Early assessment of shear wave elastography parameters foresees the response to neoadjuvant chemotherapy in patients with invasive breast cancer

Background Early prediction of tumor response to neoadjuvant chemotherapy (NACT) is crucial for optimal treatment and improved outcome in breast cancer patients. The purpose of this study is to investigate the role of shear wave elastography (SWE) for early assessment of response to NACT in patients with invasive breast cancer. Methods In a prospective study, 62 patients with biopsy-proven invasive breast cancer were enrolled. Three SWE studies were conducted on each patient: before, at mid-course, and after NACT but before surgery. A new parameter, mass characteristic frequency (fmass), along with SWE measurements and mass size was obtained from each SWE study visit. The clinical biomarkers were acquired from the pre-NACT core-needle biopsy. The efficacy of different models, generated with the leave-one-out cross-validation, in predicting response to NACT was shown by the area under the receiver operating characteristic curve and the corresponding sensitivity and specificity. Results A significant difference was found for SWE parameters measured before, at mid-course, and after NACT between the responders and non-responders. The combination of Emean2 and mass size (s2) gave an AUC of 0.75 (0.95 CI 0.62–0.88). For the ER+ tumors, the combination of Emean_ratio1, s1, and Ki-67 index gave an improved AUC of 0.84 (0.95 CI 0.65–0.96). For responders, fmass was significantly higher during the third visit. Conclusions Our study findings highlight the value of SWE estimation in the mid-course of NACT for the early prediction of treatment response. For ER+ tumors, the addition of Ki-67improves the predictive power of SWE. Moreover, fmass is presented as a new marker in predicting the endpoint of NACT in responders.


Introduction
Neoadjuvant chemotherapy (NACT) is an established therapeutic strategy for operable breast cancers and locally advanced breast cancers and allows more patients to undergo breast-preserving surgery [1,2]. A pathological complete response (pCR) to NACT is associated with increased disease-free interval. However, responses to NACT are quite variable. With the increased use of NACT, it is crucial to have an accurate prediction of tumor response to NACT.
Tissue stiffness has been demonstrated to be significantly correlated with tumor growth as cancer development and progression require extensive reorganization of the extracellular matrix (ECM) [16]. Increased deposition of collagen and other ECM molecules enhances the stiffness of tumoral stroma [17][18][19]. Changes in tumor stiffness were significantly greater in patients who had a good response to NACT compared to those resistant to NACT [20]. Breast cancer pre-and post-treatment stiffness obtained from SWE was significantly correlated with the presence of residual cancer [8,9]. A study in [21] showed that the SWE stiffness measured after 3 cycles of NACT and changes in stiffness from baseline were strongly associated with pCR after 6 cycles. The combination of the post-treatment SWE and greyscale ultrasound has also been shown to be promising for end-of-treatment identification of residual disease and thus response to NACT, with similar accuracies found in assessment by MRI [22].
In the current study, a new SWE parameter mass characteristic frequency (f mass ) was used. f mass is defined as the ratio of the averaged minimum shear wave speed (SWS) within the regions of interest (ROIs) to the largest mass dimension. The physical meaning of the new parameter can be explained as the inverse of the maximum shear wave propagation time in a breast mass. The motivation for using f mass is that in SWE, the SWS of small masses is often underestimated due to their small size compared to the wavelength. This error may lead to the false-negative diagnosis of such masses. f mass represents the SWS weighted by the inverse of mass diameter; therefore, f mass assumes a larger value for masses that are too small. Thus, one may expect f mass to be a more robust parameter in SWE than SWS itself. However, we want to emphasize that f mass is not meant to compensate for the underestimation error of SWS in a mathematical sense. Our intention is to introduce f mass as a new metric that improves the characterization of breast masses in a statistical sense. The purpose of this study was to investigate the role of SWE parameters, including mass characteristic frequency, in evaluating the breast tumor response to the NACT treatment. For ER-positive tumors, the combination of the SWE parameters with Ki-67 was further studied to improve the sensitivity and specificity of the response prediction.

Study population
This prospective study was Health Insurance Portability and Accountability Act (HIPAA) compliant and was approved by the Institutional Review Board (IRB) (IRB application #12-003329). From January 2014 to September 2020, 62 female patients (age range 27-78 years) with 62 biopsy-proven invasive breast cancers were recruited in this study. During the recruitment, patients with prior mastectomy or breast implant were excluded. One patient with a previous lumpectomy in the contralateral breast was included in this study. A signed written informed consent with permission for publication was obtained from each enrolled patient prior to the study.
