MRI HAS BEEN PREVIOUSLY SHOWN to be more accurate than other imaging techniques for determining the size and extent of most breast malignancies (ductal and lobular) when assessment is prior to therapeutic intervention (1–4). Mumtaz et al (1) reported a correlation of 0.93 for MRI vs. 0.59 for x-ray mammography when compared to the histology of the lesions. Rodenko et al (2) found good histologic correlation vs. MRI in 85% of their 20 patients with lobular carcinoma compared to only 32% for mammography. A significant number of previously unsuspected malignant lesions are also commonly reported following MRI evaluation according to these authors. In these four studies MRI-visible cancer foci, not detected by x-ray, ranged from 17% to 34%. Esserman et al (5) used a wider range for lesion extent concordance (within 50% of histologic size) in evaluating their 57 patients. MRI had 98% concordance vs. 55% for mammography. They also retrospectively assessed the MRI data and determined that an appropriate change in surgical management would likely have occurred in 42% of these women if MRI had been the primary preoperative imaging tool. In the Conrad et al (6) study of 40 women, preoperative MRI was used to change the treatment decisions; the therapy was altered in 19 (51%) patients, with only one of the changes being of questionable benefit. The additional critical data that MRI provided in these and other studies related to an improvement in the accuracy of location and size determination of the known cancer, or identification of other suspicious lesions. This is particularly important in patients who are being considered for breast conservation therapy. Determining the full extent of a malignancy always has high preoperative value, but breast-sparing lumpectomy operations necessitate the greatest degree of staging accuracy. The normal breast anatomy does not lend itself to easy determination of the boundaries of most superimposed disease. The x-ray appearance of tissue may also be difficult to interpret at a site of prior biopsy due to scarring (Fig. 1). This type of problem is frequently exacerbated by fibrous reaction in the region of the tumor after chemotherapy has been administered, making assessment via palpation and standard imaging difficult (7). Thus, a residual mass that is palpable or x-ray visible may be entirely “sterile” or have significant viable tumor (7). MRI has been shown to define the extent of disease well under many pre- and postintervention conditions, even when tumor extends within ducts significantly beyond the palpable or x-ray visible mass (5, 8, 9). Intra- or postoperative detection of residual disease beyond the initial surgical margin has significant negative consequences relating to patient health and psychology, and health care economics. In addition to improving patient care, accurate preoperative staging is also reported to produce significant cost savings, according to previous investigators (5, 10). Preoperative MRI exams that follow induction chemotherapy will be of great value if they are shown to be as accurate as MRI studies of untreated tumors. However, there have been relatively few studies performed to assess MRI accuracy after chemotherapy has been administered, and they have reported mixed results (9, 11, 12). Gilles et al (11) found that MRI successfully identified 22 of 23 tumors that had viable residua. Abraham et al (9) demonstrated similar results with MRI, missing only one of 30 postchemotherapy tumor deposits. However, in a recent study of 13 patients, Rieber et al (12) raised serious doubts about the value of MRI. MRI failed to identify cancer residua in four patients and significantly underestimated the extent of disease in two others. These investigators included multiple observations concerning the capability of MRI in this clinical area; however, they did not specifically compare MRI to the histologic measurements. This study was undertaken to determine the ability of MRI to identify and measure viable tumor residua following chemotherapy, and to compare MRI results with those derived from x-ray mammography, ultrasound (US), and palpation.
Twenty consecutive patients with breast cancer were evaluated following chemotherapy using MRI to assess the size of cancer residua and compare these data with subsequent histologic measurements of the viable tumor. This retrospective study also involved assessment of the preoperative size of the malignancy as determined by physical exam and x-ray mammogram. These values were later compared with the histology. The tumor size correlation coefficient between MRI and pathologic analysis was the highest, at 0.93. Physical exam and x-ray mammography (available for 17 patients) produced correlation coefficients of 0.72 and 0.63, respectively, compared to histologic measurement. The accuracy of MRI did not vary with the size of cancer residua. MRI is an accurate method for preoperative assessment of breast cancer residua following chemotherapy. J. Magn. Reson. Imaging 2001;13:868–875. © 2001 Wiley-Liss, Inc.
