MRI evaluation of pathologically complete response and residual tumors in breast cancer after neoadjuvant chemotherapy

Authors


Abstract

BACKGROUND.

This study investigated the role of magnetic resonance imaging (MRI) in evaluation of pathologically complete response and residual tumors in patients who were receiving neoadjuvant chemotherapy (NAC) for both positive and negative HER-2 breast cancer.

METHODS.

Fifty-one individuals, comprised of 25 HER-2 positive and 26 HER-2 negative patients, were included in the study. Serial MRI studies were acquired before, during, and after NAC. On the basis of the final MRI, response was determined to be a clinically complete response ([CCR], no enhancement), probable CCR (residual enhancement equal to or less than that of glandular tissue), or residual tumor. All patients received surgery. Pathological outcomes were categorized as 1) no residual cancer, 2) no residual invasive cancer but ductal carcinoma in situ (DCIS) present, or 3) residual invasive cancer. The pathologically complete response (pCR) was defined as no invasive cancer.

RESULTS.

Complete clinical response as seen through MRI, including CCR and probable CCR, was identified in 35 (35 of 51, 69%) patients. MRI correctly diagnosed pCR in 26 (26 of 35, 74%) patients, including 18 of 19 (95%) patients in the HER-2 positive group and 8 of 16 (50%) patients in the HER-2 negative group (P < .005). The accuracy of MRI in identifying pCR varied according to the chemotherapy agent that was administered. MRI was more accurate in identifying pCR in patients who were receiving trastuzumab and less accurate in patients receiving bevacizumab. The high false-negative rate found in HER-2 negative patients was associated with residual disease that presented as scattered cells or small foci. In cases with residual bulk tumor, the lesion size, determined by MRI, correlated highly with that found in histopathological measurements (r = 0.93).

CONCLUSIONS.

MRI may predict pCR with high accuracy in HER-2 positive patients, but it has a high false-negative rate in HER-2 negative patients, particularly in patients who are receiving antiangiogenic agents. Results indicate that the chemotherapy agent should be taken into consideration when using MRI to interpret therapeutic outcomes. More studies are needed to establish the role of MRI in managing, especially surgical planning, patients who are receiving NAC. Cancer 2008. © 2007 American Cancer Society.

The traditional clinical role of neoadjuvant chemotherapy (NAC) is to down-stage inoperable cancers to render them operable or to facilitate improved outcomes in breast conservation surgery.1, 2 With evidence of improved survival for patients who achieved a pathologically complete response (pCR) after NAC,3 NAC has received attention as a treatment option for management of breast cancer. Pathologically complete response (pCR) rates have varied widely according to chemotherapy regimens; eg, 13% of patients treated with an anthracycline-based regimen (doxorubicin and cyclophosphamide),1 and 26% of patients who in addition received the second-line regimen of partially noncross-resistant agents such as taxane had a pCR.2 There was a statistically significant correlation between pathologically complete response and both overall survival and disease-free survival.4–8 Therefore, for patients who are receiving NAC, pCR is the most relevant prognostic factor.

Noting this emerging development, the National Cancer Institute (NCI) organized a conference entitled “Preoperative Therapy in Invasive Breast Cancer,” from March 26–27, 2007, to discuss critical research into the potential role of NAC in managing breast cancer. Breast imaging, especially MRI, is identified as one such research area. Although the role of MRI for preoperative surgical planning and high-risk screening is well established,9–15 MRI application in the setting of NAC remains to be investigated.

The advantage of MRI over other modalities, including mammography, ultrasound, and palpation for monitoring response after NAC has been widely reported.16–18 Mammography involves radiation, and, therefore, frequently repeating mammograms during NAC for monitoring responses is not acceptable. Previous MRI studies have focused on bulk residual disease, showing that dynamic contrast-enhanced MRI (DCE-MRI) can provide clear tumor delineation and is the most accurate imaging modality to predict residual tumor size after NAC.16–22 However, with the more aggressive NAC protocol, a higher rate of pCR (no invasive cancer) and minimum tumor burden (no bulk tumor) has been found. Therefore, for patients who are receiving an aggressive NAC protocol, the use of MRI to predict pCR and residual minimum disease should be investigated.

