspondence to L.W.N.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Two hundred sixty-two new patients with palpable or mammographic abnormalities underwent MR imaging, and pathologic evaluation was performed. They were entered prospectively into the model, which yielded 454 patients in the construction (training) and validation (test) phases. Predictive values for previously published terminal nodes or branch points of the model were compared between the training and test data sets. Ductal enhancement morphology, regional enhancement micronodularity, regional enhancement degree, and focal mass T2 signal intensity were evaluated for model expansion. Diagnostic performance characteristics of the model were recalculated.
RESULTS: For previously published nodes, absence of a lesion visible at MR imaging, smooth masses, lobulated masses with nonenhancing internal septations, and lobulated masses with minimal or no enhancement had negative predictive values (NPVs) for malignancy similar in both data sets (96% vs 99%, 100% vs 93%, 100% vs 98%, and 100% vs 100%). Irregular masses with internal septations (100% vs 0%) and spiculated masses with no or minimal enhancement (100% vs 50%) did not. Nonseptated enhancing lobulated masses with low T2 signal intensity were added as a benign terminal node (NPV, 100%). Mild regional enhancement (NPV, 92%) was added but not considered a terminal node. Sensitivity, specificity, NPV, positive predictive value, and accuracy of the expanded model were 96%, 80%, 96%, 78%, and 87%, respectively.
CONCLUSION: Additional investigation yielded a slightly modified model, but the diagnostic performance characteristics remain high, similar to those originally published.
Index terms: Breast neoplasms, diagnosis, 00.121411, 00.121412, 00.121415, 00.12143, 00.30 • Breast neoplasms, MR, 00.121411, 00.121412, 00.121415, 00.12143, 00.30 • Computers, diagnostic aid • Magnetic resonance (MR), contrast enhancement, 00.12143, 00.30
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
The diagnostic performance characteristics of the originally validated and revised models were as follows: sensitivity, 100% and 96%; specificity, 69% and 79%; negative predictive value (NPV), 100% and 97%; positive predictive value (PPV), 75% and 76%; and overall accuracy, 83% and 86%, respectively (1).
The terminal nodes, or terminal branches, that were not considered adequately validated because of small samples sizes were irregular masses with internal septations and spiculated masses with no or minimal enhancement. In addition, minimal enhancement in focal masses was not validated by using cases entered prospectively (1).
This investigation builds on the research described. The purpose of this study was to (a) prospectively validate the nodes in the interpretation model with an increase in the patient sample size of the test data set; (b) expand the model to improve specificity while maintaining sensitivity, with identification of additional architectural features to serve as decision nodes; and (c) reevaluate the diagnostic performance of the model.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Eligibility Criteria
Eligible patients included women with palpable or mammographically visible lesions who presented to the University of Pennsylvania Medical Center and were scheduled for excisional biopsy or cyst aspiration. One hundred ninety-two patients (age range, 21–88 years; mean age, 49 years) were enrolled in the previous study. An additional 262 patients (age range, 18–85 years; mean age, 52 years) were enrolled for this study by using similar eligibility criteria; only patients presenting with mammographic or palpable abnormality were included in the study. The period for enrollment spanned from December 24, 1991, to April 16, 1999.
Malignant lesions were classified as invasive ductal, invasive lobular, invasive tubular, medullary, or colloid carcinoma or as ductal carcinoma in situ (DCIS). Benign histologic findings were classified as fibroadenoma, fibrocystic change, or other findings such as radial scar, lipoma, normal breast tissue, hyperplasia, and lobular carcinoma in situ.
MR Imaging Examination
MR imaging was performed with a 1.5-T magnet (Signa; GE Medical Systems, Milwaukee, Wis). Patients were positioned prone with the breast to be imaged in gentle compression within a four-coil-array surface coil designed by one of the authors (M.D.S.) for imaging the breast (13). This surface array coil is not commercially available, but similar signal-to-noise ratio can be achieved with commercially available breast coils, and at least one commercially available coil allows the breast to be imaged with compression. Compression of the breast allows a shorter imaging duration, but with the gradient speeds now commercially available, an entire uncompressed breast can be imaged within a time similar to that used in our original interpretation model publication (approximately 3 minutes 40 seconds for a single contrast-enhanced run).
