Prediction of pathological complete response to neoadjuvant chemotherapy for invasive breast cancers based on longitudinal ultrasound and superb microvascular imaging: a single-center retrospective study

Introduction

Neoadjuvant chemotherapy (NAC) was introduced by Frei (1982) as systemic chemotherapy for localized tumors prior to radical surgery or radiotherapy. Currently, NAC is the standard treatment for breast cancer (BC) patients to eradicate cancer cells and assess tumor response to systemic therapy (Sabeti et al., 2025). Pathological complete response (pCR) is regarded as a surrogate indicator of favorable overall, event-free, and long-term survival in BC (Early Breast Cancer Trialists’ Collaborative Group (EBCTCG), 2018). Early identification of pCR can promptly guide and adjust the treatment plan, thereby avoiding unnecessary surgery and chemotherapy cycles. Therefore, there is an urgent need to develop methods for the early prediction of pCR.

Magnetic resonance imaging (MRI) is widely regarded as the most sensitive imaging modality for assessing NAC in patients with BC (Hayward et al., 2023; Wang et al., 2023). Nevertheless, literature reports (Kim et al., 2023b) indicate that the combined use of Doppler ultrasound (US) and elastography surpasses MRI in detecting residual cancer after NAC for BC. Additionally, MRI is not suitable for patients with metal implants within their bodies or for those experiencing renal insufficiency; however, US is more convenient, cost-effective, radiation-free, and widely accessible. Therefore, it is highly recommended that patients with BC undergo regular re-evaluation of their response to chemotherapy using US after NAC (Shao et al., 2022). The echogenic rind of the breast lesion on grayscale US has a high positive predictive value (PPV) for diagnosing BC (Watanabe et al., 2021). It will be incorporated into the sixth edition lexicon of the American College of Radiology Breast Imaging Reporting and Data System (BI-RADS). Chung et al. (2022) discovered that triple-negative BC with echogenic rinds before NAC was more likely to achieve pCR. However, there were few documented reports on the changes in echogenic rinds before and after NAC. We aimed to explore whether the presence or absence of a specific factor, and its changes following NAC, were associated with pCR.

Angiogenesis plays a pivotal role in BC progression, significantly contributing to disease aggressiveness, resistance to treatment, and unfavorable prognosis (Zhang et al., 2022). Following NAC administration, the intratumoral vascular diameter and blood flow velocity decrease, posing greater challenges for detecting tumor blood supply (Kuo et al., 2008; Kumar et al., 2010). Contrast-enhanced ultrasound (CEUS) offers detailed insights into the microvasculature and hemodynamics (Zhou et al., 2020). However, this technique requires the injection of contrast agents, which not only increases the cost of examinations for patients, but also prolongs the time doctors spend on image analysis. Superb microvascular imaging (SMI) is a novel, feasible microflow imaging technique that eliminates clutter, retains low-velocity signals, and enhances sensitivity to microvasculature within the tumor without the need for a contrast agent (Kurt et al., 2023).

SMI provides two modes of vascular imaging: color superb microvascular imaging (cSMI) and monochrome superb microvascular imaging (mSMI). The cSMI simultaneously displays grayscale and color information. It features dedicated software to evaluate the vascular index (VI), which calculates the ratio between the pixels of the Doppler signal and those of the total lesion. Studies have shown that VI can help increase diagnostic accuracy in differentiating between benign and malignant breast lesions (Chae et al., 2021; Zhang et al., 2022). However, there have been few reports on the role of VI in predicting response to NAC in BC. The mSMI image is displayed side by side with the grayscale image, which reduces the background signal in the region of interest, focusing solely on the vasculature. The images are similar to those of contrast-enhanced ultrasound, but do not require contrast agents. Diao et al. (2020) reported that breast lesions with a larger scope in contrast-enhanced mode than in grayscale mode are more likely to be malignant. This phenomenon is also observable in the mSMI images. To our knowledge, there is limited understanding of its correlation with the response to NAC in BC.

Therefore, the objective of this study was to investigate whether longitudinal changes in various BC lesion parameters are associated with pCR following NAC.

Materials and Methods

Results

The multivariate model in predicting pCR

Using a multivariate logistic regression model with forward selection and the Wald test, three factors were identified as having the strongest association with pCR: ΔVI (%), ΔSD (%), and SD. As shown in Table 4, a greater rate of change in VI and SD (Δ%) was associated with a higher likelihood of patients achieving pCR. In contrast, a more extended baseline SD was inversely associated with a reduced probability of achieving pCR. Although the AUC values of ΔVI (%) and ΔSD (%) were good (0.855 and 0.906, respectively), the combined multivariate model still exhibited the best AUC (0.963), which was significantly higher than that of any single factor (Table 5) (Fig. 2).

The internal validation of the multivariate model

To evaluate the internal validity of the multivariate logistic regression model, bootstrap resampling (B = 1,000) was performed using the rms package in R. The results demonstrated strong performance in both discrimination and calibration.

