Introduction

Neoadjuvant chemotherapy (NAC) has become standard treatment in locally advanced breast cancer, and is now used for triple-negative and human epidermal growth factor receptor 2 (HER2)–positive breast cancer even in early-stage cancer. As previous trials have shown that patients with HER2-positive or triple-negative breast cancer and residual disease after NAC derive a survival benefit from switching to specific adjuvant treatments [1], the prognostic role of accurately assessing pathologic complete response (pCR) has also become increasingly important. Dynamic contrast-enhanced (DCE)–MRI has been widely used as an imaging modality for evaluating tumor response in breast cancer after NAC, and has been reported to show superior performance to mammography and ultrasonography in depicting residual disease [2, 3]. However, early prediction of response has been reported to be less accurate than post-treatment prediction [4]. Predicting response earlier, preferably even before the initiation of treatment, expands clinical applications by allowing personalized treatment modification and avoidance of ineffective treatment.

Despite growing interest in expanding the role of MRI in both screening and diagnostic imaging, access to MRI is limited mostly due to cost and scan time issues. Therefore, attempts to shorten MRI to make it more accessible have recently gained momentum, including ultrafast MRI [5]. Breast ultrafast DCE-MRI has a very high temporal resolution, ranging from approximately 4 to 7 s, and provides kinetic information at very early time points [6, 7]. Several previous studies have reported that such early kinetic information demonstrates comparable accuracy in discriminating benign breast lesions from malignant lesions compared to the kinetic curve analysis derived from conventional DCE-MRI, implying that the very early phase from ultrafast DCE-MRI may substitute for the conventional DCE-MRI delayed phase [7, 8]. In addition, a recent study has also reported that ultrafast DCE-MRI kinetic parameters were associated with tumor aggressiveness [8]. However, to the best of our knowledge, no studies have been published regarding the association between early kinetic parameters derived from pretreatment ultrafast DCE-MRI and pCR after NAC in patients with invasive breast cancer. As ultrafast MRI can be combined with conventional DCE-MRI, combining kinetic information from both could improve response prediction. In addition, as sensitivity to NAC differs according to cancer subtype with pCR rates ranging from 0.3 to 50.3% [9, 10], kinetic parameters associated with treatment response may also differ according to tumor subtype.

Thus, the purpose of our study was to investigate whether pretreatment kinetic features from ultrafast DCE-MRI are associated with pCR in patients with invasive breast cancer and according to immunohistochemistry (IHC) subtype.

Materials and methods

Study population

This retrospective study was approved by the Institutional Review Board, and the requirement for informed consent was waived. We retrospectively reviewed breast MRI studies obtained in our institution between August 2018 and June 2019. From the breast MRI database, we identified 1709 consecutive women with breast cancer who underwent preoperative breast MRI. During the study period, all women who underwent preoperative MRI for newly diagnosed breast cancer were imaged with a combined protocol including ultrafast and conventional DCE-MRI. Among them, 266 women had received NAC. Among these patients, we excluded those who did not undergo surgery at our institution (n = 3) and those with inadequate computer-aided diagnosis (CAD)–generated images for analysis (n = 7). Inadequate CAD-generated images included presence of motion artifacts (n = 5) or tumors located at the far peripheral breast (n = 2) for which automatic segmentation and angiovolume calculation by the CAD system was not possible. Finally, 256 women (mean age, 50.2 years; age range, 25–86 years) with 256 invasive breast cancers who underwent NAC and surgery were included in this study.