During the course of the study, parameters evolved slightly as a result of improving technology, which allowed shorter imaging times with the same amount of T1 weighting and signal-to-noise ratio. The protocol sequences and the ranges of parameters used were as follows: sagittal T1-weighted spin-echo imaging (400–500/minimum to 15 [repetition time msec/echo time msec]; two signals acquired; section thickness, 3 mm; intersection gap, 1 mm), fat-saturated T2-weighted fast spin-echo imaging (4,000–6,000/105–120; two signals acquired), and dynamically enhanced fat-saturated fast multiplanar spoiled gradient-echo imaging (20–28/2.4–4.4 and a flip angle of 30°-60° has evolved to 9–10/2.0–2.4 and a flip angle of 35°–45° with continued 12–18-cm field of view, 512 x 256 matrix, 2–4-mm section thickness with no intersection gap, 300–700-µm in-plane resolution, and one signal acquired). The fast multiplanar spoiled gradient-echo fat-saturation technique has evolved from a standard chemical shift selective technique to a chemical shift selective fat-inversion technique that is now the standard three-dimensional fat-saturation technique used in the imager’s software package.
The contrast-enhanced images were obtained by using a three-dimensional volume acquisition in which 28 sections were used to cover the entire breast by means of varying the section thickness. Initially, imaging after the administration of contrast material lasted for a total of 3–4 minutes per run, and three postcontrast runs were performed, with interpretation emphasis placed on the first run. Now with the shorter repetition and echo times, imaging lasts for a total of 1.5 minutes per run, and four to five postcontrast runs are performed, with interpretation emphasis placed on the first and second runs. The intravenous injection of approximately 0.1 mmol of gadopentetate dimeglumine (Magnevist; Berlex Laboratories, Wayne, NJ) per kilogram of body weight to a maximum of 20 mL has been liberalized over time to an injection of 20 mL per patient, which yields a dose of approximately 0.15 mmol/kg for most patients. Contrast material was injected during a 10-second interval. Imaging was begun after the injection of contrast material and prior to the injection of a 10-mL saline solution flush.
Model Validation
For the purposes of this study, a true-positive finding was defined as a malignant lesion that was correctly classified, and a true-negative finding was a benign lesion correctly classified. A false-positive finding was a benign lesion incorrectly classified, and a false-negative finding was a malignant lesion incorrectly classified.
Several general rules used in constructing the original interpretation model were reapplied: (a) The radiologists classified the borders as being smooth, lobulated, irregular, or spiculated by evaluating the most malignant-appearing aspect of the border, not the most prevalent characteristic of the border. (b) Because ductal enhancement had such a high past PPV for malignancy, again, any case displaying this feature, regardless of other features, was considered suggestive of malignancy. (c) Since regional enhancement had not been particularly representative of benign or malignant disease, again, if a case displayed this feature in combination with any other feature, the other feature was emphasized. (d) Minimal enhancement of a focal mass was defined as enhancement less than or equal to that of the surrounding breast fibroglandular tissue.
The initial model was constructed retrospectively by using a training data set of 98 patients; the method used was detailed in the original publication (1) and represents a form of recursive partitioning analysis (14,15). In the original publication (1), a test data set of 94 patients was then entered prospectively to begin validation of the model. For this study, 262 new patients were entered prospectively to enlarge the test data set being used to validate the model. The methods being used to construct and validate the model attempt to follow the recommendations published by Wasson et al (14) for constructing and validating clinical prediction rules.
The NPVs and PPVs of the benign terminal nodes in the model were calculated for the training and test data sets. The 2 x 2 contingency tables containing the data from which these predictive values were derived were evaluated for statistical significance as a preset criterion for node validation. An absence of a statistically significant difference demonstrated for the 2 x 2 table was considered proof of validity of the terminal node being studied. Statistical significance of the contingency tables was assessed by using a Fisher exact test as long as the sample size in the node was adequate to justify formal statistical analysis. We did not perform formal statistical testing on terminal nodes with a sample size of less than five, since diagnostic categories with such small sample sizes have been empirically and theoretically shown to have misclassification rates too high (of too little statistical power) to be considered valid (14). To our knowledge, no single statistical method has been established as the reference standard for comparing the results of the training and test data sets of a prediction rule (14).