Imaging SWE studies were conducted for each patient at three time points: before initiation of NACT, at the midcourse of NACT, and after completion of NACT but prior to surgery. A flow diagram of the study population is summarized in Fig. 1.
The 2D SWE scanning was performed by one of our two experienced sonographers, using the GE LOGIQ-E9 ultrasound clinical scanner equipped with a 2-8 MHz linear array probe (9L-D, GE Healthcare, Wauwatosa, WI) for both the conventional B-mode and SWE data acquisition. To reduce motion artifacts, patients were instructed to suspend respiration for approximately 3 s during the data acquisition. The SWE measurement was acquired within a rectangle-shaped field of view, which covered the whole lesion and the adjacent normal tissue. For each lesion, along the same orientation, at least four SWE images were obtained. One of the consistent stiffness maps was chosen to draw ROIs. Three nonoverlapping ROIs, 3 mm in diameter, were placed at the stiffest position of the lesion, with peritumoral stroma included. One ROI was placed at the surrounding normal tissue. The mean SWS, maximum SWS, minimum SWS, and standard deviation of the SWS inside each ROI were calculated by the ultrasound machine. The SWS for the tumor was represented by the average values of the three ROIs placed at the stiffest position. A new shear wave parameter, mass characteristic frequency, represented by f mass , is used in this paper: f mass = 1000V min /d, where f mass is with unit Hz, V min is the minimum SWS with unit m/s, and d is the mass size in mm and recorded as the maximum dimension of the tumor shown on the B-mode image. Figure 2 illustrates the measurements for calculating f mass .

Clinical pathologic data
The parameters for residual cancer burden (RCB) measurement for each patient were obtained from the surgical pathology report, and the RCB score was calculated with an empirical equation: RCB = 1.4(f inv dprim ) 0.17 + [4(1-0.75 LN )d met ] 0.17 [23]. RCB was on a continuous scale and was further categorized as 0 (RCB = 0), I (0 < RCB < =1.36), II (1.36 < RCB < =3.28), or III (RCB > 3.28). Categories 0 and I were regarded as responders while categories II and III were regarded as nonresponders [23,24]. Of the 62 patients included in this study, 60 underwent lumpectomy or mastectomy, and the corresponding RCB scores were calculated from the surgical reports. Two patients did not undergo surgery, one progressed on neoadjuvant chemotherapy and died before undergoing surgery, and one developed metastatic disease, and therefore, surgery was not indicated and ultimately died 3.5 years after diagnosis. The two deceased patients were categorized as non-responders. The MRI dimensions (AP, trans, and SI) were read from the clinical MRI image that was acquired close to the date of the SWE scanning. The MRI volume was calculated as the product of the three dimensions (volume = AP × trans × SI). The clinical MRI was based on the dynamic enhanced protocol using intravenous (IV) contrast administration. Obtained T1-and T2-weighted images were analyzed with computer-aided detection (CAD) image analysis. With the help from a radiologist, the maximum lesion size was read from the B-mode imaging.
Estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and Ki-67 proliferative index status of the pre-NACT tumor needle core biopsies were obtained from the clinical record. ER and PR were considered negative if less than 1% of invasive tumor cells were immunoreactive, and were considered positive if greater than or equal to 1% of invasive tumor cells were immunoreactive. As per the   [25], immunohistochemical HER2 scores or of 0 and 1+ were considered negative and a score of 3+ was scored as positive. Equivocal HER2 immunostains (HER2 scores of 2+) underwent fluorescence in situ hybridization testing for HER2 amplification and were classified as per the ASCO/CAP guidelines. Ki-67 immunostain was reported as a percentage of positively staining nuclei.
Based on the clinical biomarker data, the tumors were divided into five molecular subtypes according to the St. Gallen criteria [26]:

Statistical analysis
The measured SWS was converted to elasticity expressed in kilopascals [27]. Changes of SWE parameters were also calculated: The subscripts 1, 2, and 3 indicate the corresponding parameters measured at the first, the second, and the third visits, respectively.