At our institute, we have been conducting clinical trials that use a protocol comprised of first-line anthracycline-based and second-line taxane-based chemotherapy regimens that have proven to achieve a high pCR rate relative to other published results.23 HER-2 positive cancer is aggressive and, in the past, has been considered an unfavorable prognostic biomarker. However, with the availability and effectiveness of trastuzumab (Herceptin), a targeted treatment, HER-2 has become a favorable biomarker.24 Our second-line regimen for HER-2 positive cancer also contained trastuzumab, another drug that has been proven to achieve a high pCR rate. For HER-2 negative cancer, the pCR rate is much lower. Yet, because of the promising efficacy of the antiangiogenic agent bevacizumab (Avastin) for lung and metastatic breast cancer, we added this drug to the second-line taxane regimen for HER-2 negative patients. Because bevacizumab acts on blood vessels, it may impede the delivery of MR contrast agent and, thereby, limit the ability of DCE-MRI to detect residual disease.

The objective of this study was to investigate the role of breast MRI in the evaluation of NAC outcome by correlating MR imaging findings with histopathology at post-NAC surgery. We were particularly interested in comparing the predictive accuracy of pCR with that of minimal residual disease when no residual tumor was detected on MRI. This may provide useful information for establishing the future clinical role of MRI in surgical planning after NAC. Our analysis was also categorized by HER-2 positive and negative status and by different chemotherapy regimens. Because it is known that pCR is less likely to be achieved in HER-2 negative patients compared with HER-2 positive patients, the predictive accuracy in these 2 groups was evaluated separately. Lastly, we also investigated the predictive accuracy in patients who also received bevacizumab, because the agent may possibly compromise DCE-MRI detection of residual disease.

MATERIALS AND METHODS

Subjects

Fifty-one patients enrolled from July 2003 to April 2006 were included in this study. Selection criteria allowed patients who were enrolled into the treatment trial with biopsy-proven invasive cancer, who also had received serial MRI studies before and during therapy, and who had had their last MRI performed before surgery. They had lesions either >5 cm or <5 cm and also had lymph node involvement clinically documented by palpation. Twenty-five patients were HER-2 positive with an age range of 32 to 77 years old (mean, 51 years). Twenty-six patients were HER-2 negative, with an age range of 31 to 69 years old (mean, 48 years; P = .142). The pretreatment tumor size ranged from 0.9 cm to 8.5 cm (median, 2.4 cm). The clinical stages were stage II (n = 19), stage III (n = 18), and stage IV (n = 14). Three patients had infiltrating lobular carcinoma (1 HER-2 positive and 2 HER-2 negative); the other 48 patients had invasive ductal carcinoma. The duration from last MRI to surgery was 6–90 days (mean, 38 days) in HER-2 positive patients and 20–64 days (mean, 40 days) in HER-2 negative patients (P = .65).

This study was approved by an institutional review board and complied with the Health Insurance Portability and Accountability Act (HIPAA). All patients gave written informed consent for NAC treatment, MRI, and participation in this study.

Treatment Protocol and MRI Follow-up

All patients received biweekly doxorubicin (Adriamycin) and cyclophosphamide (AC) as their first-line regimen. After the first 2 cycles, an oncologist evaluated patient response using all information available at that time (clinical examination, patient's tolerance, sonography, etc), and the oncologist decided whether the patient should continue to receive 2 additional cycles of AC or should be switched to a taxane-based regimen. The second-line taxane-based regimen comprised paclitaxel (Taxol-T) or Nab-paclitaxel (Abraxane-Ab, a new formulation of albumin-bound nanoparticle of paclitaxel), combined with carboplatin(Ca). HER-2 positive patients also received trastuzumab (Herceptin-H). Some HER-2 negative patients also received bevacizumab (Avastin-Av). Table 1 summarizes these different regimens and the number of patients in each group.