MRI protocol

MRI examinations were performed using a 3-T MRI scanner (Ingenia or Ingenia CX, Philips Healthcare) with a dedicated 16-channel bilateral breast coil with the patient in the prone position. The dynamic series were preceded by sequences including a three-plane localizing sequence, axial bilateral modified Dixon turbo spin echo T2-weighted sequences, and an axial diffusion-weighted sequence. The ultrafast and conventional DCE-MRI protocol is shown in Fig. 1. For the dynamic series, we first obtained conventional, precontrast images, and then obtained pre- and post-contrast images with the ultrafast protocol before performing post-contrast conventional DCE-MRI. The ultrafast sequence obtained a total of 10 phase images with a 6.5-s temporal resolution, with 9 phases acquired immediately after and during contrast injection. Gadobutrol (Gadovist, Bayer Healthcare) was intravenously injected at a dose of 0.1 mmol/kg of body weight and at a rate of 2 mL/s followed by a 20-mL saline flush. Subsequently, conventional post-contrast image acquisition was performed immediately after completion of the ultrafast sequence, which consisted of five post-contrast axial phases. An enhanced T1-weighted high-resolution isotropic volume examination (eTHRIVE) sequence with a 63.7-s temporal resolution was used for conventional DCE-MRI. Therefore, the first and fifth conventional post-contrast image was obtained at approximately 58.5 s and 313.3 s after contrast injection, respectively. Details regarding acquisition parameters for axial ultrafast and conventional DCE-MRI are presented in Table 1. After the completion of axial post-contrast sequences, we obtained additional images with a T1-weighted delayed post-contrast sequence in the sagittal plane, including the full range of the axilla (TR/TE, 4.1/2.1 ms; FOV, 250 × 250 mm; matrix, 320 × 320 mm; thickness, 3 mm).

Fig. 1
figure 1

DCE-MRI protocols for ultrafast and conventional axial MRI. DCE, dynamic contrast-enhanced; DWI, diffusion-weighted imaging; s, seconds; T1W, T1-weighted

Full size image
Table 1 Acquisition parameters for ultrafast and conventional axial DCE-MRI
Full size table

Conventional MRI analysis

One dedicated breast radiologist (V.Y.P, with 7 years of subspeciality experience in breast imaging) reviewed conventional DCE-MRI according to the American College of Radiology Breast Imaging-Reporting and Data System [11], and categorized lesions into masses and nonmass enhancement (with or without a mass). Multiplicity was defined as findings highly suspicious for additional sites of malignancy within the ipsilateral breast, based on tissue confirmation or imaging. In women with multifocal or multicentric disease, only the largest biopsy-proven tumors were included for analysis. Tumor size, which was measured as the maximum diameter of the enhancing tumor, was recorded from the radiological report.

All conventional DCE-MRI data were transferred to a commercially available CAD system (CADstream, version 6.0; Merge Healthcare, Inc.). Tumor enhancement kinetics were retrospectively assessed by using the CAD system. When comparing the precontrast and first contrast-enhanced series, a 50% enhancement threshold was selected. The system automatically segmented the tumors in 3D and calculated the angiovolume (total enhancing lesion volume), peak enhancement value (the highest pixel signal intensity in the first contrast-enhanced series), and delayed enhancement profiles (proportions of washout, plateau, and persistent enhancement components). A color-coded map was generated based on changes in pixel values between initial and delayed contrast-enhanced series. The washout type consisted of a decreased pixel signal intensity at the last contrast-enhanced series of more than 10% from the first contrast-enhanced series. The persistent type consisted of an increased pixel signal intensity at the last contrast-enhanced series of more than 10% from the first contrast-enhanced series. The plateau type consisted of a change in either direction less than 10% from the first contrast-enhanced series. On the basis of delayed enhancement profiles, we recorded the percentage of washout enhancing components within each tumor.

Ultrafast DCE-MRI analysis

Analyses of ultrafast DCE-MRI parameters were performed on a workstation (Aquarius iNtuition Viewer; TeraRecon). Using 10 phases of ultrafast-DCE MR images, 9 subtracted images were obtained by subtracting the first (precontrast) phase form each of the other 9 phases. Sequential maximum-intensity projection (MIP) images were generated for each time point. Time-to-enhancement (TTE) was defined as “the time point where the lesion starts to enhance minus the time point where the aorta starts to enhance” [8]. Two readers (V.Y.P and J.H.K, with 7 and 6 years of experience, respectively) evaluated TTE in consensus based on the MIP images. When it was difficult to visually decide the time point based on MIP images, we determined it as the point where the washin curve starts to rise with the aid of the software [12]. The maximal slope (MS) was determined as the slope of the tangent (percentage relative enhancement/second [%/s]) along the steepest part of the curve [8], which was automatically calculated by the software. One reader (V.Y.P) placed a circular region of interest (ROI) of 3 × 3 mm for the strongest and most rapidly enhancing part of the tumor (henceforth, referred to as MS_3mm) and also manually drew a ROI over the enhancing portion of the representative slice of the tumor on the post-contrast ultrafast MRI (henceforth, referred to as MS).