Model Expansion
The initial interpretation model was built in a stepwise fashion so that a decision tree–shaped format would show which individual architectural features occupied the decision nodes (1). To occupy each successive node in the model, the architectural feature was chosen that would isolate the largest number of patients with a high NPV for malignancy. If a group of patients with a chance greater than 95% of having benign disease (<5% chance of malignancy) was isolated, then this was considered a benign terminal node, and no further subdivision was attempted. If the residual patients in the group had a risk of malignancy greater than or equal to 5%, then further subcategorization by using other architectural features was attempted (1). These same guidelines and the data from all 454 patients were used in our attempts to expand the model.
Four new architectural features were evaluated: ductal enhancement morphology (linear vs branching), focal mass T2 signal intensity (low vs intermediate or high), regional enhancement micronodularity (stippled vs not stippled), and degree of regional enhancement (mild vs moderate or marked, where mild enhancement was defined as enhancement equal to or slightly greater than that of breast fibroglandular tissue). Their individual diagnostic performance characteristics (sensitivity, specificity, NPV, PPV, and overall accuracy) were calculated by means of comparing whether the radiologist interpreting the examination thought the feature was present with whether the patient subsequently had malignancy. The NPVs were then used to assess the appropriateness of each feature for incorporation into the model. For our prior article (1) and for the majority of patients involved in this study, the radiologists (M.D.S. and S.G.O.) concurrently interpreted the studies, but recently they conferred only on cases they considered questionable; it is impossible to determine exactly when this transition occurred.
Model Diagnostic Performance
By using the data for all 454 patients, the sensitivity, specificity, NPV, PPV, and overall accuracy of the model were calculated by classifying data from all the patients in nodes with an NPV for malignancy of at least 95% as testing benign and data from all other patients as testing malignant. In addition, the individual diagnostic performance characteristics of architectural features associated with the model were calculated to determine whether incorporation into the model improved their predictive capabilities.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
|
|
Of the previously published nodes, four terminal nodes had NPVs for malignancy that were similar in the training and test data sets and had node sample sizes of at least five patients. These included (a) absence of a lesion visible at MR imaging (96% [22 of 23] vs 99% [68 of 69]); (b) smooth masses (100% [12 of 12] vs 93% [26 of 28]); (c) lobulated masses with nonenhancing internal septations (100% [11 of 11] vs 98% [39 of 40]); and (d) nonenhancing or minimally enhancing lobulated masses (100% [11 of 11] vs 100% [seven of seven]). These four nodes were considered validated and were retained as benign terminal nodes. See Table 1, including its note, for additional details on how these percentages were derived.
The training data set NPV for absence of a lesion visible at MR imaging is listed as 88% (22 of 25) in Table 1 but, as explained in the table note, we think 96% (22 of 23) is more reflective of this node because two of the three missed cancers were not located in the region of the breast chosen for dynamic enhancement when we used a two-dimensional enhancement technique that did not cover the entire breast. A more detailed explanation can be found in our prior publication (1). We now routinely enhance the entire breast by using the three-dimensional technique described in the Materials and Methods section.
Two previously published terminal nodes did not have test set NPVs for malignancy similar to those seen in the training set. These two terminal nodes also did not achieve a total sample size of at least five patients. These nodes were not considered valid and were deleted as terminal nodes. These included irregular masses with internal septations (100% [one of one] vs 0% [zero of two]) and nonenhancing or minimally enhancing spiculated masses (100% [two of two] vs 50% [one of two]).
Figure 2 illustrates the previously published terminal nodes that did not meet our validation criteria, as well as some of the expansion attempts that did not meet our expansion criteria. Figure 2 is included solely to show the data it contains and does not represent the model as we recommend it be used.
|
|
In the ductal enhancement node, 14 of the 28 cancers were associated with either irregular or spiculated masses. None of the seven cases of DCIS alone were associated with a focal mass. In Table 2, the ductal enhancement "other cancer" category included two invasive lobular carcinomas, one combined invasive ductal and invasive lobular carcinoma with DCIS, and one predominately intraductal adenocarcinoma.