Table 1. Accuracy of MRI in Prediction of pCR in Different Drug Groups
HER-2 statusSecond-line regimenTherapeutic drugsMR-CRpCR accuracy
  1. MR-CR indicates that both a clinically complete response (CCR) and a probable CCR were considered to be complete responses as determined on magnetic resonance imaging (MRI); pCR, pathologically complete response.

  2. Statistically significant difference was found between TCaH and TCa groups (P < .05), TCaH and AbCaAv groups (P < .01), and AbCaH and AbCaAv groups (P < .025).

HER-2 positive, n = 25TCaH (n = 17)Taxol, carboplatin, herceptin1413/14 (93%)
AbCaH (n = 7)Abraxane, carboplatin, herceptin55/5 (100%)
TCaH + AbCaAv (n = 1)
HER-2 negative, n = 26TCa (n = 20)Taxol, carboplatin127/12 (58%)
AbCaAv (n = 6)Abraxane, carboplatin, avastin41/4 (25%)

All patients had pretreatment baseline breast MRI exams, at least 2 follow-up examinations during the course of therapy, and a final examination after completion of the therapy protocol. After NAC, a definitive surgery was performed. Twenty-five patients received lumpectomy, and 26 received mastectomy. The type of surgery was decided on the basis of the surgeon's recommendation and the patient's personal choice.

Pathological Examination

Surgical specimens were cut into 5 mm slices, fixed in 10% neutral-buffered formalin, and processed for histological examination. If gross tumor was evident, each paraffin block containing the tumor was sliced and stained with hematoxylin and eosin (H&E) for evaluation. If no gross tumor was found, the tissue marker left in the breast was identified, and slides from the block containing the marker as well as the adjacent blocks were examined. Residual disease post-NAC was recorded into 1 of 3 categories, 1) no residual malignancy, no sign of cancer cells; 2) no residual invasive cancer, DCIS present; 3) residual invasive cancer. Defined as no invasive cancer present, pCR included categories 1 and 2. This definition is used at M. D. Anderson (Houston, Tx)25 and also at the NCI March 26–27, 2007 conference entitled “Preoperative Therapy in Invasive Breast Cancer”. In cases with residual invasive cancer, the pathological size was determined as the longest dimension, either the longest dimension on 1 H&E-stained slide or from the number of blocks (each 5 mm) where the malignant invasive tumor was detected, whichever was greater.

MRI Acquisition

The MRI examination was performed by a 1.5 Tesla MR scanner (Philips Medical Systems, Cleveland, Oh) with a dedicated 4-channel, phased-array, breast coil (manufactured by USA Instruments, Aurora, Oh). The imaging protocol consisted of high-resolution precontrast imaging and dynamic contrast-enhanced imaging. After a scout scan, sagittal, unilateral T1-weighted, precontrast images were acquired. After this, a 3D Spoiled Gradient Recalled (SPGR) pulse sequence with 16 frames, including 4 precontrast and 12 postcontrast sets, was prescribed for axial, bilateral, dynamic imaging (TR = 8.1 ms, TE = 4.0 ms, flip angle = 20°, slice thickness = 4 mm, matrix size = 256 by 128, field of view [FOV] = 32–38 cm). The scan time was 42 seconds per acquisition. Gadodiamide (Omniscan; GE, Fairfield, Conn, doing business as General Electric Healthcare, headquartered in the UK) contrast agent was injected (1 cc/10 lbs body weight) at the beginning of the fifth acquisition followed by 10 cc saline for flushing.

Interpretation of MRI

After the study was completed, all images were transferred to a personal computer for processing and interpretation. Subtraction images were obtained by subtracting the precontrast images acquired in Frame 3 from the 1-minute postcontrast enhanced images acquired in Frame 6, and then maximum intensity projections (MIPs) were generated from these subtraction images. The subtraction images were also color coded, representing different degrees of enhancement. Examples from 4 patients demonstrating MIPs and color-coded images before, during, and after NAC are shown in Figures 1–4. On the basis of the computer-generated MIP, 1 radiologist (J.C.), with 2 years of experience in interpreting breast MRI, performed the size measurement. Two lines were drawn (the longest dimension and the longest perpendicular dimension) to obtain the 2-dimensional sizes. When the enhancement was low on MIP, the corresponding color-enhanced subtraction images were used. The shape of the tumor may change depending on how the breast is positioned within the coil. In one study, the tumor may be round, and in another study, it may be elongated. Therefore, the 2-dimensional size was measured and then converted to 1 dimension (by calculating the square root of the area), and this was used to correlate with pathological size.