The Dxy index (Somers’ Dxy) of the model was 0.927 in the original sample, indicating excellent discriminative ability. After bootstrap correction, Dxy was slightly reduced to 0.919 with a minimal optimism of 0.008. Similarly, the R² index decreased modestly from 0.788 to 0.766, suggesting stable explanatory power.

Table 4:

The multivariate model results of independent variables to pCR using forward auto-selection method and Wald test.

Parameters	Estimated B	S.E.	Wald	Adjusted OR (95% CI)	P
ΔVI (%)	0.033	0.014	5.859	1.034 (1.006 to 1.062)	0.015
ΔSD (%)	0.123	0.029	18.003	1.131 (1.068 to 1.197)	<0.001
SD	−2.576	0.993	6.726	0.076 (0.011 to 0.533)	0.009

DOI: 10.7717/peerj.20171/table-4

Notes:

Abbreviations

pCR

pathological complete response

LD

largest diameters

SD

shortest diameters

VI

vascular index

Table 5:

The comparisons of AUCs between parameters following two cycles of NAC and multivariate model.

		P-value compared to the AUC of
Parameters	AUC	ΔVI (%)	ΔSD (%)	SD	Multivariate model
ΔVI (%)	0.855	–	0.228	0.004	<0.001
ΔSD (%)	0.906	0.228	–	<0.001	0.012
SD	0.666	0.004	<0.001	–	<0.001
Multivariate model	0.963	<0.001	0.012	<0.001	–

DOI: 10.7717/peerj.20171/table-5

Notes:

Abbreviations

AUC

area under the receiver operating characteristic curve

NAC

neoadjuvant chemotherapy

LD

largest diameters

SD

shortest diameters

VI

vascular index

In Fig. 3, the original and corrected values for each parameter show similar results. In the calibration plot in Fig. 4, the distribution of the predicted and actual probabilities closely aligns with the 45-degree diagonal line. Both of these findings demonstrate the stability of the estimation.

The decision curve analysis and nomogram

DCA was performed to assess the clinical utility of the prediction model (Fig. 5). The DCA showed that the model provided a higher net benefit than the “treat all” or “treat none” strategies across a range of threshold probabilities (0.4 to approximately 0.6), supporting the model’s potential value in predicting pCR. Figure 6 depicts the nomogram of this 3-factor multivariate logistic regression model for convenience in clinical practice.

Discussion

In the current study, we combined baseline SD with the rate of change in VI and SD (Δ%) to construct an easily accessible and cost-effective model for predicting pCR after NAC in BC patients. The model yielded favorable outcomes in terms of AUC, calibration, and DCA. This could guide the timely adjustment of treatment regimens for patients who do not achieve pCR, thereby increasing the rates of pCR and preventing toxic effects.

BC is a heterogeneous disease that is characterized by multiple molecular subtypes and genetic features. The treatment plan was primarily determined by molecular classification (Gradishar et al., 2024). Literature reports that ER-negative and/or HER2-positive status are more likely to achieve pCR (Chen et al., 2023a; Chen et al., 2024; Guo et al., 2024). Our study confirmed this as well: In the pCR group, 31 (60.8%) patients were ER-negative, and 36 (70.6%) were HER2-positive.

Currently, Chinese experts strongly recommend that individuals with BC undergoing NAC should receive routine ultrasound screenings every two cycles, primarily to monitor changes in the size of breast tumors (Shao et al., 2022). Our research revealed that, prior to NAC, both LD and SD in the non-pCR group were significantly larger than those in the pCR group, indicating statistically significant differences between the two groups. Previous studies (Liu, Tang & Chen, 2022) suggested that the larger the maximum diameter of the tumor before NAC, the lower the likelihood of achieving pCR. However, Xie et al. (2023) found that there was no statistically significant difference in the baseline lesion size, but there was a difference in the change rate of the LD and volume after two cycles of treatment. Moreover, the rate of volume change was an independent predictor of pCR. However, our research indicated that although the rates of change in both LD and SD were higher in the pCR group than in the non-pCR group after two cycles of treatment, ΔSD and SD were identified as having the strongest association with pCR. To the best of our knowledge, very few studies have reported the value of predicting pCR based on the SD of the lesion. Our research showed that ΔSD demonstrated outstanding performance, achieving an AUC value as high as 0.906. We hypothesized that because the breast is a flat organ, tumors tend to shrink in a parallel manner following NAC, irrespective of whether they undergo concentric or non-concentric shrinkage (Ballesio et al., 2017). In cases of non-concentric shrinkage, the primary lesion is divided into separate sublesions, which increases the measurement error for the LD (Park, Kim & Lee, 2022), whereas the SD is relatively easier to measure. Therefore, we considered SD and ΔSD as effective indicators for evaluating the response to NAC.