MATERIALS AND METHODS
The study was limited in nature, attempting only to compare the ability of several imaging techniques to accurately define the extent of a breast malignancy, the location of which was already known. Accordingly, data was obtained retrospectively on 20 postchemotherapy breast cancer patients, 32–66 years old (median = 45). The preoperative clinical and imaging reports were reviewed to determine the estimates of residual tumor via each modality. These data were compared to the histologic measurements of viable tumor contained in the pathologist's report.
The 20 patients included in the study had all received a full course of chemotherapy and an MRI exam prior to surgery during the 30-month period ending in October 1999. At the time of initial staging, most of our patients had large tumors: T4 (N = 6), T3 (N = 12), and T2 (N = 2), most of which were invasive ductal carcinoma (N = 18) (vs. lobular carcinoma, N = 2). Diagnostic specimens were obtained via core-needle biopsy or excisional biopsy. The chemotherapy regimen included Adriamycin and Cytotoxan in all patients for three to six cycles. Additional treatment included 5-Flurouracil (N = 6), Taxol (N = 4), and radiation therapy (N = 4). All patients had a mastectomy (N = 13) or wide local excision (N = 7) at the site of the original tumor even when no residual tumor mass was readily visible. The specimen was deemed adequate for this study if no “positive margins” were encountered, although this occasionally necessitated reexcision of positive margins in patients with partial mastectomy. The specimens were processed in the standard manner at each institution and thus were not examined using a specific protocol related to any of the imaging studies. Although the high percentage of mastectomies in our patient group allowed evaluation of other portions of the breast, the relatively thick 7–10-mm sectioning for visual inspection and palpation would not likely detect many small synchronous lesions. As a result, our study was confined to assessment of the known lesion.
Unilateral breast MRI was performed on Philips (Best, The Netherlands) 1.5T scanners (ACSII and NT platforms) using a bilateral breast coil. A series of 6-mm T1 and T2 images were initially acquired for purposes of localization and signal characterization. Only subsequent postcontrast (using intravenous gadolinium DTPA) sequences were used for tumor measurement. Low-resolution 3D axial gradient-echo (TR/TE/flip = 7.4 msec/3.6 msec/55°), T1 pre- and postcontrast images were obtained 45–60 sec following injection using one acquisition, 4-mm slice thickness, 128 × 256 matrix, and 30–38-cm FOV. The series of subtraction images from this data set were used to assess for rapid enhancement within the lesion. Most lesion measurements were obtained from the second (slightly delayed) higher-resolution gradient-echo 3D fat-saturated (spectral inversion recovery (SPIR) technique) sagittal T1 (TR/TE/flip = 35 msec/7 msec/70°) sequence. This second image series immediately followed the rapid scan using a low-to-high k-space acquisition. Spatial resolution ranged from 18–20-cm FOV with a 166 × 256 matrix and 1.8-mm sagittal partitioning. This in turn was compared with an identical precontrast scan. The measurements derived from these MR images were prospective, independent, and blind to other imaging exams. The exams were performed at an MR-only imaging facility, and other imaging studies and reports were rarely available at the time of initial interpretation. The largest single, and mean residual tumor dimensions were measured via contrasted MR (20/20 patients), physical examination (20/20 patients), mammography (17/20 patients), and sonography (8/20 patients). The mean number of days between preoperative MRI and surgery was 18.9 (standard deviation 18.8 days) with a wide range of 0–73 days. However, this included only one patient whose MRI exam and definitive surgery was separated by more than 48 days. Preoperative x-ray mammography was performed using standard craniocaudal and mediolateral views and any additional compression or oblique views needed for a diagnostic exam. The preoperative interval ranged from 1–63 days (mean = 21.8 days). The most recent preoperative physical exam data were derived from clinic notes. The presurgical interval ranged from 0–69 days (mean = 14.4 days). The eight patients who had preoperative high-resolution US studies were examined from 1 to 63 days (mean = 26.4 days) prior to surgery. There was significant variation concerning the time of assessment following chemotherapy. The mean time between the end of chemotherapy and tumor size assessment was 24, 14, 20, and 22 days for MRI, x-ray, US, and palpation, respectively. Although each patient received at least three cycles of therapy before posttherapy imaging was performed, two patients had their most recent preoperative imaging evaluation prior (3 and 8 days) to finishing all chemotherapy cycles.