Figure 1.

True-negative (CCR and pCR) of a 41-year-old woman with HER-2 positive invasive ductal carcinoma in the left breast, receiving AbCaH treatment is shown. Top left to right shows MIP pretreatment, second follow-up, and postNAC MRI studies. Bottom row shows color-coded images of MRI studies. Note the gradual shrinkage of the tumor and eventual disappearance on the final MRI (not visible on MIP). MRI diagnosis was complete clinical response (CCR). Final pathological examination did not reveal any malignancy; thus a pathologically complete response (pCR) was realized.

Figure 2.

True-negative (probable CR and pCR) of a 35-year-old woman with HER-2 positive invasive ductal carcinoma in the left breast, receiving TCaH treatment is shown. Top row shows MIP pretreatment, second follow-up, and the postNAC final MRI. Bottom row shows color-coded images in these MRI studies. Note the large heterogeneously enhanced tumor in the left breast before treatment and the subsequent marked shrinkage of tumor. There was notable enhancement of the cancer site on MIP, but the post-treatment intensity was only comparable to that of glandular tissues elsewhere. As a result, this case was designated as a probable CCR. Final pathology did not reveal any malignancy; thus pCR was achieved.

Figure 3.

False-negative (CCR and 3 foci of 1.5 mm residual cancer) of a 49-year-old woman with HER-2 negative invasive ductal carcinoma in the left breast, receiving TCa treatment is shown. Top row shows MIP pretreatment, second follow-up, and the final post-NAC MRI. Bottom row shows color-coded images these MRI studies. Note the gradual shrinkage and eventual disappearance of tumor on the final MRI (not visible on MIP). MRI diagnosis was CCR. Final pathology, however, demonstrated 3 small foci of invasive cancer, each 1.5 mm.

Figure 4.

False-positive (CCR and DCIS considered as pCR) of a 41-year-old woman with HER-2 positive invasive ductal carcinoma in the left breast, receiving TCaH plus AbCaAv treatment is shown. Top row shows MIP from pretreatment, second follow-up, and the final post-NAC MRI. Bottom row shows color-coded images of these serial MRI studies. Note that after being treated with the first-line regimen, the tumor did not appear to shrink on the first follow-up MRI but appeared to respond after switching to the second-line regimen. One 4-mm enhancing lesion was detected by MRI and was diagnosed as residual disease. However, histopathology showed DCIS, which is considered pCR by definition. Thus, this is a false-positive case by MRI.

The tumor size of each patient in all serial MRI studies (pretreatment and all follow-up) was analyzed in one sitting by the radiologist to ensure consistent determination of the tumor boundary. The radiologist was blinded to the pathology results. If the 1-dimenional size reduction after completing NAC was <30% compared with the pretreatment size, the case was classified as a nonresponder (NR). When residual tumor was present with >30% size reduction, the case was classified as a partial response (PR). Cases in which no enhanced tissues were visible were classified as clinically complete responses (CCRs). When minimal enhancement was found at the previous lesion site with weaker or comparable enhancement relative to normal glandular tissue (ipsilateral or contralateral), the case was classified as probable CCR. Both CCR and probable CCR were considered to be a complete response determined by MRI (noted as MR-CR) and MR-CR was used to evaluate the accuracy of MRI in prediction of pCR.

Statistical Analysis

A Fisher exact test was performed to compare pCR rates between groups. A 2-tailed chi-square test was used for comparison of lymph node status. A 2-tailed Student t test was used to compare difference of age and duration from the last MRI to surgery between HER2± groups. A Pearson correlation was used for comparing MRI-determined residual tumor size and pathological size. P < .05 was considered significant.

