Integrating delta radiomics and changes in multimodal ultrasound features for predicting response to neoadjuvant chemotherapy in breast cancer

Introduction

Breast cancer is one of the most common malignancies among women worldwide (1). Neoadjuvant chemotherapy (NAC) is a standard treatment for locally advanced and early-stage high-risk breast cancer because it can reduce tumor burden and increase the rate of breast-conserving surgery (2). Nevertheless, due to the genetic heterogeneity of breast tumors, patients exhibit substantial variability in their responses to NAC (3). Currently, patients who achieved pathologic complete response (pCR) after NAC have better survival (4). However, the status of pCR is known only after postoperative pathology assessment, which may delay treatment adjustment and expose patients to ineffective chemotherapy. Therefore, early response assessment is clinically important for optimizing treatment decisions.

Magnetic resonance imaging (MRI) and ultrasound (US) are commonly used for response assessment during NAC. Although MRI plays an important role in treatment monitoring, its reported specificity is limited (5), and its high cost may limit routine clinical use. In contrast, multimodal US provides information on tumor morphology, blood perfusion, and tissue stiffness, enabling longitudinal assessment during NAC (6). However, US interpretation remains highly operator-dependent, resulting in inter-observer variability and limited objectivity of conventional assessment (7).

In this context, US radiomics enables objective quantification of tumor heterogeneity and may improve response assessment during NAC. Because tumor characteristics can change over the course of chemotherapy, longitudinal imaging features may be more indicative than those obtained at a single time point. Delta radiomics captures these temporal changes and may therefore improve response prediction (8,9).

Beyond intratumoral texture changes, NAC also induces macroscopic alterations in tumor morphology. Tumor regression patterns are typically categorized as concentric shrinkage (CS) and non-concentric shrinkage (NCS) (10). Prior studies have demonstrated that the CS pattern is correlated with pCR, suggesting that shrinkage pattern may serve as a morphological imaging marker of treatment sensitivity (11). Shrinkage pattern may complement radiomics by capturing macroscopic morphological response. However, most studies of tumor shrinkage patterns have focused on MRI (12,13), and US-based studies integrating longitudinal multimodal US changes, shrinkage patterns, and delta radiomics remain limited. In addition, the utility of these imaging markers for response assessment may depend not only on their association with treatment sensitivity, but also on the treatment stage at which they are evaluated.

Nevertheless, the optimal timing for US-based response assessment during NAC remains unclear. Most US-based studies for NAC response prediction have focused on single-time-point imaging or comparisons between baseline and either early- or post-treatment scans. Because the mid-NAC stage may reflect emerging treatment sensitivity while still allowing therapeutic adjustment, this time point may be particularly valuable for response assessment (14). Accordingly, integrating longitudinal intratumoral and morphological imaging changes from pre- to mid-NAC may provide a more comprehensive basis for response prediction.

Therefore, this study aimed to integrate longitudinal changes in pre- and mid-NAC multimodal US, delta radiomics features (RFs), and US-based shrinkage patterns to develop a model for predicting major histopathological response (MHR) in breast cancer, and to preliminarily explore the associations between US-defined shrinkage patterns and immunohistochemical (IHC) characteristics, thereby supporting noninvasive response assessment. We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0342/rc).

Methods

Study population

A retrospective analysis was conducted on 101 women with pathologically confirmed breast cancer who completed 6–8 cycles of NAC at Zhangzhou Affiliated Hospital of Fujian Medical University between April 2023 and March 2026. Treatment response was assessed using the postoperative Miller-Payne grading system (15). Patients with grades 4–5 were classified as having MHR, whereas those with grades 1–3 were assigned to the non-MHR group (16).

Eligible patients met the following criteria: (I) histologically confirmed breast cancer with available IHC results; and (II) multimodal US examinations performed both before NAC and at the mid-treatment stage. Exclusion criteria were as follows: (I) incomplete NAC treatment, (II) concomitant malignant tumors or distant metastasis, and (III) incomplete clinical or pathological information.

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Zhangzhou Affiliated Hospital of Fujian Medical University (approval No. 2025KYZ361), and the requirement for informed consent was waived. The overall workflow of the study is shown in Figure 1.