In Table 2, "other cancers" displaying regional enhancement were invasive lobular carcinoma (n = 2), papillary adenocarcinoma (n = 1), and angiosarcoma (n = 1).
Both cancers with smooth borders were colloid cancers. Figure 3 shows images obtained in one of these patients. Both of the colloid cancers showed minimal or no overall enhancement, another finding more characteristic, in isolation, of benign abnormalities than of malignancies. One of these colloid cancers displayed rim enhancement, a feature classically associated with malignancies. We encountered two other colloid carcinomas in our study population. Of these, one manifested as a lobulated mass with moderate to marked enhancement and the other as an irregular rim-enhancing mass with internal septations and moderate overall enhancement.
|
|
|
|
|
With larger sample sizes, the presence of nonenhancing internal septations in an irregular mass no longer appeared reflective of benign abnormalities, as we had previously hoped. Pathologic evaluation of the three septated irregular masses revealed one intraductal papillary carcinoma, one colloid carcinoma with DCIS, and one fibrocystic change. The category of no or minimal enhancement in an irregular mass was also not reflective of benign disease, as we had hoped. The four irregular masses in this category displayed minimal enhancement and were one invasive ductal carcinoma, one invasive ductal and invasive lobular carcinoma, and two fibrocystic changes. Of note, no irregular malignancies were seen that did not enhance at all. Similarly, low T2 signal intensity in an irregular mass did not reflect benignity. All three irregular masses with low T2 signal intensity were malignant: two invasive ductal carcinomas with DCIS and one invasive lobular carcinoma that is shown in Figure 6.
|
The category of no or minimal enhancement in a spiculated lesion did not, with larger sample sizes, reliably reflect benign abnormalities and represented malignancy in one of four cases. Pathologic findings in the four cases were one invasive ductal carcinoma that had been excluded from enhanced images, one radial scar, and two fibrocystic changes. Again of note, no spiculated malignancy was seen that did not enhance at all on what was considered an adequate study. Low T2 signal intensity did not reliably suggest benign abnormality in spiculated masses. Of the two masses with low T2 signal intensity, one was DCIS and the other was fibrocystic change.
Model Expansion
Of the four new architectural features tested, low T2 signal intensity in focal masses and mild degree of regional enhancement were strongly associated with benign disease; their individual NPVs for malignancy were 81% (25 of 31) and 92% (11 of 12), respectively. The other two features investigated (ductal enhancement morphology, NPV of 14% [two of 14] and regional enhancement micronodularity, NPV of 64% [16 of 25]) did not allow us to sufficiently distinguish benign from malignant breast abnormalities to be incorporated into the model. Figures 5–12 show examples of the four architectural features tested. Table 3 summarizes the individual performance characteristics of the four features. Mild regional enhancement warranted incorporation into the model. However, its NPV was less than 95% and, thus at this point, we do not consider it a terminal node.
|
|
|
|
|
|
|
Model Diagnostic Performance
The diagnostic performance characteristics of the interpretation model calculated without the two new architectural features (mild regional enhancement and low T2 signal intensity in lobulated masses) but including all 454 patients, were sensitivity, 96% (185 of 192); specificity, 75% (196 of 262); NPV, 97% (196 of 203); PPV, 73% (185 of 252); and overall accuracy, 84% (381 of 454).
The diagnostic performance characteristics of the expanded interpretation model, including the newly identified features and all 454 patients, were sensitivity, 96% (184 of 192); specificity, 80% (210 of 262); NPV, 96% (210 of 218); PPV, 78% (184 of 237); and overall accuracy, 87% (394 of 454).
Including all 454 patients, the predictive values of the individual architectural features associated with the model were calculated to determine their predictive capabilities if they were used outside the model. These values are displayed in Table 4.
|
1. Fifty-seven of the 454 patients had nonenhancing internal septations, and 54 of these were benign, which yielded an individual NPV for malignancy of 95% (54 of 57), compared with a model NPV of 98% (50 of 51), where its use is restricted to lobulated focal masses.