Statistical regression analysis was used to correlate the results with equivalent measurements derived from histopathologic examination of the postsurgical specimens. Histologic measurement data were taken from standard written reports. There was no effort to correlate the specimen orientation with any of the preoperative images prior to sectioning and analysis. Histologic measurements reported in this article were derived from the mean tumor dimensions rather than the single largest dimension, unless otherwise noted.
Mean tumor dimensions measurements demonstrate better correlation than largest single dimension (LSD) measurements (Table 1). This is probably in part due to a reduction of sampling error when obtaining a mean. Furthermore, the known errors that will exist between imaging and the “gold standard” of histologic assessment will be reduced. This relates to the fact that the LSD measured by the different imaging modalities is often a directly visualized oblique plane, whereas that determined by the pathologist is often from a different plane of section, or is determined via a more indirect method (amalgamation of several separate specimen fragments). Also, when the largest dimension of the tumor is perpendicular to the plane of section of the specimen, the LSD must be estimated by analyzing contiguous slices. A direct correlation is often not possible, such as when an intraoperative positive margin necessitates a second excision. Averaging three orthogonal measurements would theoretically reduce this error, and is supported by our data (Table 2).
|P <.0001||P <.001||P <.05||P >.05|
|P <.0001||P <.0005||P <.01||P >.05|
|P <.005||P <.005||P <.005||P >.25|
|P <.001||P <.005||P <.0005||P <.05|
|P >.25||P <.05||P >.25||P >.1|
|Mean (CM)||Largest (CM)||N =|
|Path||3.1 ± 2.2||4.6 ± 4.1||20|
|MR||3.2 ± 2.2||3.8 ± 2.7||20|
|Palp||4.3 ± 4.2||4.8 ± 4.7||20|
|Mammo||2.1 ± 2.2||2.3 ± 2.4||17|
|Sono||2.2 ± 2.4||2.4 ± 2.9||8|
The majority of our data is displayed in the following tables and figures. The correlation coefficients between preoperative imaging measurements of lesion size and those of the pathologist are contained in Table 1. The mean diameter and largest single-diameter measurements are in the bottom left and upper right sides of the table, respectively. This data is also listed in Table 2, with standard deviations included. The correlation scattergram of MR vs. pathology measurement is Chart 1. With a P-value of < 0.0001, our data confidently demonstrates a very high correlation (r = 0.936) between MR estimates of residual tumor and pathologic specimen measurements of the viable malignancy. There was no difference noted in the ability of MRI to accurately measure large vs. small cancer residua.
Estimates of postchemotherapy lesion size by physical exam were also obtained in each patient prior to surgery. The correlation coefficient compared to the mean histologic diameter was moderate at 0.724 (P <.005). However, the wide standard deviation equaled the mean lesion size, clearly limiting confidence in this method of assessment. X-ray mammography films were obtained in only 17 of the 20 patients. This data still produced a confidence P-value of < 0.001. The marginal correlation of 0.627 compared to histologic analysis demonstrates the limitation of the x-ray technique under these circumstances. These data are visually displayed in the corresponding scattergrams (Charts 2 and 3).
The number of US examinations (N = 8) on this patient group was too small for unequivocal conclusions to be drawn.
The retrospective nature of this study, and the lack of careful correlation between imaging studies and pathologic sectioning, significantly limits our conclusions. Thus our assessment was confined to answering the question of whether MRI was superior to other current techniques for preoperative determination of the size of a known breast malignancy. Based on our results, we concluded that contrasted MRI is well suited for evaluation of the postchemotherapeutic breast lesions, especially when considering extent of residual/recurrent disease prior to the planning of breast conservation surgery. The specificity limitations of MRI are partially obviated under the circumstances encountered following treatment of a malignancy whose general location is already known. However, there were significant problems encountered in this study concerning correlation of histologic and image derived measurements. They were partially solved by the use of the mean value of the three maximal dimensions, as discussed in the Materials and Methods section. No overtly subjective data were included in this analysis; only numeric values documented in preoperative reports were used, which tended to neutralize imaging bias. However, many deficiencies still exist.
The time delay between imaging studies and definitive surgery unquestionably results in reduced correlation with pathologic specimen. The average delay and range was similar for the various modalities, and thus little bias is likely to have been introduced by the timing of the studies.