Figure 1 Flowchart of radiomics features and features of multimodal ultrasound extraction, model establishment and analysis. ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; ICC, intraclass correlation coefficient; IHC, immunohistochemistry; Ki-67, Ki-67 proliferation index; LASSO, least absolute shrinkage and selection operator; NAC, neoadjuvant chemotherapy; PR, progesterone receptor; ROC, receiver operating characteristic; SHAP, SHapley Additive exPlanations.

Clinical and IHC markers

Clinical characteristics, such as age and body mass index (BMI), were recorded. IHC data from pre-NAC biopsy specimens were also recorded, including estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and Ki-67 proliferation index (Ki-67). ER and PR positivity was defined as nuclear staining in at least 1% of tumor cells. HER2 positivity was defined as an IHC score of 3+ or a score of 2+ with fluorescence in situ hybridization confirmation of gene amplification. High Ki-67 expression was defined as ≥14%. These IHC markers were further compared among US-defined shrinkage pattern groups (3).

Multimodal US

Instruments and methods

Multimodal US examinations were performed before NAC and at the mid-treatment time point. According to the NAC protocol, patients received six or eight cycles of treatment, and imaging data were collected after completion of the third or fourth cycle. All examinations were administered by a senior sonographer with a decade of experience in breast US diagnosis, following standardized operating procedures and instrument parameter settings. The US examinations were conducted using a Siemens ACUSON Sequoia US diagnostic system equipped with an L10-4 linear array probe (frequency 4–10 MHz). Multimodal US was performed on breast lesions according to a standardized scanning procedure, and standard cross-sectional images were saved for subsequent analysis. All images were independently interpreted by two experienced sonographers who were blinded to pathological results. In the event of differing opinions, consensus was achieved after discussion to assign the results.

US observation indicators and assessment

Two-dimensional US: the lesion was examined from multiple angles and planes, and two orthogonal images showing representative features were preserved. Lesion location, maximal diameter, and internal echogenicity were recorded. Changes in two-dimensional US features between pre- and mid-NAC were compared.

Shrinkage pattern: US shrinkage patterns were classified on the basis of post-treatment changes in tumor size, shape, and internal echogenicity. Lesions were divided into CS and NCS. CS was defined as a uniform reduction of the lesion toward its center, resulting in a regular and continuous residual contour. NCS was defined as a non-uniform reduction pattern, including echogenicity change, partial retraction, and dendritic retraction (Figure 2).

Figure 2 Schematic representation of shrinkage patterns. (A,B) Concentric shrinkage. (C-H) Non-concentric shrinkage. (C,D) Partial retraction. (E,F) Dendritic retraction. (G,H) Correspond to echogenicity changes.

Color Doppler flow imaging (CDFI): intralesional vascularity was semi-quantitatively assessed according to the Adler grading system (17) and categorized into low vascularity (grades 0–I) and high vascularity (grades II–III). Changes in Adler grades between pre- and mid-NAC were recorded as increased, decreased, or unchanged.

Shear wave elastography (SWE): maximum elasticity (Emax) and mean elasticity (Emean) were measured using SWE. The region of interest (ROI) was placed on the largest cross-sectional plane of the lesion while avoiding necrotic, cystic, and visibly calcified areas. Each parameter was measured three times, and the mean value was used for analysis. Relative changes in SWE parameters were calculated as follows: ΔSWE = (X_Mid − X_Pre)/X_Pre, where X represents Emax or Emean.

Contrast-enhanced ultrasound (CEUS): SonoVue was used as the contrast agent. Following bolus injection of 4.8 mL via the antecubital vein, the contrast agent was flushed with 5 mL of normal saline. Continuous dynamic imaging was acquired for at least 120 s. Qualitative features, including enhancement area, perforating vessels, enhancement shape, and enhancement margin, were assessed at pre- and mid-NAC, and treatment-related changes were compared.