In addition, we measured the volume of each tumor on ultrafast DCE-MRI. U1 was defined as the time point at which the lesion starts to enhance and U2 was defined as the subsequent time point after the lesion starts to enhance. After visually reviewing the cases, we found that the tumor volumes generally did not significantly differ between U2 and later ultrafast DCE-MRI phases. Therefore, we chose to obtain volume data from time points U1 and U2 (Figs. 2 and 3). One breast radiologist (V.Y.P) manually segmented the tumors at U1 and U2 on subtraction images using the software (Aquarius iNtuition Viewer). The enhancing portion of the tumor was manually outlined on each slice containing the tumor, and then the tumor volume was automatically calculated by the software. To evaluate interobserver reproducibility, another breast radiologist (J.H.K) independently performed tumor segmentation on 40 randomly chosen lesions. In addition, we calculated the ratios of tumor volume from each time point: volume ratio U1/U2, the ratio of tumor volume at U1 to tumor volume at U2; volume ratio U1/CAD, the ratio of tumor volume at U1 to angiovolume calculated by the CAD system (CADstream); and U2/CAD, the ratio of tumor volume at U2 to angiovolume calculated by the CAD system.

Fig. 2
figure 2

Images in a 52-year-old woman with triple-negative invasive ductal carcinoma who achieved pCR after NAC. a Pretreatment axial conventional early contrast-enhanced MRI scan shows the enhancing breast cancer in the left upper inner breast. b Maximal slope was calculated by placing a circular ROI of 3 × 3 mm at the most rapidly enhancing part of the tumor, and also by manually placing a ROI over the enhancing portion of the tumor on ultrafast DCE-MRI. Tumor volume was measured on ultrafast DCE-MRI at time point U1 (c) and U2 (d), resulting in a U1/U2 volume ratio of 0.82. e Post-treatment axial conventional early contrast-enhanced MRI scan shows no residual enhancement around the clip marker, which had been previously inserted within the tumor (arrow)

Full size image
Fig. 3
figure 3

Images in a 54-year-old woman with HER2-positive invasive ductal carcinoma. a Pretreatment axial conventional early contrast-enhanced MRI scan shows heterogeneous nonmass enhancement in the right outer central breast. b Maximal slope was calculated by placing a circular ROI of 3 × 3 mm at the most rapidly enhancing part of the tumor, and also by manually placing a ROI over the enhancing portion of the tumor on ultrafast DCE-MRI. Tumor volume was measured on ultrafast DCE-MRI at time point U1 (c) and U2 (d), resulting in a U1/U2 volume ratio of 0.36. e Post-treatment axial conventional early contrast-enhanced MRI scan shows remaining nonmass enhancement. Surgery revealed residual invasive ductal carcinoma and in situ

Full size image

Clinicopathologic data collection

Clinicopathologic data (patient age, preoperative clinical lymph node status, treatment, and histopathological information) were collected from medical records. Clinically node positive was defined as positive signs of axillary lymph node metastases on either preoperative axillary ultrasound or cyto-/histopathology [13]. The histologic type and IHC subgroup were assessed based on pretreatment core biopsy specimens. The positivity for each of the biomarkers was defined according to the American Society of Clinical Oncology (ASCO)/College of American Pathologists (CAP) guidelines and is summarized in Supplemental Table 1 [14, 15]. The tumors were stratified according to IHC receptor status into HR-positive (ER- and/or PR-positive and HER2-negative), HER2-positive, and triple-negative subgroups. The final histopathological results of surgical specimens were reviewed to determine pathological T and N categories. A pCR was defined as the absence of residual invasive tumor cells in the breast and axilla.

