Neoadjuvant Chemotherapy Response Research Articles

Abstract A051: Multi-omic explainable machine learning improves cancer treatment outcome prediction

Abstract Background: Advancements in multi-omics data integration and explainable Machine Learning (ML) have shown promise in precision oncology. Multi-omic data used to train ML models may include genomics, transcriptomics and histopathology to characterize cancer cells and the tumor microenvironment (TME). Explainability methods, such as SHAP, have enabled researchers and clinicians to unravel the decision-making rationale of ML models predicting cancer progression and treatment response. We developed an explainable ML framework that incorporates multi-omic features of cancer and the TME. This framework was applied to predict patient response to neoadjuvant chemotherapy (NAC) in breast cancer and immune checkpoint inhibitor (ICI) in melanoma. Methods: For breast cancer, we used the cohort from Sammut et al. [1] (n=157 training, n=75 test). For melanoma, we assembled a cohort comprising 229 patients (n=138 training, n=53 test cutaneous, n=38 test non-cutaneous) from five independent studies. We improved the performance of the ensemble ML models in Sammut et al. [1] by implementing a shared-learning architecture to enable component models to influence each other as training progresses. We applied this ensemble (Ens:LR+RF+SVM) to predict NAC response in breast cancer and ICI response in melanoma by integrating clinical, DNA sequencing, RNA sequencing and histopathology (only for breast cancer) data. For melanoma, we also trained three single ML models (LR, RF, and SVM) and another ensemble (Ens:LR+RF), and introduced a novel dual utility of SHAP for feature-selection during training and biomarker threshold identification during validation. Results: The ensemble model trained on the multi-omic breast cancer features achieved ROC-AUC of 0.88 and showed a potential 25% reduction in false positives (i.e., incorrect predictions of good response) compared to its predecessor from Sammut et al. [1]. In the melanoma cohort, the Ens:LR+RF model achieved ROC-AOC of 0.77 but was outperformed by the RF model, ROC-AUC 0.78. SHAP revealed unique interactions between each ML model and the feature space, resulting in distinct training feature sets per model. During validation, the intersection between feature values and SHAP scores revealed numerical thresholds underpinning good versus poor responses of clinically meaningful biomarkers such as neoantigen load (>2.25 good, <2.25 poor, values in log10 scale). Across these two studies, we developed and open-sourced a scalable and versatile ML workflow (xML-workFLow) for rapid experimentation in biomedical research. Conclusions: This work showcases the potential of multi-omics explainable ML in advancing precision oncology to improve treatment outcome prediction. With further experimental validation, the use of explainable ML to determine numerical thresholds could guide the development of companion diagnostics and inform combination therapeutic strategies.

Prediction of Early Neoadjuvant Chemotherapy Response of Breast Cancer through Deep Learning-based Pharmacokinetic Quantification of DCE MRI.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To improve the generalizability of pathologic complete response (pCR) prediction following neoadjuvant chemotherapy using deep learning (DL)-based retrospective pharmacokinetic quantification (RoQ) of early-treatment dynamic contrast-enhanced (DCE) MRI. Materials and Methods This multicenter retrospective study included breast MRI data from four publicly available datasets of patients with breast cancer acquired from May 2002 to November 2016. RoQ was performed using a previously developed DL model for clinical multiphasic DCE-MRI datasets. Radiomic analysis was performed on RoQ maps and conventional enhancement maps. These data, together with clinicopathologic variables and shape-based radiomic analysis, were subsequently applied in pCR prediction using logistic regression. Prediction performance was evaluated by area under the receiver operating characteristic curve (AUC). Results A total of 1073 female patients with breast cancer were included. The proposed method showed improved consistency and generalizability compared with the reference method, achieving higher AUCs across external datasets (0.82 [CI: 0.72-0.91], 0.75 [CI: 0.71-0.79], and 0.77 [CI: 0.66-0.86] for Datasets A2, B, and C, respectively). On Dataset A2 (from the same study as the training dataset), there was no significant difference in performance between the proposed method and reference method (P = .80). Notably, on the combined external datasets, the proposed method significantly outperformed the reference method (AUC: 0.75 [CI: 0.72- 0.79] vs 0.71 [CI: 0.68-0.76], P = .003). Conclusion This work offers a novel approach to improve the generalizability and predictive accuracy of pCR response in breast cancer across diverse datasets, achieving higher and more consistent AUC scores than existing methods. ©RSNA, 2025.