2. Lack of or minimal enhancement in a focal lesion led to prediction of benignity in 92% (60 of 65) of all cases, versus 100% (18 of 18) when restricted by the model to lobulated nonseptated masses.
3. Low T2 signal intensity led to prediction of benignity in only 81% (25 of 31) of all cases, compared with 100% (three of three) when restricted by the model to lobulated nonseptated enhancing masses.
Although not associated with the model, we also calculated the individual PPV of peripheral rim enhancement and again found it to be highly predictive of malignancy, with a predictive value of 86% (103 of 120). Of the benign rim-enhancing lesions, two fibroadenomas and one fibrocystic change manifested as lobulated masses with nonenhancing internal septations, and two fibroadenomas, one fibrocystic change, and one other benign histologic finding manifested as smooth masses. The other 10 were classified falsely within the interpretation model as being malignant.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Absence of a lesion visible at MR imaging, that is, absence of any discrete abnormality at MR imaging to correspond to the palpable or mammographically visible abnormality, remained highly predictive of benign findings, with an NPV for malignancy of 96% (90 of 94). Thus, it was retained as a terminal node. As before, the true NPV for this node is thought to be higher (98% [90 of 92]) because two of the cancers were not imaged after the administration of contrast material owing to a limitation in our early two-dimensional technique, which does not exist with our current three-dimensional technique (1).
Smooth masses (NPV for malignancy of 95% [38 of 40]), lobulated masses with nonenhancing internal septations (NPV for malignancy of 98% [50 of 51]), and lobulated masses that displayed no or only minimal contrast enhancement (NPV for malignancy of 100% [18 of 18]) also remained highly predictive of benign findings and were retained as terminal nodes. Malignancies that masqueraded in these three terminal nodes tended to be rare but predictable diseases. The two smooth masses were colloid cancers, and the one lobulated septated mass was an adenoid cystic cancer. Unfortunately, no architectural feature could be identified that routinely distinguished these cancers from the benign entities they mimicked.
The category of lobulated enhancing masses with low T2 signal intensity was added as a benign terminal node. The presence of low T2 signal intensity in lobulated (or smooth) focal masses was not prevalent enough to warrant it replacing any of the previous nodes in the model. However, its individual NPV for malignancy of 81% (25 of 31) was high enough to warrant attempted incorporation into the model as a subcategory of nonseptated enhancing lobulated masses. As a node in the model, its NPV increased to 100% (three of three), although the sample size was small (n = 3).
For irregular and spiculated masses, nonenhancing internal septations, minimal enhancement, and low T2 signal intensity were not suggestive of benign abnormalities. Although it was hoped with the originally published model that internal septations and absent or minimal enhancement would be benign subcategories for irregular and spiculated masses, the increased sample sizes needed to validate these terminal nodes did not support their retention, and they are no longer considered worthwhile subcategories. Lack of enhancement in irregular and spiculated masses may still reliably lead to prediction of benign abnormalities, but sample sizes adequate to prove this were not achieved with this study. Low T2 signal intensity was not an effective subcategory of irregular or spiculated masses.
Of the four new architectural features tested, only low T2 signal intensity in focal masses (NPV of 81% [25 of 31]) and mild degree of regional enhancement (NPV of 92% [11 of 12]) were strongly associated with benign abnormalities; ductal enhancement morphology and regional enhancement micronodularity were not. Mild regional enhancement was incorporated into the model. However, its NPV was not more than 95% and, thus, we do not consider it a terminal node. As described earlier, low T2 signal intensity in lobulated masses was incorporated as a benign terminal node.
Certain architectural features remained highly predictive of malignancy, including ductal enhancement (PPV for malignancy of 85% [28 of 33]), spiculated borders (PPV for malignancy of 91% [72 of 79]), irregular borders (PPV for malignancy of 84% [52 of 62]), and peripheral rim enhancement (PPV for malignancy of 86% [103 of 120]).