Radiomics

RF extraction and delta calculation

For each lesion, the largest tumor cross-sectional slice on Two-Dimensional US images acquired at pre- and mid-NAC was chosen for analysis. Image segmentation was completed in 3D Slicer (version 5.10.0) by two senior radiologists independently, with pathological information masked. The lesion boundary was manually traced to cover the tumor region. Using PyRadiomics, a series of quantitative features were extracted. To test segmentation robustness, 30 lesions were randomly selected for repeat annotation by both readers. Intraclass correlation coefficient (ICC) <0.75 was discarded. Delta RFs were calculated as RF_Delta = (RF_Mid − RF_Pre)/RF_Pre.

RF selection and construction of the delta radiomics score (delta Rad-score)

Feature selection was performed exclusively in the training cohort. Delta RFs retained after ICC filtering were standardized using z-score normalization, and near-zero variance features were removed. Among pairs of features with a Pearson correlation coefficient >0.90, one feature was removed. Least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation was then applied for feature selection and dimensionality reduction. The optimal penalty parameter was determined according to the 1-standard-error criterion (lambda.1se), and features with non-zero coefficients were retained. The delta Rad-score was computed from the selected delta RFs based on their assigned weights.

Model development, validation, and SHapley Additive exPlanations (SHAP) interpretation

The 101 patients were randomly divided into a training cohort (n=71) and a validation cohort (n=30) by stratified sampling. In the training cohort, IHC variables and US-related parameters were first screened by univariable analysis and then entered into multivariable logistic regression to identify independent predictors of MHR. An US-IHC model, a delta-radiomics model based on the delta Rad-score, and a combined model integrating both were then developed.

Model performance was assessed using discrimination, calibration, and clinical utility metrics, including the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, Brier score, calibration curves, and decision curve analysis. Differences in the AUC were compared using the DeLong test. A nomogram based on the combined model was constructed for individualized prediction of MHR, and SHAP analysis was performed to improve model interpretability. The validation cohort was used for independent assessment.

Statistical analysis

Statistical analyses were performed using R software (version 4.5.0). Continuous variables were compared using the independent-samples t-test or Mann-Whitney U-test, and categorical variables using the Chi-square test. A two-sided P<0.05 was considered statistically significant.

Results

Baseline clinicopathological features and US changes by MHR status

A total of 101 patients with breast cancer who received NAC were included in this study. With MHR predefined as the primary endpoint, 54 patients achieved MHR after treatment. The MHR rate was 53.52% in the training group and 53.33% in the validation group. As shown in Table 1, the primary and validation cohorts were well-balanced regarding baseline clinical characteristics, immunomic profiles, and multimodal US features, and no statistically significant intergroup differences were identified.

Association between shrinkage patterns and IHC biomarkers

As shown in Table 2, ER-negative (50.0% vs. 17.6%, P=0.001) and PR-negative tumors (62.0% vs. 19.6%, P<0.001) were more common in patients with CS than in those with NCS. HER2 status showed no association with CS (P=0.87). A higher proportion of elevated Ki-67 expression was observed in CS (96.0% vs. 82.4%), a however, no statistically significant difference was identified (P=0.06).

Univariable logistic regression analysis of predictors of MHR

Results of the univariable logistic regression are summarized in Table 3. Six variables were significantly associated with MHR, including shrinkage pattern [odds ratio (OR), 33.214; P<0.001], ER status (OR, 0.070; P<0.001), PR status (OR, 0.070; P<0.001), Δmaximum diameter (OR, 0.953; P<0.001), change in enhancement area (OR, 2.189; P=0.02), and ΔEmax (OR, 0.989; P=0.03). The remaining variables did not reach statistical significance (all P>0.05).

Multivariable logistic regression analysis of factors associated with the odds of MHR

In multivariable analysis, shrinkage pattern, PR status, Δmaximum diameter, and enhancement area change were independently related to MHR (Table 4). Among them, shrinkage pattern had the strongest effect (OR,92.613; P=0.001). PR positivity (OR, 0.017; P=0.04) and larger Δmaximum diameter (OR, 0.950; P=0.03) were linked to lower odds of MHR, whereas enhancement area change shows an association with higher odds (OR, 7.286; P=0.01). ER status and ΔEmax were not significant in the model (both P>0.05).