Single-cell transcriptomic profiling of immune landscape in triple-negative breast cancer during neoadjuvant chemotherapy

Triple-negative breast cancer (TNBC) is the most aggressive subtype, typically requiring neoadjuvant chemotherapy (NAC) as an obligatory component of the treatment regimen. Achieving a pathological complete response to NAC is associated with improved long-term outcomes for patients with TNBC. The functional status of the immune system plays a critical role in NAC efficacy. Herein, we presented the investigation of systemic and local immune landscape during the initial course of NAC treatment and identify factors that contribute to chemotherapy resistance of TNBC. Using single-cell RNA sequencing, we demonstrated that the transcriptional profile remained stable in a patient who responded to NAC, while a non-responder exhibited significant dysregulation in the expression of genes involved in stress response, apoptosis, immune cell proliferation, and differentiation within lymphocyte and monocyte populations. During the first course of NAC, circulating cytotoxic CD8 T cells in the non-responder patient overexpressed granzymes B and H, granulysin, and perforin. In contrast, expression of these factors decreased in CD8 T cells within the tumor. Finally, we identified for a first time a signature of myeloid-derived suppressor cells (MDSC) within the S100АhighMHClow monocyte population and calculated an MDSC score for both the responder and the non-responder TNBC patients. An elevated MDSC score in the non-responder was validated using data from an independent cohort of patients with poor NAC response. Our data underscores the importance of immune system functionality in determining chemotherapy efficacy in TNBC.

Medical semantic knowledge-integrated multitask learning network for report generation and neoadjuvant chemotherapy response prediction.

Automatically generated reports and predictions of neoadjuvant chemotherapy (NAC) responses can support physicians in clinical decision-making. However, previous research has not thoroughly considered the connection between the detailed descriptions within these reports and their impact on predicting NAC responses. This study aims to develop an effective medical semantic knowledge-integrated multitask learning (MSKMTL) network to enhance the accuracy of NAC response predictions and improve the quality of automatically generated medical reports. We propose a MSKMTL network designed to increase report generation and NAC prediction by leveraging semantic knowledge within medical reports. The knowledge base was trained with DCE-MRI and the corresponding reports to capture the interrelationships of paired visual-text representations. During the training stage, semantic knowledge is encoded via a pretrained BERT model and stored in the knowledge base. We use contrastive learning to alignthe feature space of the image with the related report. In the inference stage, MSKMTL retrieves the stored semantic knowledge by querying image features to facilitate the generation of an impression section of the report and predict the NAC response. We evaluated the performance of the proposed network using metrics associated with both report generation and prediction tasks on a breast DCE-MRI dataset and a public IU x-ray dataset. For report generation, the BLEU-1, BLEU-4, ROUGE, and METEOR scores are 0.408, 0.186, 0.426, and 0.402, respectively. For NAC prediction, the area under the curve, recall, precision, and F1 score are 0.849, 0.750, 0.795, and 0.794, respectively. The network also showed strong performance with the IU x-ray dataset. Our study shows that, compared to other methodologies, integrating medical semantic knowledge increases prediction accuracy and generates superior reports.

Radiomics-based machine learning in prediction of response to neoadjuvant chemotherapy in osteosarcoma: A systematic review and meta-analysis.

Radiomics-based machine learning in prediction of response to neoadjuvant chemotherapy in osteosarcoma: A systematic review and meta-analysis.