Statistical analysis

The clinicopathologic factors, ultrafast and conventional DCE-MRI features obtained from pretreatment MRI data, were compared on the basis of pCR. Continuous variables were compared using the Mann–Whitney U test or Student t test. Categorical variables were compared by using the \({\chi }^{2}\) test or the Fisher exact test. Two-sided p < 0.05 was considered to indicate a statistically significant difference. For IHC subtypes, post hoc pairwise analysis was performed with a Bonferroni method for multiple comparison adjustment; p < 0.017 (0.05/3) was needed to for a statistically significant difference.

Logistic regression analysis was performed to investigate association between variables and pCR. Multivariable logistic regression was performed for variables with a p value < 0.05 at univariable analysis. We also performed separate logistic regression analysis for each IHC subgroups (HR-positive, HER2-positive, triple-negative). For the kinetic parameters that showed a significant difference, receiver operating characteristic (ROC) curve analysis was performed to determine the optimal cutoff values by using the maximum Youden index.

Intraclass correlation coefficient (ICC) was used to evaluate interobserver agreement and absolute tumor volume agreements between time points with the following interpretation: a value of 1.0 was considered perfect agreement; 0.81–0.99, almost perfect agreement; 0.61–0.80, substantial agreement; 0.41–0.60, moderate agreement; 0.21–0.40, fair agreement; and 0.20 or less, slight agreement [16]. All statistical analyses were performed with commercially available software (SPSS [version 25.0, IBM], SAS [version 9.4, SAS Inc.], and MedCalc [version 20.011, MedCalc Software]).

Results

Baseline characteristics and pCR outcomes

Clinical and imaging characteristics of the 256 patients are shown according to pCR status in Table 2. Of the 256 cancers, 230 (89.8%) were invasive ductal carcinoma, 10 (3.9%) were invasive lobular carcinoma, 10 (3.9%) were mucinous carcinoma, 3 (1.2%) were invasive papillary carcinoma, and 3 (1.2%) were metaplastic carcinoma. According to IHC profiles, 84 (32.8%, 84 of 256) were classified as HR-positive, 85 (33.2%, 85 of 256) were HER2-positive, and 87 (34.0%, 87 of 256) were triple-negative breast cancer. Most HR-positive patients received taxane-containing NAC regimens, and HER2-positive patients received HER2 targeted therapy. In triple-negative cancers, most patients received anthracycline and taxane-based NAC. Further details are presented in Supplemental Table 2.

Table 2 Clinical characteristics and MRI features based on pathological complete response status
Full size table

Of the 256 invasive breast cancers, 106 (41.4%) achieved pCR. The rate of pCR was significantly different between HR-positive and HER2-positive tumors (13.1% [11 of 84] vs. 63.5% [54 of 85], p < 0.001), and HR-positive and triple-negative tumors (13.1% [11 of 84] vs. 47.1% [41 of 87], p < 0.001), but there was no significant difference between HER2-positive and triple-negative tumors (63.5% [54 of 85] vs. 47.1% [41 of 87], p = 0.031).

Analysis of all invasive breast cancer patients

Patients who failed to achieve pCR had a significantly higher proportion of clinically positive lymph nodes than those who achieved pCR (76.0% vs. 62.3%, p = 0.018) (Table 2). At univariable logistic regression analysis, clinical lymph node status was the only variable associated with pCR (OR, 0.521; 95% CI, 0.303, 0.897, p = 0.019). None of the imaging parameters was associated with pCR when including all patients regardless of IHC subtype (Supplemental Table 3).

Table 3 Univariable logistic regression analysis of variables associated with pCR in patients with HR-positive (n = 84) and HER2-positive invasive breast cancers (n = 85)
Full size table