Feature selection and construction of the delta Rad-score

Radiomics analysis yielded 831 features from US ROIs obtained at pre-NAC and mid-NAC US ROIs. After interobserver consistency testing, 748 features with ICC >0.75 were kept. Delta features were then generated, and LASSO regression selected three features with non-zero coefficients derived from the Gray Level Dependence Matrix (GLDM) and Neighboring Gray Tone Difference Matrix (NGTDM): original_gldm_GrayLevelNonUniformity, wavelet_LLL_ngtdm_Busyness, and original_ngtdm_Strength (Figure 3). The delta Rad-score was calculated as follows:

ΔRad-score=−0.18509+0.614631×original_gldm_GrayLevelNonUniformity−0.916007×original_ngtdm_Strength+0.314828×wavelet_LLL_ngtdm_Busyness[1]

Figure 3 Feature selection extraction. (A,B) LASSO analysis of delta radiomics features; (C) Delta radiomics features’ coefficients. 1se, one standard error; GLDM, Gray Level Dependence Matrix; LASSO, least absolute shrinkage and selection operator; LLL, low-low-low; min, minimum; NGTDM, Neighboring Gray Tone Difference Matrix.

Diagnostic performance of the models

In the training cohort, the combined model showed the best discriminative ability for predicting MHR, with an AUC of 0.949, compared with 0.941 for the US-IHC model and 0.751 for the delta radiomics model (Figure 4A). It also achieved the highest accuracy (0.887) and specificity (0.895), while its sensitivity was identical to that of the US-IHC model (both 0.879). DeLong testing revealed no significant difference between the combined model and the US-IHC model (P=0.42), whereas both models outperformed the delta radiomics model (both P<0.001).

Figure 4 Performance evaluation and visualization of the predictive models. (A,B) Receiver operating characteristic curves of the three models. (C,D) Decision curve analysis. (E,F) Calibration curves. (G) Nomogram derived from the combined model for individualized prediction of major histopathological response. DCA, decision curve analysis; delta Rad-score, delta radiomics score; IHC, immunohistochemistry; MHR, major histopathological response; PR, progesterone receptor.

In the validation cohort, the AUCs of the combined model, US-IHC model, and delta radiomics model were 0.911, 0.893, and 0.670, respectively (Figure 4B). As shown in Table 5, the combined model and US-IHC model had the same accuracy (0.800). The combined model showed slightly higher specificity (0.750 vs. 0.688), whereas the US-IHC model showed higher sensitivity (0.929 vs. 0.857). DeLong analysis again demonstrated no significant difference between these two models (P=0.63), while the combined model outperformed the delta radiomics model (P=0.02).

The combined model also yielded the lowest Brier scores in both cohorts (0.092 in the training cohort and 0.143 in the validation cohort). Good concordance was observed between predicted probabilities and actual outcomes in the calibration analysis, and decision curve analysis further supported the greater clinical utility of the combined model across different threshold probabilities (Figure 4C-4F). A nomogram based on the combined model was further developed to facilitate individualized prediction of MHR (Figure 4G).

SHAP-based interpretation of the combined model

The SHAP‑based interpretation results for the prediction model are presented in Figure 5. As shown in Figure 5A, PR and shrinkage pattern were the two most influential predictors, with mean absolute SHAP values of 2.447 and 2.305, respectively. These were followed by Δmaximum diameter (1.281), enhancement area change (1.113), and delta Rad-score (0.693). The SHAP summary plot demonstrated distinct distributions and directional effects of these variables on model output across the cohort (Figure 5B).

Figure 5 SHAP interpretation of the combined model. (A) Global importance of the top five predictors. (B) Summary plot of feature effects across all patients. (C)Waterfall plot of feature contributions in a representative case. (D,E) Dependence plots for delta Rad-score and Δmaximum diameter. (F) SHAP distribution for PR. (G) SHAP distribution for enhancement area change. (H) SHAP distribution for shrinkage pattern. delta Rad-score, delta radiomics score; PR, progesterone receptor; SHAP, SHapley Additive exPlanations.

At the individual level, the SHAP waterfall plot for a representative patient illustrated how distinct features jointly determined the final prediction (Figure 5C). In this case, enhancement area change (larger → other) provided the largest positive contribution, whereas PR-positive status, NCS, a smaller reduction in maximum tumor diameter, and a higher delta Rad-score decreased the predicted probability of MHR.