MRI accuracy is overstated since the location of malignancies was known via prior biopsy and pretherapy MRI examination in 13 patients. Essentially, all focally enhancing tissue was assumed to be viable cancer if it was in the area previously known to be involved by tumor. This is a key issue concerning the accuracy of our MRI data, under less controlled circumstances, this approach to interpretation will almost certainly not work as well. However, MRI postchemotherapy assessment was performed without benefit of other imaging data, and thus no other known bias was introduced.
X-ray measurement did not include any effort to estimate viable tumor extent vs. “scar.” Even a mammogram of a relatively “clear” fatty breast that easily demonstrates abnormal density does not readily determine the proportion of the abnormality that is malignant, vs. fibrosis or nonviable tumor residua (Figs. 2 and 3). Glandular tissue blended with and masked the actual tumor size on a number of x-ray studies (Fig. 4).
Palpation estimates of tumor size are subjective, depending on the size and consistency of the breasts, location of the tumor, and experience of the examiner. This measurement is, at best, an approximation and is difficult to precisely quantify. It cannot identify the proportion of viable tumor vs. scar or other tissue alteration in the area of the malignancy (7). In this study, palpation was not performed blindly. The examiners had the benefit of contemporaneous mammograms or US examinations or associated reports. There is a high likelihood that the palpation measurement accuracy is overestimated in this study. In several instances documented in the clinic notes, the estimated tumor size was markedly changed on clinic visits immediately following receipt of x-ray mammography reports that differed from previous physical exam assessment. New palpation estimates were always closer to that reported mammographically. Only the last preoperative palpation estimate was used for our data set for the purpose of consistency and because there was no means to determine whether any examiner's opinion was actually changed by an x-ray report. However, this apparent bias likely accounts for the correlation coefficients of these two assessment parameters (x-ray and palpation) being closer to each other than to the histologic findings. The palpation method still overestimated the tumor size by roughly one third, despite x-ray mammography underestimating by a similar proportion.
Pathology measurements were used as the “gold standard,” but they also revealed very significant inaccuracies in this study. Maximal dimension varies depending on histologic section plane, yet no correlation was made with specimen orientation/location vs. preoperative imaging studies. Most histologic measurements were 2D, frequently not including the craniocaudal direction. The common pattern of scattered viable tumor in a zone of normal and necrotic tissue is a difficult scenario in which to compare measurements. Our use of the mean of the three orthogonal plane measurements from the preoperative studies and histologic specimens (when available) proved to be a better method of comparison (vs. longest dimension). This partially solved the problem created by not orienting the specimen similar to that of the imaging studies prior to sectioning.
The correlation between palpation and histologic assessment was fair at best. Small size was not the only causative problem, as three of seven residual masses less than 2 cm were accurately measured on physical exam. The other four were not detected. A few lesion size estimates were markedly in error, with > 5 cm discrepancies noted. Both grossly larger and smaller palpation estimation errors occurred.
X-ray mammography/histologic comparison showed moderate statistical correlation, but with several significant problem areas. Six of the 17 lesions identified at pathology were not visualized on the mammograms. Of these, three were less than 1 cm, but the other three were > 3 cm, demonstrating a “dangerous” x-ray deficiency when the malignancy is not calcified. Mammography underestimated lesion size by > 1.5 cm in five of 17 patients, but overestimated by > 1.5 cm in only one patient. The overestimates related either to inclusion of nearby benign calcifications or inclusion of adjacent dense tissue (parenchyma or scar) with the viable tumor (Fig. 2).
Although statistically insignificant due to small sample size (N = 8), the US estimates of tumor size correlated relatively poorly with pathology measurements, on average underestimating size by over 30%.
Despite significant known measurement error, the correlation between MRI and histologic assessment is very strong, and might be even more precise if sectioning of the specimen was in the same orientation in which the imaging had been performed. The extremely high confidence level (P <.0001) concerning MR estimates of residual tumor and pathologic specimen measurements contained in our data set made it unnecessary to evaluate additional patients for this portion of the study.