RESULTS

Surgical Types Between HER-2 Positive and HER-2 Negative Patients

In the HER-2 positive group, 13 patients received mastectomy, and 12 had lumpectomy. In the HER-2 negative group, 12 patients received mastectomy, 13 had lumpectomy, and 1 patient had initial lumpectomy followed by mastectomy. The choice of lumpectomy in these 2 groups was not statistically different (12 of 25 vs 13 of 26; P = 1).

MR-CR and pCR in Patients Receiving Different Chemotherapy Regimens

The MR prediction accuracy of pCR in different drug groups is summarized in Table 1. Overall, there was no significant difference for MRI-determined complete response (MR-CR) in different groups. The pCR prediction accuracy in HER-2 positive patients was 93% (13 of 14) in the Taxol, Carboplatin, Herceptin (TCaH) group and 100% (5 of 5) in the Abraxane, Carboplatin, Herceptin (AbCaH) group, which was higher than that of 58% (7 of 12) in the Taxol, Carboplatin (TCa) group and 25% (1 of 4) in the Abraxane, Carboplatin, Avastin (AbCaAv) group of HER-2 negative patients. It was noted that the pCR rate was the lowest in the group receiving bevacizumab (Avastin), only 25%.

MRI Tumor Response and pCR Between HER-2± Groups

The therapeutic responses evaluated by MRI (NR, PR, probable CCR, and CCR) and final pCR between HER-2 positive and negative patients are summarized in Table 2. A pathologically complete response was achieved in 28 of 51 (55%) patients. Of these, 22 patients showed no evidence of malignant cells, and 6 (4 HER-2 positive and 2 HER-2 negative) patients showed DCIS without an invasive component. The overall pCR rate was higher, 19 of 25 (76%), in the HER-2 positive group compared with 9 of 26 (35%) in the HER-2 negative group (P < .005). MRI diagnosed complete clinical response (MR-CR) in 35 of 51 (69%) patients, including 28 patients without any enhanced lesions (CCR) and 7 patients with very mild enhancements defined as probable CCR.

Table 2. Tumor Responses Determined by MRI and Pathology According to HER-2 ± Status
 NRPRProbable CCRCCRpCR
  • NR indicates nonresponder; PR, partial responder; CCR, clinically complete response; pCR, pathologically complete response.

  • *

    Significant difference between HER-2 positive and negative groups (P < .01).

HER-2 positive, n = 250 (0%)6 (24%)4 (16%)15 (60%)19 (76%)*
HER-2 negative, n = 261 (4%)9 (34%)3 (12%)13 (50%)9 (35%)*
All patients, N = 511 (2%)15 (29%)7 (14%)28 (55%)28 (55%)

Diagnostic Accuracy of MRI

The diagnostic accuracy of MRI, including true-negative (ie, accurate prediction of pCR), true-positive, false-negative, and false-positive rates are summarized in Table 3. Of the 35 MR-CR cases, MRI correctly diagnosed 74% (26 of 35) pCR. Among these, pCR prediction accuracy was much higher, 95% (18 of 19), in the HER-2 positive group compared with 50% (8 of 16) in the HER-2 negative group (P < .005). Figure 1 demonstrates 1 patient diagnosed by MRI as CCR, who had confirmed pCR. Of the 7 probable CCRs, 71% (5 of 7) were proven to have pCR, and 29% (2 of 7) had residual invasive cancer. Figure 2 shows images from 1 patient who had confirmed pCR that was diagnosed by MRI as probable CCR. MRI showed a false-negative diagnosis in 26% (9 of 35) patients, 8 in the HER-2 negative group and 1 in the HER-2 positive group. Figure 3 shows images from 1 patient who was diagnosed by MRI as CCR but who also had residual invasive cancer presenting as 3 small foci of 1.5 mm. Lastly, MRI showed a false-positive diagnosis in 2 patients, 1 HER-2 positive and 1 HER-2 negative. Figure 4 presents images from 1 patient who had a 4-mm residual tumor on MRI with DCIS on pathology. This case is considered pCR by definition, and it is, thereby, a false-positive case.