For continuous variables, both delta Rad-score and Δmaximum diameter showed inverse relationships with SHAP values (Figure 5D,5E). Specifically, lower delta Rad-score and greater reduction in maximum tumor diameter were associated with higher SHAP values, indicating a greater likelihood of MHR. Among categorical variables, PR-negative status was associated with positive SHAP values, whereas PR-positive status contributed negatively to the prediction of MHR (Figure 5F). Similarly, CS showed a strong positive contribution, while NCS exerted a negative effect on model output (Figure 5H). For enhancement area change, the larger → other category had the strongest positive impact on MHR prediction, whereas larger → larger and other → larger were associated with lower or negative contributions (Figure 5G).

Discussion

In this retrospective study, we developed three models. Among them, the combined model performed best in predicting MHR, suggesting good potential for predicting MHR in clinical practice. Further analysis identified delta Rad-score, PR, Δmaximum diameter, shrinkage pattern and enhancement area change as key predictors of MHR. These variables provide complementary information on tumor biology, molecular characteristics, and treatment-related morphological changes, suggesting that these factors may help identify, at an early stage, patients who are less likely to benefit from NAC.

By comparing US characteristics between the pretreatment and mid-NAC stages, we found that, among enhancement area changes, the transition from Larger to other showed a more pronounced positive effect on MHR prediction. This association may be related to treatment-induced pathological changes, including tumor cell apoptosis, attenuation of stromal reaction, and reduced neovascularization, which together may lead to decreased microvascular perfusion and a more confined extent of enhancement within the lesion (18). Accordingly, a non-increased CEUS-enhanced area may indicate better control of viable tumor burden and a more favorable pathological response. We also found that Δmaximum diameter could effectively reflect treatment response, which is consistent with previous studies (19,20). Its pathological basis may lie in the combined effects of reduced tumor cell activity, disruption of the microvascular network, ischemic necrosis, and stromal fibrosis, all of which contribute to lesion shrinkage on imaging. Taken together, although the CEUS-enhanced area and maximum diameter may share overlapping treatment-related pathological mechanisms, they capture post-NAC tumor regression from two complementary perspectives: perfusion and morphology.

In the present study, the three selected delta radiomic features were all texture-related parameters: original_gldm_GrayLevelNonUniformity, wavelet_LLL_ngtdm_Busyness, and original_ngtdm_Strength. These features reflect treatment-induced changes in intratumoral heterogeneity from different aspects, including gray-level distribution, local texture variation, and structural complexity (21,22). A higher delta Rad-score was associated with a lower probability of achieving MHR. Among the selected features, original_gldm_GrayLevelNonUniformity and wavelet_LLL_ngtdm_Busyness were positively associated with the delta Rad-score, whereas original_ngtdm_Strength showed a negative association and had the largest absolute regression coefficient, indicating a relatively greater contribution to the model. Overall, increases in GrayLevelNonUniformity and Busyness may indicate sustained heterogeneity during treatment, which could be related to poorer response. By contrast, higher Strength may reflect more organized structural change and increased sensitivity to NAC. Together, these findings imply that failure to achieve MHR may be associated not only with initial tumor characteristics but also with treatment-related remodeling of intratumoral architecture.

Analysis of IHC markers in the primary tumor showed that PR-negative patients were more likely to achieve MHR after NAC, consistent with previous studies (23,24). This association may reflect the intrinsic biological heterogeneity of breast cancer. As a downstream marker of estrogen signaling, loss of PR often indicates not only altered hormone receptor status but also a shift toward a more proliferation-driven and biologically aggressive phenotype. However, in the setting of NAC dominated by cytotoxic agents, tumors with higher proliferative activity and lower hormone dependence may be more sensitive to chemotherapy, thereby increasing the likelihood of achieving MHR (25). In contrast, we did not observe significant associations of Ki-67 positivity or HER2 positivity with MHR, which differs from some previous reports (26,27). This discrepancy should be further examined in larger cohorts with more refined molecular subtype stratification.