Statistical analysis was conducted with RStudio (RStudio, PBC, Boston, MA). The Kruskal-Wallis test and Pearson's chi-squared test were used in the statistical analysis to calculate the p value for continuous data and count data, respectively. p < 0.05 was considered indicative of a statistically significant difference. Leave-one-out cross-validation (LOOCV) [28] was used to assess the effect of multiple factors on the prediction of the response to NACT. Receiver operating characteristic (ROC) curve analysis was used to calculate the area under the curve (AUC) and determine the cutoff values, as well as the corresponding sensitivity and specificity. The optimal cutoff was defined as the point closest to the point (0, 1) on the ROC curve.

Results
Clinical parameters during the NACT Patient demographic and tumor characteristics of the 62 patients enrolled in this study are presented in Table 1. Overall, as expected, compared to other histologic subtypes, patients with invasive ductal carcinoma had higher rates of response to NACT (p = 0.03), and higher tumor grade (grade III) had a higher response rate of 67.7% compared to 32.3% to lower grade tumors (grade I/II). Among different ER-positive molecular subtypes, a significant difference was found for the response rate (p = 0.03) with the highest response rate seen in Luminal B (HER2+) type cancers. Table 2 summarizes the MRI volume (V MRI ), mass size (s), and the SWE parameters, including the new parameter f mass . The averaged MRI volume, shear wave elasticity, and mass size decreased during the NACT treatment for both the responder group and the non-responder group. Tumor response was significantly correlated with the values of E mean-ratio1 measured during the first SWE visit; s 2 , E mean2 , and E max2 measured during the second SWE visit; and V MRI3 , For all three visits, non-responders showed higher averaged elasticity, elasticity ratio, and lower mass characteristic frequency. No significant difference was found in the change of elasticity, although there was a trend for the averaged change in stiffness in the responder group to be higher. A significant difference was found in the change of the mass characteristic frequency measured between the first and the third visits (f mass1-3 , p < 0.001). Figures 3 and 4 show the typical SWS maps for a responder and a non-responder for the three SWE studies, respectively; E mean and f mass for different molecular subtypes measured during the three visits are shown in Figs. 5 and 6, respectively, indicating that stiffness decreased significantly for the responders, while remained high for the non-responders; the f mass remained low for non-responders and increased significantly for responders.
Leave-one-out cross-validation for the NACT response prediction Selected SWE parameters measured at each visit were combined using the LOOCV to predict the NACT treatment response, and the models were denoted as the noninvasive models. The E mean_ratio1 and s 1 were combined for the first visit, the E mean2 and s 2 were combined for the second visit, and the E max_ratio3 and f mass3 were combined for the third visit. The corresponding ROCs are shown in Fig. 7a. The AUCs, optimal cutoffs, and the corresponding specificity and sensitivity are summarized in Table 3. The AUC (0.84, 0.95 CI 0.78-0.97) for the third visit was the highest among the three visits. A significant difference was found for the AUC between the first and the third visits (p = 0.04). No significant difference was found between the second and the third visits (p = 0.29). Therefore, the AUC (0.75, 0.95 CI 0.62-0.88) for the second visit gave the second best prediction, given the balance between timing and accuracy.
For ER-positive tumors, the Ki-67 index obtained from the pre-NACT biopsy was then added to the selected SWE parameters included in the noninvasive models with the LOOCV to predict NACT response. The corresponding models were denoted as the mixed models, with ROCs shown in Fig. 7b. The AUCs, optimal cutoffs, and the corresponding specificity and sensitivity are summarized in Table 3. Among the mixed models, the ROC for the first visit showed the best prediction (0.80, 0.95 CI 0.65-0.96). With an optimal of 0.68, the specificity was 0.93 and the sensitivity was 0.70.