The diagnostic performance characteristics of the interpretation model calculated without the two new architectural features (mild regional enhancement and low T2 signal intensity, lobulated masses), were sensitivity, 96% (185 of 192); specificity, 75% (196 of 262); NPV, 97% (196 of 203); PPV, 73% (185 of 252); and overall accuracy, 84% (381 of 454). These findings are consistent with the values originally published with the model and, we believe, strongly support the model’s validity.
The diagnostic performance characteristics of the updated interpretation model (including the new features) were sensitivity, 96% (184 of 192); specificity, 80% (210 of 262); NPV, 96% (210 of 218); PPV, 78% (184 of 237); and overall accuracy, 87% (394 of 454). The improvement in specificity with minimal change in sensitivity suggests that the identification of new architectural features is feasible and worthwhile. This study did not attempt to validate the new benign terminal node that resulted from our model expansion attempts (low T2 signal intensity in enhancing lobulated masses). We continue to accrue new patients, now from two separate medical centers, and plan to use that data to validate the newly added node.
In addition to validating the new terminal node, other important aspects of validating the interpretation model still remain. Its universality needs to be proved by (a) using different readers and (b) using different MR machinery. We also hope to prove the model’s universality with our ongoing investigations. However, at this point, it is reassuring that with increased sample sizes the main premises of the interpretation model remained true.
We still believe that architectural feature analysis and, in particular, our interpretation model based on architectural features, is an important tool for distinguishing benign from malignant breast MR abnormalities. Its diagnostic performance already rivals those reported by investigators with both similar and different philosophic approaches and, as we have shown, our model has potential for continued growth and improvement.
Reported sensitivities for breast MR imaging have been consistently high, usually in the range of 94%–100% (3,16). The sensitivity of our model (96%) is similar to these. Reported specificities have been generally lower and highly variable, ranging from 37% to 97% (1,3,4,16–19). The true specificity of our model, which should be in the range of 75%–80%, is still surpassed by those reported by some investigators who have concentrated on contrast enhancement kinetics. Kaiser and Zeitler (17) reported specificities as high as 97% by using a rapid rate of enhancement as the sole criterion for malignancy. Kuhl et al (18) found a specificity of only 37% when they used enhancement rate alone as the criterion, but they achieved a specificity of 83% by using the shape of the signal intensity time course curve. It should be noted, however, that readers in the article by Kuhl et al (18) were shown a single image of the MR abnormality from which they could have gleaned some rudimentary architectural feature data and that the authors themselves recommend the use of both architectural and time course data to optimize accuracy of interpretation.
The few studies (7,18) in which architectural feature versus enhancement kinetic analysis were directly compared in the same patients suggest that architectural feature analysis will remain an important part of any breast MR evaluation. Mussurakis et al (7) compared the use of architectural features with the use of contrast enhancement kinetics and found architectural features to be superior to quantitative kinetic indices alone (P = .02) (3). They studied lesion conspicuity, signal intensity, contour, and enhancement pattern.