In our study, CS on post-NAC US was identified as an independent predictor of MHR, which is consistent with previous MRI-based findings (11). Tumor shrinkage pattern reflects not only the morphological response to neoadjuvant therapy but also the biological sensitivity of the tumor to treatment. According to the NCCN guidelines, post-NAC tumor regression pattern plays an important role in surgical planning (28). CS is generally associated with a higher likelihood of complete tumor excision with negative margins, thereby facilitating breast-conserving surgery, whereas NCS may be associated with an increased risk of residual peritumoral disease and less favorable local control (29). Taken together, tumor shrinkage pattern represents a clinically meaningful imaging phenotype with potential value in treatment response assessment, surgical decision-making, and prognostic evaluation.

Whether US-defined shrinkage patterns are associated with underlying tumor biology, however, remains insufficiently understood. In our study, ER and PR negativity occurred at elevated frequencies among lesions displaying concentric tumor shrinkage vs. non-concentric counterparts, this molecular correlation has been consistently reported in prior MRI-based investigations (12,30). This observation supports the notion that US shrinkage phenotypes may partly reflect tumor biology. A possible explanation is that ER-negative and PR-negative tumors generally exhibit higher proliferative activity and greater chemosensitivity, which may allow a more rapid reduction in tumor burden after cytotoxic therapy and lead to a more localized decrease in lesion volume on US. In addition, relatively abundant peritumoral vascularity may facilitate drug delivery and enhance treatment response (31). Nevertheless, shrinkage pattern is unlikely to be determined by ER/PR status alone. Other factors, including HER2 status, Ki-67 level, molecular subtype, intratumoral heterogeneity, and the tumor microenvironment, may also contribute. Therefore, the association of ER negativity and PR negativity with CS on US may reflect the combined effects of tumor biology and chemosensitivity.

One of the discussion points of the study is that while there are established guidelines on the treatment of patients who did not achieve pCR (32), currently there is no recommended consensus on the treatment of patients who did not respond to NAC at mid-treatment. It remains unclear if this group of patients should be planned for earlier surgery or for a change of the chemotherapy regimen. Conversely, if there was good response noted at mid-treatment, the dilemma then arises if the patient should continue with the rest of the chemotherapy regimen. As a result, with the development of prediction models for NAC response, the clinical treatment of these groups of patients should be investigated too.

This study has certain limitations that warrant consideration. First, only a limited number of cases were included in the validation cohort, resulting in wide confidence intervals for the AUC and extreme values for some performance metrics, which may have reduced the stability of model evaluation. Second, the effects of NAC may vary across molecular subtypes; thus, large-scale stratified studies are needed to assess the applicability of the model in different subgroups. Finally, we did not directly compare our model with advanced imaging modalities such as MRI, which limited a full assessment of its relative advantages. Future prospective, multicenter investigations will be necessary to refine the model and confirm its clinical usefulness.

Conclusions

In conclusion, our study showed that a combined model integrating PR, shrinkage pattern, Δmaximum diameter, enhancement area change, and delta Rad-score, based on changes in multimodal US features between the pretreatment and mid-NAC stages, could effectively predict response to NAC in breast cancer patients at mid-treatment. More importantly, our findings suggest that combining conventional imaging assessment with artificial intelligence-driven radiomics analysis may provide a more comprehensive characterization of treatment-related tumor heterogeneity. If supported by future large-scale multicenter prospective evidence, this strategy may serve as a useful tool for individualized and adaptive neoadjuvant treatment in breast cancer.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0342/rc

Data Sharing Statement: Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0342/dss

Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0342/prf

Funding: This work was supported by the Medical Innovation Project of Fujian Provincial Health Commission (No. 2022CXA055) and the Natural Science Foundation of Fujian Province (No. 2023J011813).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0342/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Zhangzhou Affiliated Hospital of Fujian Medical University (approval No. 2025KYZ361) and individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Ma Y, Zhu Y, Zhu J, et al. Ultrasound and immunohistochemical predictors of neoadjuvant chemotherapy response in breast cancer. Transl Cancer Res 2025;14:6289-99. [Crossref] [PubMed]
2. Xu M, Shan D, Chen X, et al. Pretreatment Amide Proton Transfer-Weighted Imaging Histogram Analysis Combined With ER-Negative and HER2-Positive Expression Predicts Pathologic Complete Response After Neoadjuvant Chemotherapy in Breast Cancer. Cancer Med 2025;14:e71420. [Crossref] [PubMed]
3. Wang M, Huang Z, Tian H, et al. Longitudinal Ultrasound Delta Radiomics for Early Stratified Prediction of Tumor Response to Neoadjuvant Chemotherapy in Breast Cancer. Acad Radiol 2025;32:7119-33. [Crossref] [PubMed]
4. Lim LY, Miao H, Lim JS, et al. Outcome after neoadjuvant chemotherapy in Asian breast cancer patients. Cancer Med 2017;6:173-85. [Crossref] [PubMed]
5. Mao N, Dai Y, Zhou H, et al. A multimodal and fully automated system for prediction of pathological complete response to neoadjuvant chemotherapy in breast cancer. Sci Adv 2025;11:eadr1576.
6. Wang S, Liu J, Song L, et al. Integrative multimodal ultrasound and radiomics for early prediction of neoadjuvant therapy response in breast cancer: a clinical study. BMC Cancer 2025;25:1156. [Crossref] [PubMed]
7. Hao T, Ji X, Zhao Q, et al. Ultrasound and Clinicopathological Features-Based Machine Learning Model for Predicting Neoadjuvant Therapy Efficacy in Breast Cancer. Cancer Rep (Hoboken) 2026;9:e70600. [Crossref] [PubMed]
8. Alshuhri MS, Al-Mubarak HF, Qaisi A, et al. MRI Delta Radiomics to Track Early Changes in Tumor Following Radiation: Application in Glioblastoma Mouse Model. Biomedicines 2025;13:815. [Crossref] [PubMed]
9. Feng X, Shi Y, Wu M, et al. Predicting the efficacy of neoadjuvant chemotherapy in breast cancer patients based on ultrasound longitudinal temporal depth network fusion model. Breast Cancer Res 2025;27:30. [Crossref] [PubMed]
10. Zou J, Zhang L, Chen Y, et al. Neoadjuvant Chemotherapy and Neoadjuvant Chemotherapy With Immunotherapy Result in Different Tumor Shrinkage Patterns in Triple-Negative Breast Cancer. J Breast Cancer 2024;27:27-36. [Crossref] [PubMed]
11. Yi C, Li L, Ma J. Tumor enhancement and shrinkage pattern in dynamic contrast-enhanced magnetic resonance imaging for predicting pathologic complete response after human epidermal growth factor receptor 2 (HER2)-targeted therapy in breast cancer. Quant Imaging Med Surg 2024;14:6734-44. [Crossref] [PubMed]
12. Wang J, Liu Y, Liu T, et al. Tumor shrinkage patterns and optimal timing of response assessment during neoadjuvant therapy for breast cancer: a study based on multiparametric MRI. Cancer Imaging 2025;26:2. [Crossref] [PubMed]
13. Wang L, Wang Y, Yang L, et al. Using baseline MRI radiomics to predict the tumor shrinkage patterns in HR-Positive, HER2-Negative Breast Cancer. Front Oncol 2025;15:1539644. [Crossref] [PubMed]
14. Bozer A, Altın L, Eren Güzel H. Predictive Value of Dimensional and Functional MRI Parameters on Mid-Treatment MRI for Pathologic Complete Response in Breast Cancer. Eur J Breast Health 2025;22:44-53. [Crossref] [PubMed]
15. Wang XY, Huang JX, Liu FT, et al. Predicting Breast Cancer Pathologic Complete Response after Neoadjuvant Chemotherapy Using Bimodal US and MRI. Radiol Imaging Cancer 2025;7:e240493. [Crossref] [PubMed]