Table 3. Overall Diagnostic Performance of MRI
 True negative*True positiveFalse negative*False positive
  • *

    When true-negative and false-negative results were combined, pCR (pathologically complete response) prediction accuracy was much higher in the HER-2 positive group (18 of 19; 95%) compared with the HER-2 negative group (8 of 16; 50%; P < .005).

HER-2 positive, n = 2518 (72%)5 (20%)1 (4%)1 (4%)
HER-2 negative, n = 268 (31%)9 (34%)8 (31%)1 (4%)
All patients, N = 5126 (51%)14 (27%)9 (18%)2 (4%)

The 9 false-negative cases (1 HER-2 positive and 8 HER-2 negative) were likely due to residual disease present as scattered invasive cancer cells or foci smaller than 3 mm in size. Our MRI did not detect such residual enhancement, and it did not provide a correct diagnosis. In our study, this pattern of residual disease occurred more frequently in HER-2 negative patients.

In addition to false-negative cases, we had 2 false-positive cases. One had MRI enhancement in a region of DCIS (Fig. 4). The other false-positive MRI showed a 1-cm enhanced lesion, but pathological examination of the mastectomy specimen did not find any evidence of malignancy. As the specimen has since been discarded, the corresponding area could not be examined to verify the histopathology finding.

Correlation of Residual Tumor Size Determined on MRI Versus Pathology

Sixteen cases of residual cancer detected by MRI included 15 partial responders and 1 nonresponder, as shown in Table 2. MRI residual-lesion size was in the range of 0.4 cm to 3.7 cm, and the pathological size was 0.4 cm to 6.0 cm. A Pearson linear regression was used to correlate MRI and pathological size after we excluded 1 outlier. On MRI, a patient with lobular cancer had a 0.5-cm residual tumor, but this patient actually had scattered residual cancer cells distributed throughout a 6.0-cm region by histopathology. This case, again, demonstrates the limitation of MRI in detecting scattered cancer cells. After excluding this case, the remaining 15 cases correlating MRI and pathological size showed a Pearson correlation coefficient of r = 0.93; P < .000001.

Axillary Lymph Node Status

Axillary lymph node dissection was performed in 16 of 25 HER-2 positive patients and in 17 of 26 HER-2 negative patients. The pathological examination found positive lymph nodes in 4 of 16 HER-2 positive patients and in 9 of 17 HER-2 negative patients, a finding that was not statistically different (P = .15). Of the 4 HER-2 positive patients, 1 was pCR, and 3 did not achieve pCR. Of the 9 HER-2 negative patients, 2 were pCR, and 7 were non-pCR.

DISCUSSION

The final outcome after neoadjuvant chemotherapy was evaluated on the basis of pathological findings. Achieving pathologically complete response (pCR) is considered the ultimate goal for a favorable prognosis.3–8 Factors that are significant predictors of pathological response to neoadjuvant chemotherapy include mitotic index, histological grade, HER-2, Ki67, p53, S-phase fraction, and hormone-receptor status.26–28 High rates of pathologically complete remission were associated with absence of estrogen-receptor and progesterone-receptor expression.29 Because of the small number of cases in this study, we could not perform a multivariate analysis to evaluate all these biomarkers. By using trastuzumab (Herceptin) as a targeted therapy for HER-2, we found a statistically significantly higher pCR rate in HER-2 positive patients. Nineteen of 25 (76%) HER-2 positive patients and 9 of 26 (35%) HER-2 negative patients achieved pCR. Therefore, HER-2 positive breast cancer has become a favorable response predictor. In evaluating patients who were receiving anthracycline-based neoadjuvant chemotherapy, we found the pCR rate to be very low, ranging from 3% to 28%, in published reports.28, 30, 31 With this chemotherapy regimen, overexpression of HER-2 was not predictive of pathological response. This contradiction was likely due to the finding that trastuzumab was not yet available at that time.

