| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Breast Imaging |
1 From the Departments of Radiology (A.T., A.H., T.B., T.W.V., S.S., M.T.), Gynaecology (M.S.), and Pathology (H.A.L.), Johannes Gutenberg University of Mainz, Langenbeckstrasse 1, D-55131 Mainz, Germany. Received March 2, 2001; revision requested April 16; final revision received February 6, 2002; accepted March 6. Supported by a grant from the Deutsche Forschungsgemeinschaft (Th 315/7-1). Address correspondence to A.T. (e-mail: teifke@radiologie.klinik.uni-mainz.de).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Breast tissue specimens were obtained in 464 of 967 patients who had undergone dynamic gadolinium-enhanced T1-weighted fast low-angle shot three-dimensional MR imaging of both breasts. A comparison of sensitivity, specificity, and predictive values of the prospectively recorded findings of mammography, ultrasonography (US), and MR imaging with the histopathologic results was performed with receiver operating characteristic (ROC) curve analysis. MR imaging examination findings that caused a false-negative diagnosis were reviewed to identify possible sources of error.
RESULTS: Histopathologic analysis revealed 244 benign and 354 malignant lesions. The sensitivity values for mammography, mammography combined with US, MR imaging alone, and the combination of all three modalities were 73.7%, 88.1%, 88.4%, and 95.5%, and the areas under the ROC curves were 0.744, 0.829, 0.850, and 0.876, respectively. Twenty-eight (8.4%) of 334 invasive and 13 (65%) of 20 intraductal carcinomas were missed with MR imaging. In eight cases, motion artifacts (n = 1), tumor location near or beyond the outer boundary of the field of view (n = 3), inadequate infusion of the contrast material (n = 1), and masking of the tumors by intensively enhanced surrounding glandular tissue (n = 3) were identified as adequate explanations for the false-negative results. The remaining missed breast cancers (n = 33) exhibited very diffuse growth patterns or were 5 mm or smaller.
CONCLUSION: MR imaging did not depict 41 of 354 malignant tumors for several reasons.
© RSNA, 2002
Index terms: Breast, cysts, 00.3199, 00.721 • Breast, MR, 00.121411, 00.121412 • Breast neoplasms, 00.311, 00.327, 00.81, 00.813 • Breast neoplasms, diagnosis, 00.327, 00.81, 00.813 • Breast neoplasms, US, 00.1298 • Breast radiography, comparative studies, 00.11
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
The purpose of our study was to assess the prevalence and characteristics of malignant breast lesions that were not identified by using MR imaging and to evaluate the diagnostic value of mammography and ultrasonography (US) in these cases.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
The indications for MR imaging are listed in Table 1. The most common indication was the search for additional tumor foci (41% of examinations) where a high degree of suspicion for malignancy already existed on the basis of results of conventional breast imaging.
|
The study protocol was approved in its entirety by our regional review board. All patients gave informed consent to participate in this study.
Conventional Diagnostic Imaging
In all patients, high-quality mammograms (obtained with different equipment and films, because from different referral centers) were available. Prior to MR imaging, these mammograms had been reviewed, and a clinical examination and US (Sonolayer SSA-250A with SMA-736SA annular array probe; Toshiba Medical Systems, Tokyo, Japan) of both breasts had been performed by one of three radiologists (A.T., A.H., S.S.) experienced in all methods of breast imaging.
In our clinic, US examination is a standard procedure in all patients with mammographic abnormalities or inhomogeneous and dense breast parenchyma. US was performed as transverse and sagittal scanning of both breasts. Furthermore, areas suggestive of breast cancer were scanned in a radial and antiradial orientation with and without compression. Criteria for malignancy were ill-defined or irregular borders of solid lesions, posterior acoustic shadowing, architectural distortion, and solid intracystic or intraductal lesions.
In chronologic order, each of these examiners recorded the findings prospectively and rated every lesion that was discovered by using mammography and/or US separately and collectively by using BI-RADS analogous categories (10) for final assessment. These were as follows: BI-RADS category 1, negative; BI-RADS category 2, benign finding; BI-RADS category 3, probably benign finding; BI-RADS category 4, suspicious abnormality; and BI-RADS category 5, highly suggestive of malignancy. Categories 1 to 3 were considered benign, and categories 4 and 5 were considered malignant for the calculation of sensitivity and specificity.