16. Lin R, Chen Y, Wang Y, et al. Integrated conventional ultrasound and radiomics model for predicting major pathological response to neoadjuvant therapy in triple-negative breast cancer. BMC Med Imaging 2025;25:472. [Crossref] [PubMed]
17. Chen L, Su Y, Wang Y, et al. Ultrasound Radiomics-Based Machine Learning and SHapley Additive exPlanations Method Predicting Pathological Prognostic Stage in Breast Cancer: A Bicentric and Validation Study. Clin Breast Cancer 2025;25:e954-e967.e6.
18. Xie Y, Chen Y, Wang Q, et al. Early Prediction of Response to Neoadjuvant Chemotherapy Using Quantitative Parameters on Automated Breast Ultrasound Combined with Contrast-Enhanced Ultrasound in Breast Cancer. Ultrasound Med Biol 2023;49:1638-46. [Crossref] [PubMed]
19. Liu YS, Li ZJ, Wang H, et al. Clinical value of ultrasound combined with the nutritional risk index in predicting the efficacy of neoadjuvant chemotherapy for breast cancer: An observational study. Medicine (Baltimore) 2026;105:e47070. [Crossref] [PubMed]
20. Wen X, Chen J, Zhong J, et al. Model based on ultrasound and clinicopathological characteristics for early prediction of pathological complete response to neoadjuvant chemotherapy in breast cancer. Quant Imaging Med Surg 2024;14:8840-51. [Crossref] [PubMed]
21. Chitalia RD, Kontos D. Role of texture analysis in breast MRI as a cancer biomarker: A review. J Magn Reson Imaging 2019;49:927-38. [Crossref] [PubMed]
22. Zwanenburg A, Vallières M, Abdalah MA, et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology 2020;295:328-38. [Crossref] [PubMed]
23. Park GE, Lee S, Min EJ, et al. Predicting Pathologic Complete Response Using Breast MRI After Completion of Neoadjuvant Chemotherapy in Breast Cancer Patients With Residual Calcifications on Mammography. J Korean Med Sci 2025;40:e248. [Crossref] [PubMed]
24. Wang Y, Jiang X, Jiao Y, et al. The utilization of CEUS and SWE for predicting pathological complete response to neoadjuvant chemotherapy for invasive breast cancer. Sci Rep 2026;16:7434. [Crossref] [PubMed]
25. Li Z, Wei H, Li S, et al. The Role of Progesterone Receptors in Breast Cancer. Drug Des Devel Ther 2022;16:305-14. [Crossref] [PubMed]
26. Zhang Y, Qiu S, Wang X. Factors Influencing Pathological Complete Response After Neoadjuvant Chemotherapy in Breast Cancer: A Single-Center Retrospective Study Focusing on ER and HER-2 Status. Int J Breast Cancer 2026;2026:6694023. [Crossref] [PubMed]
27. Zhang Y, Yuan MQ, Jia TL, et al. Utilization of ultrasound combined with Ki-67 and CESM quantitative parameters to predict pathological complete response following neoadjuvant chemotherapy in triple-negative breast cancer. Transl Oncol 2026;63:102589. [Crossref] [PubMed]
28. Gradishar WJ, Moran MS, Abraham J, et al. Breast Cancer, Version 3.2022, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 2022;20:691-722. [Crossref] [PubMed]
29. Huang Y, Song X, Chen Y, et al. Intratumoral Microbiome-related MRI Model for Predicting Breast Cancer Shrinkage Pattern Following Neoadjuvant Therapy. Radiology 2025;316:e243545. [Crossref] [PubMed]
30. Bozer A, Yilmaz C, Çetin Tunçez H, et al. Correlation of Histopathological and Radiological Response Patterns and Their Prognostic Implications in Breast Cancer After Neoadjuvant Chemotherapy. Breast Cancer (Dove Med Press) 2024;16:1005-17. [Crossref] [PubMed]
31. Ling DC, Sutera PA, Iarrobino NA, et al. Is Multifocal Regression a Risk Factor for Ipsilateral Breast Tumor Recurrence in the Modern Era After Neoadjuvant Chemotherapy and Breast Conservation Therapy? Int J Radiat Oncol Biol Phys 2019;104:869-76. [Crossref] [PubMed]
32. Bossuyt V, Spring L. Pathologic evaluation of response to neoadjuvant therapy drives treatment changes and improves long-term outcomes for breast cancer patients. Breast J 2020;26:1189-98. [Crossref] [PubMed]