The success of trastuzumab in treatment of HER-2 positive breast cancer has led to its widespread use, and it is becoming the current clinical standard. The combination of trastuzumab and chemotherapy has been found to be significantly superior to chemotherapy alone in terms of both response rate and survival.32 In addition, the combination of trastuzumab, which targets the HER-2 positive cells, and anthracycline, which targets topoisomerase alpha II coamplified cancer cells, has been shown to be synergistic.33 The NAC treatment protocol described in this study has been used at our institution since 2003. Although it achieved a high pCR rate in HER-2 positive patients, it had a much lower pCR in HER-2 negative patients.23 In late 2005, with the intention to improve the pCR, bevacizumab (Avastin-Av) was added to the treatment regimen of HER-2 negative patients. Bevacizumab is a recombinant monoclonal antibody that binds and inactivates all isoforms of vascular endothelial growth factor (VEGF) to inhibit angiogenesis as well as tumor growth and proliferation. In animal models, administration of bevacizumab blocked the growth of human tumor xenografts and reduced both the size and number of metastases.34 Bevacizumab exerts its effect through several potential mechanisms, including inhibition and regression of new tumor vessel growth, alteration of vascular function and tumor blood flow (“normalization”), and direct impact upon tumor cells.35 Although the efficacy of bevacizumab in breast cancer needs further evaluation, preliminary results from a large, randomized, clinical trial for patients with previously untreated recurrent or metastatic breast carcinoma conducted by a network of researchers led by the Eastern Cooperative Oncology Group (ECOG) has shown that those patients who received bevacizumab in combination with standard chemotherapy had a longer tumor-free period than patients who received the same chemotherapy without bevacizumab.

Although research on the accuracy of MR imaging in predicting pCR has been previously published, prediction rates related to HER-2 status have not addressed.21, 22, 36, 37 In this study, MRI correctly detected 74% (26 of 35) pCR, including 95% (18 of 19) in the HER-2 positive group and 50% (8 of 16) in the HER-2 negative group (P < .005). MRI had false-negative diagnoses in 26% (9 of 35) of patients (1 HER-2 positive and 8 HER-2 negative), and false-positive diagnoses in 2 patients (1 HER-2 positive and 1 HER-2 negative). The overall accuracy was higher for HER-2 positive patients. The large number of false-negatives in HER-2 negative patients was due to presentation of residual disease as scattered cells or small foci (<3 mm), which made MRI diagnosis very difficult, and these lesions were missed on MRI because of the lack of enhancement. This finding suggests that interpretation of residual disease on MRI may need to be more conservative for HER-2 negative patients.

We also found that the predictive accuracy of pCR by MRI in patients who received bevacizumab (Avastin) was the lowest of all treatment-regimen groups. The predictive accuracy in the bevacizumab-containing group (AbCaAv) was only 25% (1 of 4), which was significantly lower compared with 93% (13 of 14) for TCaH (P < .01) or 100% (5 of 5) for AbCaH (P < .025), lower but not statistically significant compared with 58% (7 of 12) for TCa. This was probably because of the reduced tumor vascularity associated with bevacizumab treatment and decreased enhancement of residual cancer.

Regarding the role of MRI for evaluating residual tumor size, our result was in accordance with other studies,38–41 showing agreement of MRI and histopathology (r = 0.93). In our study, 1 lobular cancer was an outlier, showing a 0.5-cm residual lesion on MRI but a 6-cm region of scattered cancer cells on pathology. This case suffered from the same limitation of MRI as other false-negative cases, reflecting the difficulty of MRI in detecting this pattern of residual cancer after NAC. If conservative surgery had been recommended on the basis of the MRI that showed 0.5-cm residual disease, the patient would have had positive margins and inadequate treatment. As more studies become available, the detection of residual cancer by MRI after NAC may be helpful in creating guidelines for surgical planning as identified as critically needed research at the March 2007 NCI conference. The current standard of care includes a definitive surgery after NAC. Our findings may suggest that 1) a complete response on MRI in HER-2 positive patients is very reliable; thus MRI can be used for planning a lumpectomy; 2) when bulk residual disease is detected, an appropriate surgery can be planned on the basis of the detected tumor size; and 3) for HER-2 negative patients with minimal residual disease shown on MRI, lumpectomy can be planned, but margins need to be carefully examined.