Habitat-Based Radiomics for Revealing Tumor Heterogeneity and Predicting Residual Cancer Burden Classification in Breast Cancer.

Habitat-Based Radiomics for Revealing Tumor Heterogeneity and Predicting Residual Cancer Burden Classification in Breast Cancer.

Using [18F]FDG PET/CT to Identify Optimal Responders to Neoadjuvant Therapy in Breast Cancer—Results from a Prospective Patient Cohort

Background/objectives: Pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in breast cancer (BC) is a solid indicator of favourable prognosis, potentially also being useful for more conservative patient management. We aim to explore the potential of [18F]FDG PET/CT as a non-invasive method to predict response to NAC. Methods: In this prospective, observational cohort study, we enrolled BC patient candidates for NAC who underwent baseline and preoperative [18F]FDG PET/CT. NAC response was determined using final histopathology. PET images were assessed qualitatively and semi-quantitatively, and the findings correlated with NAC response. Results: In total, 133 BC patients were included. The visual analysis of preoperative PET/CT detected residual disease (RD) with high specificity (>93%) and moderate sensitivity, based on pCR/RD classification and RCB index. Semiquantitative measures (SUVmax, TBR) were significantly higher in non-responders across the classification methods (p < 0.001 for all). Conclusions: These findings highlight the potential of preoperative [18F]FDG PET/CT as a complementary tool for identifying excellent responders to NAC across BC subtypes or response criteria. This could inform personalised treatment and potentially allow for surgery to be omitted in selected patients.

Application of N-NOSE for Evaluating the Response to Neoadjuvant Chemotherapy in Breast Cancer Patients.

Background: Breast cancer remains a leading cause of cancer-related deaths despite advances in its diagnosis and treatment. Accurate evaluation of the response to neoadjuvant chemotherapy (NAC), especially in HER2-positive and triple-negative subtypes, is critical. The current methods, including imaging and liquid biopsies, have limitations. N-NOSE, a novel urine-based cancer screening test using Caenorhabditis elegans (C. elegans) chemotaxis, offers a non-invasive alternative. This study investigates the potential of N-NOSE to predict the NAC response in breast cancer patients for improved treatment evaluations. Materials and Methods: This prospective study enrolled 36 breast cancer patients undergoing NAC and surgery to assess the predictive power of the N-NOSE method using urine samples. A chemotaxis analysis of C. elegans was used to calculate the index reduction scores (IRS1-3), reflecting the changes in tumor-related odorants across the treatment stages. Results: Between August 2020 and May 2023, 36 breast cancer patients were enrolled to evaluate the predictive value of N-NOSE IRSs for NAC response. A pathological complete response (pCR) was achieved in 36.1% of the patients. Among the three IRS types analyzed in the 35 patients, IRS3, which showed the IRS at pre-treatment minus that after surgery, showed the highest predictive performance for a pCR, with an AUC of 0.75, indicating its potential utility as a non-invasive biomarker for treatment response evaluations. Conclusions: Index reduction scores evaluated using the N-NOSE method may reflect the efficacy of NAC in breast cancer patients. Future large-scale and multi-institutional prospective studies are warranted.

Expression of CD8+ T Lymphocyte and Foxp3+ T Lymphocyte as Risk Factor for Response to Neoadjuvant Chemotherapy in Stage III Breast Cancer