The AUCs for both the invasive models and the mixed models during the three visits are compared in Fig. 7c. The AUCs for the NACT response prediction with the MRI volume during the three visits were also plotted. The corresponding AUCs were 0.62 (0.95 CI 0.45-0.78), 0.73 (0.95 CI 0.46-1.00), and 0.73 (0.95 CI 0.54-0.92), respectively. No significant difference was found among the three AUCs. Since most patients only had the first MRI imaging before NACT, the number of MRI volume results available for the second and third visit was relatively small. Therefore, we also included the AUCs of the MRI volume adapted from [24] for reference. In contrast to the calculation method used in this study, tumor volume in [24] was computed by summing all voxels with percentage enhancement (PE) above a nominal threshold value of 70%, and PE = ((S 1 -S 0 )/S 0 ) × 100%, where S 0 , S 1 , and S 2 represented the signal intensities on the precontrast, early postcontrast, and late postcontrast images, respectively. Data are mean ± standard deviation; data in parentheses are mass numbers E mean mean elasticity, E max maximum elasticity, E mean_ratio ratio of the mean elasticity between the mass and the surrounding normal tissue, E max_ratio ratio of the maximum elasticity between the mass and the surrounding normal tissue, f mass mass characteristic frequency The subscripts 1, 2, and 3 indicate the corresponding parameters measured at the first, the second, and the third visit, respectively *p < 0.05; the difference is statistically significant

Discussion
The results of our study demonstrate that SWE aids accurate assessment and early prediction of tumor response to NACT; for ER-positive tumors, combining the Ki-67 index with some SWE parameters can further improve the response prediction. Our study showed that the averaged stiffness decreased for both the responders and the non-responders. Moreover, the changes in stiffness for both the E mean and E max between the first two visits were larger than the changes between the second and the third visits. The slower drop in stiffness during the later course of NACT treatment may be due to hypoxia, which is associated with increased matrix stiffness in the non-necrotic area of the tumors [29,30]. tumor (grade III invasive ductal carcinoma) measured a before NACT, b during the mid-course of NACT, and c after NACT but before surgery. The mean elasticity, maximum elasticity, and mass characteristic frequency are shown in d. This is a responder with RCB score of 0. NACT, neoadjuvant chemotherapy; RCB, residual cancer burden Fig. 4 Shear wave speed map for a 59-year-old female with Luminal A tumor (grade II invasive ductal carcinoma) measured a before NACT, b during the mid-course of NACT, and c after NACT but before surgery. The mean elasticity, maximum elasticity, and mass characteristic frequency are shown in d. This is a non-responder with an RCB score of 1.6. NACT, neoadjuvant chemotherapy; RCB, residual cancer burden Moreover, it has been shown that residual tumor appears to be stiffer than the fibrous tissue left for the pCR [22]. These changes in stiffness can also be detected with SWE during the NACT treatment [8,31].
The f mass is a newly introduced SWE parameter for breast cancer characterization, and a lower f mass value is significantly correlated with poor histologic prognostic factors. The average f mass value for the non-responders stayed relatively constant throughout the NACT treatment. Recalling the definition of f mass , this result indicated that the change of SWS and mass size during the NACT was such that the net effect on f mass was minimal for the non-responders. This study showed that the averaged f mass for the responders was higher than that for the non-responders during all three visits. For the third visit, a significant difference was found among the responders and non-responders. Among the responders, there was no significant difference between f mass1 and f mass2 . When compared to the average f mass1 , the average f mass3 significantly increased by 70%, indicating that the Fig. 5 The mean shear wave elasticity measured before NACT, at the mid-course of NACT, and after NACT but before surgery for different molecular subtypes: a-c Luminal A type, d-f Luminal B (HER2-) type, g-i Luminal B (HER2+) type, j-l HER2+ type, and m-o TN type. NACT, neoadjuvant chemotherapy; TN, triple-negative effect of size reduction was dominant on f mass (compared to the change of minimum SWS) after the mid-course of NACT. A similar trend was observed in each molecular subtype. However, the average E mean decreased continuously throughout the NACT. When compared to the first measurement, the average E mean decreased by 48% during the second visit and by 62% during the third visit. Therefore, f mass could be used as a useful indicator to determine the end-of-treatment point for the NACT for responders. Currently, response to NACT is routinely assessed using MRI, clinical ultrasound, and mammography [22]. The addition of f mass could further facilitate the individualized NACT treatment plan. We plan to extend our study to a larger number of patients and with more frequent f mass measurements in the course of NACT to further investigate the role of f mass in predicting earlier response to NACT.