MR Imaging
MR imaging was performed with a 1.0-T imager (Magnetom Impact Expert 42SP/AS; Siemens Medical Systems, Erlangen, Germany) and the manufacturer’s double-breast coil with the patient in the prone position. At first, T2-weighted coronal turbo spin-echo images (repetition time msec/echo time msec, 5,432/90; field of view [FOV], 175 x 350; resolution, 1.39 x 1.38 x 3 mm) were obtained. The main part of the protocol was performed with a dynamic contrast-enhanced coronal T1-weighted fast low-angle shot three-dimensional (3D) sequence (15/7; flip angle, 30°; FOV, 350 x 175 x 119; spatial resolution, 1.82 x 1.37 x 1.86 mm; sections, 64) with an acquisition time of 93 seconds. One measurement was acquired before and five measurements after manual bolus injection of gadopentetate dimeglumine (Magnevist; Schering, Berlin, Germany) in a dose of 0.1 mmol per kilogram of body weight. The bolus was injected into an antecubital vein, followed by a flush with 30 mL of saline. The injection was performed during a 5-second interval between the first two measurements. This dynamic examination was followed by that with a T1-weighted sequence and a higher spatial resolution (24/12; flip angle, 30°; FOV, 320 x 160 x 116; resolution, 0.66 x 0.99 x 1.0 mm).
The system software was used to calculate subtraction images from the first as well as the third postcontrast study and the precontrast dynamic study. T2-weighted images and subtraction images were documented on film. For enhancing lesions, intensity-time curves were calculated in regions of interest with a workstation (SPARC Station; Sun Microsystems, Mountain View, Calif) and a software package (MRVision; MRVision, Menlo Park, Calif). The regions of interest (2–9 pixels) were placed within the tumor area with the highest signal intensity enhancement by one of three radiologists (A.T., A.H., S.S.).
In addition, all data were viewed on a workstation (ISG viewing station; ISG Technologies, Mountain View, Calif), where interactive reconstructions in other planes, maximum intensity projections, and cine mode were possible.
The MR imaging examinations were evaluated prospectively by two of the three radiologists in a consensus reading by using the same rating scale described earlier. Mammographic and US findings were available during evaluation. Absence of contrast medium uptake or a bilateral slow and diffuse enhancement was considered negative for cancer. A unilateral diffuse or segmental enhancement was considered suspicious for cancer. In the case of focal enhancement, criteria for a benign lesion were a steadily rising time-signal intensity curve, well-circumscribed shape, internal septations, and homogeneous enhancement. Criteria for malignancy were maximum signal intensity in the first two postcontrast measurements; a stellate, irregular, or linear configuration; and an inhomogeneous or rim enhancement. A lesion was rated as probably malignant if it revealed morphologic or kinetic features of malignancy and as highly suggestive of malignancy in cases where both features were present.
A final rating was performed for every lesion by using the information of all three diagnostic methods. Furthermore, the radiologists subjectively evaluated the degree to which motion artifacts (none, weak, medium, strong) affected the 464 MR imaging examinations. They also estimated the degree (none, moderate, strong) of general enhancement in the second postcontrast measurement in the surrounding glandular tissue.
Histopathologic Correlation
Nonpalpable lesions were marked with a guide wire before surgery, and this procedure included MR-guided localization by using a special coil (Biopsy Mamma Coil; Noras, Würzburg, Germany). Tissue processing and interpretation of the slides were performed in the same pathology laboratory. All excised specimens were marked in three dimensions, sectioned serially in 5-mm layers, embedded entirely, placed on 5 x 5-cm slides, and stained by using the hematoxylin-eosin stain, according to standard protocols. Only breast specimens from patients who underwent complete mastectomy were not submitted in entirety. In this case, all areas that were macroscopically suspicious for cancer or areas that were localized with the guide wire were sectioned and examined histologically. Moreover, multiple representative sections from all quadrants and the retroareolar region were removed.
The diameters of all focal malignant lesions were measured on the histologic slices. Finally, all clinical colleagues (A.T., A.H., S.S., H.A.L., M.S.) reviewed and compared the histologic slides and the MR and conventional images during interdisciplinary sessions. In addition, this team reviewed the accuracy of the written diagnostic reports.