Analysis according to IHC subgroup

Among HR-positive tumors, 11 (13.1%, 11 of 84) achieved pCR. At univariable logistic regression analysis, a lower MS derived from ultrafast DCE-MRI (OR, 0.982; 95% CI, 0.966, 0.999, p = 0.040), lower peak enhancement (OR, 0.993; 95% CI, 0.987, 0.999, p = 0.031), and a lower percentage of washout based on conventional DCE-MRI kinetic analysis (OR, 0.904; 95% CI, 0.821, 0.995, p = 0.039) were associated with pCR. MS_3mm showed borderline significance (OR, 0.989; 95% CI, 0.978, 1.000, p = 0.052) (Table 3). Due to the small number of pCR cases (n = 11) in this subgroup, multivariable analysis was not performed. On the basis of the results of ROC curve analysis, the optimal cutoff values for predicting pCR were as follows: MS, 118.6%/s (area under the curve [AUC]: 0.697 [95% CI: 0.587, 0.792], sensitivity: 90.9% [95% CI: 58.7, 99.8], specificity: 46.6% [95% CI: 34.8, 58.6], p = 0.007); peak enhancement, 261% (AUC: 0.733 [95% CI: 0.626, 0.824], sensitivity: 63.6% [95% CI: 30.8, 89.1], specificity: 83.6% [95% CI: 73.0, 91.2], p = 0.019); and percentage of washout, 1% (AUC: 0.772 [95% CI: 0.668, 0.857], sensitivity: 63.6% [95% CI: 30.8, 89.1], specificity: 80.8% [95% CI: 69.9, 89.1], p < 0.001).

Among HER2-positive tumors, 54 (63.5%, 54 of 85) achieved pCR. At univariable logistic regression analysis, only clinical lymph node status was associated with pCR (OR, 0.349; 95% CI, 0.123, 0.991, p = 0.048) (Table 3). None of the imaging parameters showed association with pCR in this subgroup.

Among triple-negative tumors, 41 (47.1%, 41 of 87) achieved pCR. At univariable logistic regression analysis, a higher volume ratio U1/U2 (OR, 11.787; 95% CI, 2.056, 67.539, p = 0.006) and a smaller clinical tumor size (OR, 0.962; 95% CI, 0.931, 0.994, p = 0.020) were associated with pCR. At multivariable analysis, the volume ratio U1/U2 (OR, 14.811; 95% CI, 2.228, 98.483, p = 0.005) and clinical tumor size (OR, 0.955; 95% CI, 0.918, 0.992, p = 0.019) were independently associated with pCR (Table 4). On the basis of ROC curve analysis, the optimal cutoff value for volume ratio U1/U2 was 0.53 (AUC: 0.659 [95% CI: 0.550, 0.757], sensitivity: 92.7% [95% CI: 80.1, 98.5], specificity: 47.8% [95% CI: 32.9, 63.1], p = 0.008).

Table 4 Univariable and multivariable logistic regression analysis of variables associated with pCR in patients with triple-negative invasive breast cancer (n = 87)
Full size table

Interobserver agreement for tumor volume

Tumor volume measured by ultrafast DCE-MRI showed almost perfect agreement between the two readers for both time point U1 and U2 (ICC = 0.994 for tumor volume at U1; ICC = 0.993 for tumor volume at U2). The ICC was 0.939 for tumor volume ratio U1/U2, indicating almost perfect agreement. In addition, tumor volumes measured at time point U2 showed better agreement with CAD-generated angiovolumes calculated from conventional DCE-MRI (ICC = 0.929) than those measured at time point UI (ICC = 0.760) (Supplemental Table 4).

Discussion

Although none of the pretreatment kinetic features was associated with pCR when analyzing the entire study population collectively, we found that different ultrafast DCE-MRI features were associated with pCR according to IHC subtype. Our study results suggest that although ultrafast DCE-MRI can aid in pCR prediction, a tailored approach of applying different parameters according to IHC subtype would be beneficial.