BACKGROUND: Neoadjuvant chemotherapy (NAC) in breast cancer is usually utilized to eradicate micro-metastasis, induce apoptosis in tumor cells, and reduce the primary tumor size, enabling surgical intervention. Recent studies have shown that tumor-infiltrating lymphocytes (TILs), especially cytotoxic CD8+ T cells and immunosuppressive Foxp3+ regulatory T cells, influence tumor response to treatment. However, their role as predictive markers for NAC response remains unclear. Therefore, this study was performed to investigate whether high expression of CD8+ and low expression of Foxp3+ T lymphocytes are associated with better response to NAC in stage III breast cancer patients.METHODS: Total of 60 biopsy samples from stage III breast cancer patients were included, comprising 30 subjects in the response group (+) and 30 subjects in the non-response group (−). The expression levels of CD8+ and Foxp3+ T lymphocytes in tumor tissue were assessed semi-quantitatively by immunohistochemistry (IHC), using a scoring system based on the proportion and intensity of positively stained cells (Black’s grading criteria).RESULTS: Stage III breast cancer with high expression of CD8+ T lymphocytes was significantly associated with a better response to NAC (p=0.004; OR=6.882). Meanwhile, low expression of Foxp3+ T lymphocytes was not significantly associated with chemotherapy response (p=0.067; OR=3.250). A higher tumor grade was also associated with an improved response to treatment. The probability of achieving a positive response to NAC in subjects presenting with high CD8+ expression, low Foxp3+ expression, and high tumor grade was estimated at 96.98%.CONCLUSION: The combination of high expression of CD8+ T lymphocyte, low expression of Foxp3+ T lymphocyte and high tumor grade might be useful to predict good response to NAC in stage III breast cancer.KEYWORDS: CD8+ T lymphocyte, Foxp3+ T lymphocyte, neoadjuvant chemotherapy, breast cancer

Abstract P1-02-19: Spatially defining T-cell clonal evolution in post-neoadjuvant breast cancer and association with molecular residual disease

Abstract Background: Tumor-infiltrating lymphocyte (TIL) enrichment in primary breast cancers is predictive of treatment response, and specific CD4 and CD8 subsets are associated with neoadjuvant chemotherapy (NAC) and immunotherapy response. Spatially defined T-cell receptor (TCR) evolution patterns in resection tissues associated with NAC response and survival have not been described. Methods: To define such patterns, we applied high-throughput TCR profiling to pre- and post-NAC tissues and associated immune dynamics with tumor evolution and clinical outcomes. We generated TCR repertoires of 29 primary breast cancers and 768 spatially catalogued, post-NAC specimens from matched cases (median 24 post-treatment samples/case) using Sequence Affinity capture &amp; analysis By Enumeration of cell-free Receptors (SABER) applied to tissue-based Cancer Personalized Profiling by Deep Sequencing (t-CAPP-Seq). To profile each primary tumor’s