An expected true NPV of more than 95% has been our criterion for a benign terminal node, and we have hoped that abnormalities with sufficiently high NPVs might not require biopsy but instead might be followed up for stability or resolution. However, whether MR imaging will actually be able to help reduce the number of biopsies performed in benign tissue will depend on a number of things: (a) the NPV that society thinks is acceptable and the NPV that an individual patient thinks is acceptable in the diagnosis of breast cancer and whether MR imaging can achieve these NPVs for a given palpable or mammographically visible finding; (b) patient preferences concerning the comparative diagnostic methods, none of which are 100% accurate; and (c) the cost-effectiveness of available alternatives at any given time. Nevertheless, if breast MR imaging fulfills its potential as a breast cancer staging tool and screening modality for patients at high risk, many incidental breast MR abnormalities will be encountered for which this interpretation model should be useful.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Author contributions: Guarantors of integrity of entire study, L.W.N., M.D.S.; study concepts and design, L.W.N., M.D.S.; literature research, L.W.N., S.G.O.; clinical studies, L.W.N., M.D.S., S.G.O.; data acquisition and analysis/interpretation, L.W.N., M.D.S., S.G.O.; statistical analysis, L.W.N.; manuscript preparation and definition of intellectual content, L.W.N.; manuscript editing, revision/review, and final version approval, L.W.N., M.D.S., S.G.O.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  L. J. Solin, S. G. Orel, W.-T. Hwang, E. E. Harris, and M. D. Schnall Relationship of Breast Magnetic Resonance Imaging to Outcome After Breast-Conservation Treatment With Radiation for Women With Early-Stage Invasive Breast Carcinoma or Ductal Carcinoma in Situ J. Clin. Oncol., January 20, 2008; 26(3): 386 - 391. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N. H. G. M. Peters, I. H. M. Borel Rinkes, N. P. A. Zuithoff, W. P. T. M. Mali, K. G. M. Moons, and P. H. M. Peeters Meta-Analysis of MR Imaging in the Diagnosis of Breast Lesions Radiology, December 1, 2007; 246(1): 116 - 124. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. A. Harvey, R. E. Hendrick, J. M. Coll, B. T. Nicholson, B. T. Burkholder, and M. A. Cohen Breast MR Imaging Artifacts: How to Recognize and Fix Them RadioGraphics, October 1, 2007; 27(suppl_1): S131 - S145. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. Kuhl The Current Status of Breast MR Imaging * Part I. Choice of Technique, Image Interpretation, Diagnostic Accuracy, and Transfer to Clinical Practice Radiology, August 1, 2007; 244(2): 356 - 378. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. J. Macura, R. Ouwerkerk, M. A. Jacobs, and D. A. Bluemke Patterns of Enhancement on Breast MR Images: Interpretation and Imaging Pitfalls RadioGraphics, November 1, 2006; 26(6): 1719 - 1734. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Tozaki, T. Igarashi, and K. Fukuda Breast MRI using the VIBE sequence: clustered ring enhancement in the differential diagnosis of lesions showing non-masslike enhancement. Am. J. Roentgenol., August 1, 2006; 187(2): 313 - 321. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. Erguvan-Dogan, G. J. Whitman, A. C. Kushwaha, M. J. Phelps, and P. J. Dempsey BI-RADS-MRI: a primer. Am. J. Roentgenol., August 1, 2006; 187(2): W152 - W160. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. G. Orel, M. Rosen, C. Mies, and M. D. Schnall MR Imaging-guided 9-gauge Vacuum-assisted Core-Needle Breast Biopsy: Initial Experience Radiology, December 1, 2005; 238(1): 54 - 61. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
F. Pediconi, C. Catalano, R. Occhiato, F. Venditti, F. Fraioli, A. Napoli, M. A. Kirchin, and R. Passariello Breast Lesion Detection and Characterization at Contrast-enhanced MR Mammography: Gadobenate Dimeglumine versus Gadopentetate Dimeglumine Radiology, October 1, 2005; 237(1): 45 - 56. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. K. Kuhl, H. H. Schild, and N. Morakkabati Dynamic Bilateral Contrast-enhanced MR Imaging of the Breast: Trade-off between Spatial and Temporal Resolution Radiology, September 1, 2005; 236(3): 789 - 800. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Ghai, D. Muradali, K. Bukhanov, and S. Kulkarni Nonenhancing Breast Malignancies on MRI: Sonographic and Pathologic Correlation Am. J. Roentgenol., August 1, 2005; 185(2): 481 - 487. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. E. Deurloo, S. H. Muller, J. L. Peterse, A. P. E. Besnard, and K. G. A. Gilhuijs Clinically and Mammographically Occult Breast Lesions on MR Images: Potential Effect of Computerized Assessment on Clinical Reading Radiology, March 1, 2005; 234(3): 693 - 701. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. A. Bluemke, C. A. Gatsonis, M. H. Chen, G. A. DeAngelis, N. DeBruhl, S. Harms, S. H. Heywang-Kobrunner, N. Hylton, C. K. Kuhl, C. Lehman, et al. Magnetic Resonance Imaging of the Breast Prior to Biopsy JAMA, December 8, 2004; 292(22): 2735 - 2742. [Abstract] |