Analysis of False-Negative MR Imaging Findings
In the event that the malignant tumors were not correctly described in the MR imaging reports, a second retrospective analysis of the same MR images was performed by two radiologists (A.T., A.H.) to find out whether the missed lesion really was not visible or if there was an interpretation error. If the tumors could also not be found retrospectively, we searched for technical errors. If technical problems could be ruled out, the histomorphologic patterns (diameter, growth pattern) of the malignancies were reviewed to find an explanation for the inability to identify the tumor at MR imaging.
Data Analysis
Sensitivity, specificity, and positive and negative predictive values were calculated for each diagnostic method by using software (SPSS for Windows, release 10.0.7; SPSS, Chicago, Ill). BI-RADS categories 1 to 3 were considered benign, and categories 4 and 5 were considered malignant for the calculation of sensitivity and specificity. In addition, a receiver operating characteristic curve analysis was performed by using the same software to demonstrate the different sensitivity and specificity pairs that were based on the BI-RADS categories. The area under the receiver operating characteristic curve (Az) was calculated by using the nonparametric method as described in the SPSS manual to evaluate the global accuracy of the different imaging modalities.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
|
|
|
|
|
|
|
|
|
|
|
Of the 28 invasive lesions missed by using MR imaging, 11 were not seen at mammography or US but were accidentally discovered at histologic examination.
The remaining 13 of the 41 malignant tumors missed at MR imaging were DCIS with different histopathologic grades and diameters from 3 to 5 mm (Table 4). Seven of the 13 cases of DCIS were depicted owing to microcalcifications by using mammography, and one was depicted as an intracystic growing lesion by using US. The remaining five cases of DCIS were only discovered in the vicinity of invasive carcinomas at pathologic examination.
Movement artifacts were entirely absent in only 167 (36%) of 464 MR imaging examinations. In these cases, the artifacts were scored as weak in 209 (45%), as medium in 56 (12%), and as strong in 32 (7%) of the 464 examinations. In 195 (42%) of the 464 examinations, no early enhancement was seen in the glandular tissue. Moderate uptake of the contrast agent was observed in 218 (47%), and strong pooling was seen in 51 (11%) of the 464 examinations.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
With these restrictions, the sensitivity of MR imaging was 88.4% in 354 malignant lesions (including 20 cases of DCIS), with a specificity of 59.4%. In the study by Harms and co-workers (6), a precise pathohistologic correlation was also used as the basis for evaluating the diagnostic value of MR imaging. The authors compared 30 mastectomy specimens from patients with high-grade suspicion of malignancy and achieved a sensitivity of 94% in 47 malignant lesions that were an average diameter of 26 mm. Every area of enhancement that was positioned over that of the glandular parenchyma was characterized as suspiciously malignant. The authors themselves point out that their method would be difficult to put into clinical practice because of its low specificity of 37%.
When the results of mammography and US were regarded separately, MR imaging exhibited the highest sensitivity of the three diagnostic methods in our study. At MR imaging, 41 (28 invasive, 13 intraductal) of 354 malignant lesions were not correctly described. Of those 41 lesions, 25 were discovered by using mammography and/or US. Not only the best sensitivity (95.5%) but also the largest Az (0.876) was achieved when all three methods were combined, although specificity decreased to 47.1%. As far as the low specificity is concerned, one has to keep in mind that only patients with a biopsy-proved diagnosis were selected for this study. Patients in whom no biopsy was performed because of the combined results of the three imaging modalities were not included. Including these patients would have increased specificity.
Further investigation of the false-negative findings at MR imaging disclosed that in at least five of the examinations, errors were avoidable. It was obvious in one case that only a small dose of contrast agent had been administered. The reason was probably an unnoticed extravasation during injection of contrast material. Regardless of motion artifacts in another case, nonsubtracted images were not adequately evaluated, and, therefore, an 8-mm carcinoma was overlooked.
It is known that image subtraction for elimination of the fat signal is likely to cause disturbances. Even minor patient movement is enough to cause a simulation of or a hiding of tumors. The subtraction technique was used in this study because no alternative method was available for stable fat suppression with adequate magnetic field homogeneity, as well as spatial and time resolution. In 7% of the examinations, the motion artifacts were so pronounced that they could in no way be evaluated properly by using the subtraction images. Therefore, it would be desirable to have an integrated breast-fixating device in all commercial breast coils (12).