None of our patients had complete resolution of the neoplasm; histologic exams always revealed at least some scattered residua, as small as 1 mm. The MRI technique detected individual foci of enhancement of a similar size in these patients, although a one-to-one correlation was clearly not established in this study. This differs from the findings of Rieber et al (12), who reported four false negative results in only 13 patients. One of the likely reasons for this was their use of relatively low-resolution images (4-mm-thick partitioning with 1.4 × 2.2-mm pixels). Our study used a maximal thickness of 2 mm and pixel size of 1.2 × 0.8 mm. Gilles et al (11) reported only one false negative MRI in their 18 patients with invasive carcinoma; the missed lesion was only 2 mm in diameter. This again points toward the importance of high spatial resolution examination, as they used 1.25 × 0.625 mm in-plane resolution (the sensitivity of the thicker 3 mm/1.5 mm gap slices was improved by the use of subtraction images). Proving that any small foci of MRI enhancement unequivocally corresponds to a specific small macroscopic or microscopic histologic finding remains a serious problem. It was not addressed directly in our study beyond the demonstration of strong correlation of MRI and histologic measurements in all size categories. It remains possible that our criteria of “any strong enhancement foci in the region of prior tumor” may have produced an occasional situation in which both false positive and false negative MRI interpretations occurred in the same patient, but were not appreciated. Since no rigorous documentation of location of each lesion was possible in this study, this could occur when small benign enhancement foci were adjacent to minimally or nonenhancing malignant cells. Although this is unlikely to occur frequently, it is prudent to limit the conclusions of this study to the comparison with other techniques. MRI was clearly superior to physical examination and other imaging techniques for evaluation of postchemotherapy breast malignancy.
MRI is less limited by the presence of dense (or normal) glandular tissue, especially following chemotherapy. This is particularly true when small, scattered tumor residua are present. It is our impression that background breast glandular signal is reduced following chemotherapy, possibly secondary to a direct suppressive effect of the chemotherapeutic agents on physiologic breast activity. This potentially has an effect of making areas of tumor enhancement “stand out” more than usual, especially if they fail to respond to the therapy. The accuracy of MRI may actually improve following chemotherapy if a lower level of enhancement is accepted as abnormal. Rieber et al (12) reported a significant decrease in the degree of enhancement in most viable malignancies following chemotherapy. These authors found that in four of their 13 patients, MRI failed to detect the residual disease. However, only tumors that responded to therapy were missed at follow-up exam, and there was a three- to fourfold decrease in volume in this group. If a pattern of scattered small/tiny foci of viable residua was encountered, a lower-resolution technique such as was used in their study would likely fail to detect measurable enhancement due to a volume-averaging effect. In their study a 4-mm slice thickness was used in conjunction with a 1.4 mm × 2.1 mm in-plane resolution. We encountered this type of pattern (foci of residua smaller than 5 mm) in 20% of our patients. Our apparent success in detecting small tumor residua likely related to two major factors. Our decision to use a high-sensitivity, low-specificity diagnostic criteria evidently combined well with the higher-resolution technique utilized to produce the high degree of correlation with histologic analysis. Other investigators have also had success correctly identifying residual viable breast malignancy by using either a high-sensitivity (11) or high-resolution (9) technique. Thus, initially one is led to conclude that an appropriate MRI evaluation for this clinical problem would include this combined approach. Therefore we recommend using a high-resolution fat-saturated T1-weighted postcontrast technique. Currently, we suggest interpreting nearly any nonvascular focal enhancement within the area of a known carcinoma that has been recently treated as suspicious for viable residua. Under other conditions, this diagnostic combination clearly may not work as well.
In this study MRI was shown to provide the most accurate delineation of the extent of residua in a known breast malignancy following chemotherapy. We can conclude that MRI examination following chemotherapy will likely allow more precise breast-conservation surgery than with the assistance of x-ray mammography (and palpation) alone. Its use should be strongly considered in any scenario in which knowledge of the size and extent of the lesion will have significant operative or postoperative consequences. Use of MRI in this circumstance is likely to provide considerable improvement in patient care and also likely to result in improved cost efficiency. We believe that contrasted MRI can become the modality of choice for evaluating breast cancer in the preoperative, postchemotherapeutic patient group.
We are grateful for the help provided by Tami Walter and Sandra Davis (manuscript preparation), Alison Russell (image digitization), Virginia Barnett (illustration preparation), and Jerri Payne, PA (patient care).