genomic and immunologic response to NAC, we designed personalized oligonucleotide panels using pretreatment whole exome sequencing and a fixed panel of recurrently mutated genes to capture emergent alterations, as well as a comprehensive panel of TCR-β regions. Leveraging information from fragmented TCRs in FFPE tissues, we spatially enumerated TCR clonotype count and composition in each post-NAC sample and compared the relative abundance of clones shared with each pretreatment tumor. For a subset of cases, we enumerated TILs and tumor cellularity in post-NAC samples (N = 119). To interrogate clonotype composition, we computed Jaccard similarity indices between each pre- and post-NAC sample pair. We associated post-NAC clonal abundance and clonotype composition with tMRD burden and genomic composition. Finally, we associated TCR repertoire dynamics with quantitative tMRD assessment and survival. Results: TIL density was strongly correlated with the presence of post-NAC residual disease (R=0.94, p&amp;lt;0.001 for histologic invasive carcinoma cellularity; R=0.6, p&amp;lt;0.001 for mean tMRD allele frequency (AF)). Because TILs are defined in the presence of histologic residual disease, we defined tMRD-TILs by TCRs sequenced in tMRD positive tissues, which demonstrated greater repertoire diversity than non-tMRD TCRs (greater ShannonE and inverse Simpson diversity indices, p=.007 and p=1.3e-5, respectively), suggesting a spatially defined tumor neoantigen response. Within cases with low tMRD burden (mean AF ≤3%), which experienced longer progression-free survival than those with high tMRD burden (p=0.034), we observed greater TCR clone count in the post-NAC tissue (p=.012), greater shared clone count with the pretreatment tumor (p=5.7e-8), and greater similarity of the post-NAC clonotypes to the pretreatment tumor (p=5.1e-6, Wilcoxon test of Jaccard indices). Presence of a pretreatment CSPP1 lesion was associated with decreased TCR repertoire diversity in chemotherapy-treated resection tissues (ShannonE index, p=0.00056), and CSPP1-mutated tMRD was associated with inferior PFS (HR 23.45, p=.0007) Conclusions: By spatially defining changes in the genomic and immunologic microenvironment following NAC, we found superior molecular response is correlated with preservation of the clonal relationship between tMRD-TILs and the primary tumor with simultaneous generation of a rich TCR repertoire. Ongoing investigations involve computationally predicting neoantigen targets of clonally dominant tMRD-TILs through NAC for in vivo validation, representing an avenue for refining selection of molecularly targeted adjuvant therapies. Citation Format: Julia Ransohoff, Mia A. Carleton, Sofia Miron Barroso, Gregory Bean, Alisha Maltos, James Ford, George Duran, Michael Khodadoust, Melinda L. Telli, Ash A. Alizadeh, David M. Kurtz. Spatially defining T-cell clonal evolution in post-neoadjuvant breast cancer and association with molecular residual disease [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P1-02-19.