One malignant lesion was positioned outside the field of view (Fig 2). In two other misdiagnosed cases, the malignant lesions were abutting the chest wall. Because of an apparent weak enhancement, these tumors were classified as benign. It was not taken into account that, outside the coil volume, the absolute signal intensity does not increase, but that the relative signal intensity increases get smaller after contrast agent injection. Phantom measurements exhibited a reduction in signal intensity of approximately 50% over the 10 outer sections in comparison with the slab center in the dynamic fast low-angle shot 3D sequence used in this study (Beier T, unpublished observation, 1998). This error can be avoided by means of oversampling or slab enlargement. These compensatory techniques are combined with a substantial increase in measuring time, which in turn limits the use of the time course of enhancement as a diagnostic criterion. Boetes et al (13) also attributed one of six false-negative findings to the position of the tumor near the chest wall. Improved coil design might diminish this problem.
In the remaining 36 of 41 false-negative findings, a plausible explanation for the lack of tumor delineation might have been the histologic features of the tumors and/or of the surrounding tissue.
The uptake of contrast agent in three carcinomas was not markedly different from that in the surrounding glandular tissue. It is known that a generalized intensive enhancement can mask tumors. This phenomenon must be especially taken into account during the first and last week of the menstrual cycle in patients between the ages of 35 and 50 years or in ductal hyperplasia (14). Heywang (15) advised not including MR imaging in the diagnostic work-up in case of diffusely enhancing breast tissue, which was observed in 122 (23%) of 525 patients with uncertain mammographic findings. Consequently, MR imaging is not recommended in young patients or in patients with secretory diseases, inflammatory processes, proliferating changes, or microcalcifications (7,15). Unfortunately, it is usually in these same groups that patients have unclear findings at conventional imaging. In our study, in 11% of the examinations early enhancement was observed, and this finding severely limited the evaluation.
Either diffuse growth or a small tumor diameter characterized the remaining invasive carcinomas that were undetected at MR imaging. Missed additional multifocal or multicentric tumor foci, maximally 5 mm in diameter, were also not detected by using mammography or US. In our study, the section thickness was set at 1.86 mm to find small tumors. It is important to mention that this does not guarantee that carcinomas 2 mm in diameter can be reliably detected. If the tumor is smaller than twice the section thickness, then partial-volume effects may affect detection. Furthermore, very small tumors can be easily masked by motion artifacts or can be mistaken for vessel cross sections. Apart from these limitations, it is generally assumed that tumor angiogenesis does not begin before the tumor has reached a diameter of about 3 mm, and tumor angiogenesis is the prerequisite for contrast agent pooling in malignant lesions (16).
Holland and colleagues (17) found additional tumor foci in 63% of T1 and T2 carcinomas in mastectomy specimens. Such a high prevalence has not been observed in any MR imaging study to date, to the best of our knowledge. This suggests that the number of false-negative MR imaging results in breast diagnostic imaging is much higher than the data generally reported in the literature (1–8) suggest. However, it is not improbable that these small undetected tumors can be treated with radiation therapy after excision of the main malignant tumor. This would mean that these additional tumors are clinically less relevant. This does not hold true for following up unidentified invasive carcinomas.
Tomczak and colleagues (18) also failed to observe an enhancement in one of five inflammatory carcinomas, and Fischer and colleagues (19) failed to observe it in one of eight inflammatory carcinomas. In our case with a false-negative diagnosis, there was no circumscribed tumor found histologically but rather a multitude of very small tumor emboli in dilated lymphatic channels. It is a reasonable assumption that these tumors attain their nutrients through diffusion rather than through genuine tumor vessels and thus will not allow enhancement.
It is also plausible that densely packed circumscribed tumor cell formations with marked vascularization are associated with stronger enhancement compared with tumors that grow in linear cell arrangements or in small tumor nests with few vessels. The normal tissue between those tumor cell nests causes interfering partial-volume effects.