In addition to the response of a primary tumor, we also investigated the lymph node status in patients who received axillary lymph node dissection.42 The pathological response of both primary tumor and lymph nodes has been shown to have a significant impact on prognosis and long-term outcome in patients with locally advanced breast carcinoma.7, 42 Even if pCR of the primary tumor is achieved, lymph node metastasis indicates a substantial risk of cancer recurrence.7, 8 Positive lymph nodes were found in 4 of 16 HER-2 positive patients who received axillary lymph node dissection. Of these 4 (of 16 dissections), 1 achieved pCR in the primary tumor, and 3 did not achieve pCR. Positive nodes were found in 9 (2 pCR and 7 non-pCR) of 17 HER-2 negative patients. Therefore, HER-2 negative patients were more likely to have positive nodes compared with their HER-2 positive cohort; and non-pCR patients were more likely to have positive nodes compared with pCR patients. Because of the small number of cases, the comparison did not reach the significance level. Further studies are needed to evaluate the association of HER-2 status and primary pCR with lymph node response.

Several grading classifications are used to assess pathological responses after NAC. The most debatable issue is whether DCIS should be considered as pCR. Because of a lack of universally accepted pathological response criteria, the definition of pCR should be noted when comparing results of post-NAC published studies.3 A study at the M. D. Anderson Cancer Center (Houston, Tx) showed that, with a median follow-up of 61 months, there was no statistical difference in disease-free survival in those with DCIS only and in those with no residual invasive or in situ disease. There was also no significant difference in 5-year overall survival between these 2 groups of patients.25 This inclusion of DCIS in the pCR group was the consensus of the NCI conference, March 2007. However, although the impact of DCIS may be negligible with regard to prognosis, it may be an important risk factor for local recurrence after breast conservation therapy.43 With the cases in this study, we will be able to separately follow patients with and without DCIS in the future.

Several limitations existed in our study. Only patients receiving the chemotherapy regimens of our particular clinical trial that including anthracycline and taxane-based regimens were studied. Because different chemotherapy regimens may affect tumor detection by MRI, results may not be generalizable. The spatial resolution of the MRI images used in this study was suboptimal, which limited the ability to detect small tumor foci. The relatively low spatial resolution was selected in 2002 to cover the whole breast and also to have a temporal resolution of 42 seconds, which was critical in the analysis of our enhancement kinetics (not presented in this study). Even with an MRI system upgrade, we chose to alter the protocol so that results would be comparable. We have moved the study to a 3.0-Tesla scanner, and our current protocol has improved spatial resolution as well as fat suppression. Postcontrast enhanced images can now be directly evaluated without subtraction, and they are, thus, less prone to motion problems. One major difficulty in breast MRI is to spatially correlate MRI findings with pathological findings. Although MRI-guided biopsy has become available, it is much more complicated compared with mammography or ultrasound-guided biopsy; thus it is not widely used. Furthermore, a biopsy after NAC before definitive surgery may not be clinically justified. In this study, we took advantage of serial MRI studies to follow gradual tumor response. It would be very difficult to evaluate the final MRI by referencing only the pretreatment study without the intermediate follow-up studies.

Conclusion

In this study, we investigated the role of MRI in evaluating breast cancer response after NAC. Complete response determined by MRI was highly correlated to pCR in HER-2 positive patients, but complete response had a high false-negative rate in HER-2 negative patients. The major difficulty was detecting minimal residual disease that presented as scattered cells or small foci, particularly in patients who were receiving antiangiogenic agents. Conversely, in cases with bulk disease, the tumor size determined on MRI was highly correlated with pathological size. Our results suggest that detection of minimal residual disease is difficult, especially in HER-2 negative patients. More studies and long-term treatment outcomes are needed to establish the role of MRI in the management of patients who are receiving NAC.

Acknowledgements

The authors thank Mrs. Becky Semon, RT, for MRI acquisition and Mr. Long Vu for his assistance with statistical analyses.

Ancillary