Abstract P2-10-07: Practice patterns and survival analysis of early-stage HER-negative breast cancers with low and intermediate levels of hormone receptor expression: a 2018-2020 US National Cancer Database analysis

Abstract Background: Breast cancer (BC) with low hormone receptor (HR) expression (1-10%) is classified as HR-Low BC. We previously demonstrated that pathologic complete response (pCR) rates and biologic features of HR-Intermediate HER2-negative BC (11-30%), much like HR-Low BC, are similar to triple negative breast cancer (TNBC). Here, we report treatment patterns and overall survival (OS) for HER2-negative BC by level of HR expression from the US National Cancer Database (NCDB). Methods: Patients (pts) with stage I-III HER2-negative BC diagnosed in 2018-2020 were categorized into four groups by estrogen receptor (ER) and progesterone receptor (PR) expression: ER&amp;lt;1% and PR&amp;lt;1% (HR-Neg), ER 1-10% and/or PR 1-10% (HR-Low), ER &amp;gt;11-30% and/or PR&amp;gt;11-30% (HR-Int), ER &amp;gt;30% and/or PR &amp;gt;30% (HR-High). Pts with undocumented HR status or without curative intent surgery were excluded. The primary outcome was OS by HR expression. Key secondary outcomes were treatment patterns and OS by neoadjuvant chemotherapy (NAC) response and endocrine therapy (ET) use. Categorical variables were compared between the groups using a Chi-square test. OS survival was explored using Kaplan Meier estimates, log-rank tests and univariate/multivariable cox proportional hazard models. Results: 395,757 incident cases of early-stage HER2-negative breast cancer were identified, including 8857 (2.2%) HR-Low and 4375 (1.1%) HR-Int BC. HR-Low and HR-Int cohorts reported more advanced stage at diagnosis and were more likely to be node-positive with higher Ki-67 scores than HR-High BC (p&amp;lt;0.001 for all). HR-Low and HR-Int groups consisted of proportionally more Black patients than HR-High (20%, 19% vs 9%; p&amp;lt; 0.001). Oncotype DX results were more likely to be obtained with increasing levels of HR expression (HR-Neg 1%, HR-Low 11%, HR-Int 25%, HR-High 43%; p&amp;lt; 0.001). ET use was more frequent in HR-High than HR-Low or HR-Int BC (87%, 46% &amp; 71%; p &amp;lt;0.001). The use of immunotherapy increased with lower expression of HR (HR-Neg 5%, HR-Low 4%, HR-Int 3%, HR-High 1%; p&amp;lt; 0.001). Additionally, NAC use was more frequent in HR-Low and HR-Int than HR-High (31%, 24% &amp; 5%, p&amp;lt; 0.001). pCR rates were correlated with significantly higher OS at the 3 year (y) timepoint for all levels of HR expression (p&amp;lt; 0.001). NAC response had a significantly greater impact on 3y OS in the HR-Low (pCR 96.5%, no pCR [NR] 62.2%) and HR-Int (pCR 94.5%, NR 56.8%) groups compared to HR-High (pCR 96.7%, NR 91.5%). Multivariable analyses showed significantly worse OS in HR-Neg (hazard ratio [HazR]: 2.73), HR-Low (HazR: 2.70), and HR-Int (HazR: 2.64) compared to HR-High tumors after adjusting for age, tumor size, nodal status and grade (p&amp;lt; 0.001 for each). 5y OS in HR-Low, HR-Int, and HR-High groups were 78.5%, 81.4%, and 89.0% respectively. Among HR-positive BC, ET use was associated with longer OS in multivariable analyses (HazR: 0.85) with 5y OS for HR-Low (ET 82.3%, no ET 74.7%), HR-Int (ET 84.4%, no ET 73.1%), and HR-High (ET 90.0%, no ET 80.7%) groups (p&amp;lt; 0.001 for each). Conclusions: HR-Low and HR-Int HER2-negative BC are rare subtypes with distinct biologic features compared to HR-High BC. This real-world analysis reveals significant differences in the survival outcomes, management, and treatment responses of HR-Low and HR-Int BCs. ET use is less common with lower HR expression yet associated with improved 5y OS. The use of immunotherapy is more frequent with lower HR expression; however, requires further investigation in HR-Low and HR-Int BC. The binary categorization of HR does not adequately address disease heterogeneity. This may contribute to outcome disparities in patients with HR-Low and HR-Int, HER2-negative BC. Citation Format: Dionisia Quiroga, Julie A Stephens, Gilbert Bader, Mathew A Cherian, Ashley P Davenport, Margaret E Gatti-Mays, Kai CC Johnson, Daniel Stover, Robert Wesolowski, Nicole Williams, Nerea Lopetegui-Lia, Arya M Roy, Samilia Obeng-Gyasi, Bridget A Oppong, Sachin R Jhawar, Sagar Sardesai. Practice patterns and survival analysis of early-stage HER-negative breast cancers with low and intermediate levels of hormone receptor expression: a 2018-2020 US National Cancer Database analysis [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P2-10-07.

Abstract P3-12-03: Optimizing Treatment Sequence for Inoperable LABC: Long-Term Outcomes of Surgery First vs. Neoadjuvant Chemotherapy in a Real-World Setting