In our experience, patients who have infiltrating carcinoma with EIC and DCIS are also problematic candidates for MR imaging. Boyages et al (20) reported that approximately 30% of the invasive ductal cancers are accompanied by EIC. In their study, the recurrence rate of ductal carcinoma without EIC was approximately 8% after breast-conserving surgery and radiation therapy, but this rate increased to 27% when EIC was present. The reason for this substantial difference is thought to be caused by undetected intraductal carcinoma portions left behind in the breast (20). However, findings in some studies indicate that the prognosis for EIC tumors is not worse when the excision border is free of tumor (21). Thus, for adequate patient treatment, exact information about tumor expansion is paramount. In our experience, MR imaging cannot reliably generate this information.
To our knowledge to date, researchers in only four studies have reported depiction of DCIS at MR imaging in at least 20 cases (22–25). All of these studies were retrospective in nature and included tumors with invasive components that cannot, thus, be classified as DCIS, according to the TNM classification of that time (26). In our prospective report of findings, 13 (65%) of 20 purely intraductal carcinomas were not detected. In agreement with these observations, Fischer et al (22) found an enhancement typical for malignancy in only 15 (43%) of 35 cases of DCIS. Gilles et al (23) observed different results in their study. Of 36 cases of DCIS, only two were not correctly detected by using a dynamic spin-echo sequence with 3-mm sections and gaps of 1.5 mm. DCIS was, on the average, 45 mm and showed invasion in 33% of cases (23). Five of the 22 tumors in the study by Soderstrom et al (25) were pure DCIS (average diameter, 52 mm). All of these five cases were depicted as clumped or linear enhancements. In imaging DCIS, it is probable that the high spatial resolution of the rotating delivery of excitation off resonance technology has considerable advantages. Information about the specificity is not given in the report (25).
Sittek et al (24) retrospectively found 14 of 20 cases of DCIS in the form of focal contrast enhancement. However, the changes in signal intensity and the average increase in signal intensity were not different from those of adenosis. On the basis of histomorphologic features, one may surmise that dissociated ducts with a diameter of 1–2 mm that contain DCIS cannot be reliably diagnosed with MR imaging with our technique. The prospectively detected DCIS in our study had a larger circumscribed tumor volume and did not have a particular malignancy grade.
Further systematic studies will be necessary to determine if a field strength higher than 1 T, a single coil technique, an improved coil design, or a higher dosage of paramagnetic contrast agent would lead to better results in imaging DCIS, as well as in detecting very small or diffuse growing invasive carcinomas.
In conclusion, MR imaging is the method with the highest sensitivity in breast imaging, but it fails to depict every carcinoma of the breast. The best sensitivity and Az were achieved by using a combination of mammography, US, and MR imaging. False-negative findings at MR imaging of the breast may result from not only observer errors or technical problems but also histomorphologic characteristics of certain breast cancers. In particular, small (<4–5-mm) invasive, diffusely growing, or intraductal carcinomas can be missed at MR imaging. Therefore, one should not rely solely on a negative MR imaging finding when mammography yields even the slightest evidence of diffusely or intraductally growing tumors.
|
FOOTNOTES |
|---|
Author contributions: Guarantors of integrity of entire study, A.T., M.T.; study concepts and design, A.T.; literature research, A.T., A.H.; clinical studies, A.T., M.S., H.A.L.; experimental studies, T.B.; data acquisition, A.T., S.S., T.W.V.; data analysis/interpretation, A.T., T.B., T.W.V.; statistical analysis, T.B., T.W.V.; manuscript preparation, A.T., T.W.V.; manuscript definition of intellectual content, A.T.; manuscript editing, A.T., T.W.V.; manuscript revision/review, A.T., A.H., M.T.; manuscript final version approval, A.T., M.T.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  S. Orel Who Should Have Breast Magnetic Resonance Imaging Evaluation? J. Clin. Oncol., February 10, 2008; 26(5): 703 - 711. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. E. Noz and G. Q. Maguire Jr. Reply: 18F-FDG PET in Detecting Primary Breast Cancer J. Nucl. Med., October 1, 2007; 48(10): 1752 - 1752. [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]
|
|
|
|
|
|
L. Moy, F. Ponzo, M. E. Noz, G. Q. Maguire Jr., A. D. Murphy-Walcott, A. E. Deans, M. T. Kitazono, L. Travascio, and E. L. Kramer Improving Specificity of Breast MRI Using Prone PET and Fused MRI and PET 3D Volume Datasets J. Nucl. Med., April 1, 2007; 48(4): 528 - 537. [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]
|
|
|
|
 |
|
 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] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