We found that a lower MS derived from ultrafast DCE-MRI, lower peak enhancement, and a lower percentage of washout based on conventional DCE-MRI were associated with pCR in HR-positive tumors. As previous studies have reported that a higher peak enhancement or MS were associated with poor histological prognostic factors and that the proportion of washout type was greater in patients achieving pCR, these results may seem unexpected [17]. However, such differences can most likely be attributed to differences in study samples. To our knowledge, there are no studies that have specifically investigated associations between pretreatment MRI kinetic parameters and pCR in only HR-positive cancers. Most previous studies have encompassed invasive breast cancers of all IHC subtypes, of which HER2-positive or triple-negative tumors show both higher rates of aggressive kinetic features and pCR. In addition, although some previous studies have classified HR-positive/HER2-positive tumors as luminal B, these would not have been included in our HR-positive subgroup [10]. In contrast to HER2-positive and triple-negative invasive breast cancer, NAC is mainly performed for locally advanced cases in HR-positive/HER2-negative breast cancer. Therefore, our HR-positive subgroup would be primarily composed of relatively large and aggressive (i.e., node-positive) luminal A breast cancer, of which baseline tumor characteristics may differ from HR-positive tumors in general. In addition, due to the poor response to NAC in HR-positive tumors, the number of pCR cases was small (n = 11). Future studies with larger study samples may be able to identify additional MRI predictors of response in this subgroup.

In the HER2-positive subgroup, only clinically negative lymph node status was associated with pCR. Our results are consistent with previous studies reporting association between clinical lymph node status and pCR in invasive breast cancer [18, 19]. Similar to previous studies [9, 10], the pCR rate was highest in this subgroup (63.5%). Considering the high pCR rate and that HER2-positive breast cancers have been reported to show significantly shorter bolus arrival time than luminal type cancer [20], differences in early pretreatment kinetic features may not be able to sufficiently differentiate pCR vs. non-pCR groups in these tumors.

In the triple-negative subgroup, we found that a smaller clinical tumor size was associated with pCR. Our results are consistent with previous studies, including a study reporting that the chance of pCR is 39.7% for clinical T2 triple-negative breast cancer and 26% for clinical T3 triple-negative breast cancer [21,22,23]. In addition, we found that a higher volume ratio U1/U2 at ultrafast DCE-MRI was associated with pCR in triple-negative breast cancers. In a previous study comparing size assessment between ultrafast and conventional DCE-MRI, tumor size at U2 was comparable to the histopathological tumor size, whereas the median tumor size at U1 was significantly smaller than tumor size at U2 [24]. Therefore, a high volume ratio U1/U2 indicates that the majority of the tumor shows very early enhancement. Multiple studies have reported association between rapid early enhancement kinetic parameters and tumor aggressiveness, such as high histological grade, hormonal receptor negativity, and high Ki-67 level [20, 25, 26]. Thus, tumors with a high volume ratio U1/U2 are likely mostly composed of aggressive components with high neovascularization that respond better to chemotherapy. Also, as triple-negative breast cancer has been reported to have a low proportion of coexisting DCIS [27], early enhancement characteristics may more solely reflect characteristics of invasive components which are related with pCR. Similar to our results, a recent study also reported that diameters in the late phase of ultrafast DCE-MRI showed no significant difference with the extent of pathological residual IDC [28]. Therefore, our study results support the potential for ultrafast DCE-MRI to substitute delayed MRI phases while preserving accuracy in pCR prediction, possibly in a combined abbreviated protocol.

However, despite efforts to accurately predict pCR earlier, mid-treatment and post-treatment MRIs are still the current standard of care for evaluating treatment response to NAC [4, 29]. In a recent study, conventional kinetic features were shown to improve the performance of mid-treatment MRI in tumor response and prognosis prediction [29]. Likewise, ultrafast kinetic features may hold higher potential in aiding pCR prediction in mid-treatment MRI, as enhancement features further differ according to the degree of response. Further studies including comparison between ultrafast and conventional MRI kinetic features or combined models would be informative in better understanding the additive value of ultrafast sequences.

Our study has several limitations. First, a patient selection bias might have occurred due to the retrospective design of our single-institution study. Second, we used angiovolumes calculated by a CAD system for tumor volume measurements at conventional DCE-MRI, which may be less accurate than manual segmentation [30]. Third, we did not evaluate tumor volumes beyond the time point of U2 at ultrafast MRI. However, a previous study reported that the median tumor sizes at U2 and at later time points were not significantly different from each other [24].

In conclusion, we found that an ultrafast DCE-MRI feature was associated with pCR in invasive breast cancer but specific features differed according to IHC subtype. A higher volume ratio of U1/U2 derived from ultrafast DCE-MRI was independently associated with pCR in triple-negative invasive breast cancer.