Predictions of response to NACT with the noninvasive models which combine the SWE parameters show promising results for all three visits. To balance the timing and accuracy, the parameters obtained during the mid-course of NACT could be used for evaluating the Fig. 6 The mass characteristic frequency measured before NACT, at the mid-course of NACT, and after NACT but before surgery for different molecular subtypes: a-c Luminal A type, d-f Luminal B (HER2-) type, g-i Luminal B (HER2+) type, j-l HER2+ type, and m-o TN type. NACT, neoadjuvant chemotherapy; TN, triple-negative NACT treatment outcome. Similarly, a previous study showed that the optimal time for early evaluation to identify patients who would be responsive to treatment was after 2 cycles of treatment and immediately before the third cycle of the therapy [29].
The Ki-67 index is an important predictive factor for the effectiveness of NACT [32,33]. Breast cancer with a high Ki-67 index level has repeatedly been shown to respond better to chemotherapy. A previous study also showed that there was no pathological complete response in cases with Ki-67 < 25% [34]. In this study, we found that the average Ki-67 value was higher in responders than in non-responders. Mixed models, which combined the Ki-67 index obtained before NACT with the noninvasive models, were proposed for ER-positive tumors. When compared to the noninvasive model for the second visit, an earlier prediction with improved accuracy could be achieved with the mixed model. Similarly, the study by Ma et al. has shown the potential value of adding Ki-67 to shear wave parameters for better and earlier prediction of the response to therapy [35].
Both the noninvasive models and the mixed models were generated with the leave-one-out cross-validation analysis, which included internal validations to quantify any optimism in the predictive performance and adjust the models for overfittings [36,37]. Therefore, the models proposed in this paper for predicting the response to NACT have high reproducibility and stability.
Studies showed that MRI volumetric assessment was more accurate than the diameter for earlier detection of treatment response [22,24]. The volume from clinical MRI was also recorded in this study for all three visits. Though the MRI volume data was limited in this study, the ROCs were comparable to the results from a previous study with a larger patient number. Moreover, both the AUCs from the noninvasive model for the second visit and from the mixed model for the first visit were higher than the MRI volume prediction at the corresponding Fig. 7 a ROCs for the three noninvasive models generated with the leave-one-out cross-validation, which is based on the combination of E mean_ratio1 and s 1 for the first visit, the combination of E mean2 and s 2 for the second visit, the combination of E max_ratio3 and f mass3 for the third visit. b ROCs for the three mixed models which were generated with the Ki-67 index added to the noninvasive models for ER-positive tumors. c Comparisons of the AUCs for the ROCs for NACT response prediction with the MRI volume from reference [23], the MRI volume in this study, the noninvasive models, and the mixed models. The 1st visit for the MRI volume measurement in [23] was after the first cycle of NACT treatment. ROC, receiver operating characteristic; SWE, shear wave elastography; PR, progesterone receptor; NACT, neoadjuvant chemotherapy; AUC, area under the curve; MRI, magnetic resonance imaging; IHC, immunohistochemical Therefore, a future study based on a larger population will be helpful for comparing the results from the MRI prediction and the proposed models in this study.
There are some limitations in this study. Firstly, the sample size was relatively small; moreover, as shown in the flowchart, some patients did not complete all three visits for the SWE studies, leading to some missing data. However, the leave-one-out cross-validation has been applied to compensate for the sample number limitation. Secondly, this is a one-center study. Thus, a multicenter study with a larger population is required to further investigate the role of SWE parameters in NACT response prediction.
In summary, this study investigated the application of SWE in discriminating the responders from nonresponders to chemotherapy, before NACT, during the mid-course of NACT, and after NACT but prior to surgery. In conclusion, when the noninvasive model is used for response prediction, a balance between the timing and accuracy is achieved when the E mean2 and s 2 are measured during the mid-course of the treatment. For ER-positive tumors, an even earlier and more accurate response prediction could be obtained with the combination of Ki-67 index, E mean_ratio1 , and s 1 measured before the treatment using the mixed model. Moreover, this study also shows that f mass is useful in determining the endpoint of the NACT.

Conclusions
Our study findings highlight the value of SWE estimation in the mid-course of NACT for the early prediction of treatment response. For ER+ tumors, the addition of Ki-67 improves the predictive power of SWE. Moreover, f mass is presented as a new marker in predicting the endpoint of NACT in responders. These results may facilitate personalizing the treatment regimens of patients with breast cancer receiving NACT. Furthermore, the role of these SWE parameters can be validated in the future by carrying out a multicenter prospective study with a larger patient population.