Echogenic rinds refer to a thick band of high echo rings surrounding the entire or part of a breast mass, which is a recognized malignant sign of breast tumors, with a positive predictive value as high as 88%–97.5% (Watanabe et al., 2021; Kim et al., 2023a). Some studies suggest that echogenic rinds are related to the displacement of surrounding tissues, edema, and infiltration of new microvessels or lymphatics, indicating that the tumor grows faster (Park et al., 2022). Tumors that proliferate rapidly are more likely to achieve pCR (Chung et al., 2022). Chung et al. (2022) found that triple-negative breast cancers with echogenic rinds before NAC were more likely to achieve pCR. Our results are consistent with these findings. A total of 33 lesions with echogenic rinds were identified, 21 of which achieved pCR. Among these, six patients had triple-negative breast cancer. Additionally, we conducted a longitudinal study to monitor the changes in echogenic rinds following two cycles of treatment. In the pCR group, echogenic rinds disappeared in 17 lesions. In contrast, in the non-pCR group, only four lesions showed disappearance, with no instances of regenerative echogenic rinds. Although there was a statistically significant difference between the two groups, this phenomenon was not confirmed to be an independent predictive factor. Cui et al. (2021) found that neither the presence nor absence of a pretreatment echogenic rind, nor the disappearance of the echogenic rinds after two cycles of treatment, was associated with pCR. However, Chen et al. (2023b) found that narrowing or disappearance of echogenic rinds was an independent predictor of pCR. However, the value of echogenic rinds in predicting the efficacy of NAC in BC remains controversial. Further research and discussions are required.

Tumor angiogenesis refers to the abnormal proliferation of blood vessels that penetrate cancerous tumors (Park & Seo, 2017). Clinical evaluation of tumor vascularity can assist in the diagnosis of breast cancer, determine a treatment strategy, and forecast the prognosis. Our research indicated that the VI was slightly elevated in the pCR group compared to the non-pCR group before NAC; however, this difference was not statistically significant. Qi et al. (2023) and Lee & Chang (2024) agreed with us, believing that breast cancer is a hypervascular heterogeneous mass, and the VI based on a two-dimensional (2D) view cannot comprehensively assess the entire tumor vascularity. Moreover, the vascularity of large tumors may be underestimated because the Doppler signal diminishes with depth. Study (Mo et al., 2023) suggest that NAC affects the angiogenic activity of tumors, potentially impeding their further development. Therefore, we believe that monitoring and analyzing changes in vascular characteristics could serve as biomarkers for evaluating the response to NAC. It was exciting to observe that after two cycles of NAC, the change rate of VI in the pCR group was significantly higher than that in the non-pCR group, demonstrating medium predictive potential. The AUC value was 0.855 (95% CI [0.785–0.926]). NAC can affect both the cells and stroma of the tumor, including its microvessels. Consequently, the response of the tumor vascular bed allows for indirect assessment of the tumor’s sensitivity to drug reactions (Pavlov et al., 2023). Therefore, the alteration rate of VI can serve as a valuable imaging biomarker for predicting pCR to NAC in breast cancer.

mSMI and cSMI are the two manifestations of SMI. However, compared with cSMI, reports on mSMI are relatively scarce. Diao et al. (2020) research found that the diagnostic performance of mSMI appears to be equivalent to that of CEUS. However, their concern was the difference in diagnostic efficacy between CEUS and mSMI in penetrating blood vessels or vascular features. We observed that in 76 lesions (71.03%), the lesion scope depicted in the mSMI images was larger than that shown in grayscale US. This characteristic also serves as a key indicator for CEUS in the diagnosis of breast cancer. We further investigated whether the scope continued to expand after two cycles of NAC treatment. The results indicated that 20 lesions did not show further expansion in the pCR group, whereas three lesions did not expand in the non-pCR group. We speculate that this may be due to the effect of chemotherapy drugs causing atrophy and collapse of the tumor neovasculature, which also reflects a reduction in tumor invasiveness. However, the mSMI0 and its changes were not included as indicators in the final predictive model. Further studies with a larger number of patients may help provide additional value to mSMI.

This study had several limitations. First, this was a single-center retrospective study with a small sample size. Second, while the molecular classification of breast cancer is closely related to the efficacy of NAC, we did not conduct a more in-depth stratified analysis based on the molecular subtypes. Third, the measurement of the quantitative parameter VI required scanning of the section with the most abundant blood flow. However, it was challenging to ensure that the same section was selected during the different treatment cycles. Finally, this study was fundamentally tied to two-dimensional ultrasound imaging. Future research should incorporate volumetric imaging of tumors, thereby providing better insights into microvascular changes with NAC.

Conclusion

In summary, the rate of change in SD following two cycles of NAC has emerged as a crucial dependent factor for predicting pCR. The nomogram constructed from the combined predictive models of ΔSD (%), SD, and ΔVI (%) demonstrated satisfactory performance, indicating that the nomogram is a reliable method for predicting pCR after NAC. However, further research using larger validation datasets is essential before this model can be practically applied.

Supplemental Information