Abstract Background: Breast cancer is the most common cancer in Chinese women. Locally advanced breast cancer (LABC) is a subtype with a high incidence and poses a significant treatment challenge. Traditionally, inoperable LABC patients receive neoadjuvant chemotherapy (NAC) followed by surgery. However, not all patients benefit from NAC, raising questions about the need for a more precise definition of inoperability and the potential benefits of upfront surgery. Objectives: This retrospective cohort study aimed to assess the feasibility and efficacy of upfront surgery compared to NAC as the initial treatment approach for inoperable LABC patients. The study also evaluated the impact of various factors on disease-free survival (DFS) and overall survival (OS). Methods: The study included 243 patients diagnosed with inoperable LABC (excluding T3N1M0) at PUMCH between January 2017 and December 2018. Patients were divided into two groups: initial surgery (n=187) and NAC (n=56). Propensity score matching (PSM) was used to minimize selection bias. The primary endpoint was investigator-assessed DFS and OS. Secondary endpoints included confirmed objective response rate and subgroup analyses based on age, tumor stage, lymph node stage, molecular subtypes, and treatment response. Survival was estimated using the Kaplan-Meier method with log-rank tests for comparisons. Cox proportional hazards models were used for univariate analyses in different subgroups (Figure 1). Results: The median follow-up was 60.9 months. The 5-year OS was 89.6% for the upfront surgery group and 81.9% for the NAC group (p=0.12). Similarly, the 5-year DFS was 73.0% and 67.1%, respectively (p=0.24). No significant difference in survival outcomes was observed between the two groups overall (Figure 2). Subgroup analyses revealed that upfront surgery resulted in significantly better OS for patients younger than 60 years old (HR=0.32; 95% CI, 0.10 to 0.96; P=0.0429) and those with stage IIIA disease (HR=0.22; CI, 0.06 –0.86; P = 0.03). There was a trend towards better OS for patients with tumors smaller than 5cm receiving upfront surgery (HR=0.37; 95% CI 0.13 –1.03; p=0.056). NAC response also played a role, with patients experiencing progressive disease (PD) or stable disease (SD) after NAC having a significantly worse DFS (HR, 0.27; 95% CI, 0.09 to 0.79; P = 0.017) and OS (HR, 0.09; 95% CI, 0.02 to 0.48; P =0.004) compared to those who responded well to NAC (Figure 3). Conclusion: This study suggests that upfront surgery may be a viable and potentially beneficial treatment option for a subset of inoperable LABC patients, particularly those younger than 60 years old, with stage IIIA disease, or smaller tumors. Additionally, the response to NAC can be a valuable factor in guiding treatment decisions. These findings highlight the need for a more nuanced approach to LABC treatment, considering patient characteristics and NAC response. Further research is warranted to refine the definition of inoperability and develop personalized treatment strategies for LABC patients. Funding:To be added.This study was conducted according to ICH-GCP guidelines and was approved by the Ethics Committee of PUMCH. This study is registered with ClinicalTrials.gov. Figure 1. The flowchart of this study. Figure 2. The survival analyses of the upfront surgery group and neo-adjuvant chemotherapy group. Figure 3. The forest plot of the subgroup analysis of this study. Citation Format: Yu Song, Bowen Liu, Yan Lin, Ying Xu, Xin Huang, Yidong Zhou, Qiang Sun. Optimizing Treatment Sequence for Inoperable LABC: Long-Term Outcomes of Surgery First vs. Neoadjuvant Chemotherapy in a Real-World Setting [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P3-12-03.

Abstract P1-04-28: Comparison of tumor immune microenvironment of TNBC with different treatment outcome with neoadjuvant chemotherapy: Spatial analysis using multiplex fluorescent immunohistochemistry

Abstract Background: Triple-negative breast cancer (TNBC) is known for its highly aggressive nature and poor prognosis. However, it has been noted to exhibit high immunogenicity compared to other breast cancer types, making it more responsive to immunotherapy. Recently, neoadjuvant chemotherapy (NAC) has become the standard treatment for early high-risk TNBC. Higher stromal tumor-infiltrating lymphocytes (TIL) predict increased pathologic complete response (pCR) and superior clinical outcomes in TNBC treated with NAC. Despite this general trend, high TIL does not always predict a favorable response, nor does low TIL always predict a poor response. This suggests that individuals with varying treatment outcomes might have different immune dynamics, even under similar TIL levels. Objective: This study aimed to examine the tumor immune microenvironment (TIME) of pretreated TNBC samples that subsequently underwent NAC using multiplex immunofluorescence (mIHC). Our investigation explored the role of TIME in NAC response and elucidated differences in TIME among patients with similar TIL levels but varying treatment outcomes, as well as those with differing TIL levels but similar treatment responses. Methods: We retrospectively collected medical records of 16 patients with stage II-III early TNBC treated with NAC at Gangnam Severance Hospital, Yonsei University, between January 2019 and August 2021. Two pathologists (SSJ and YJC) evaluated stromal TIL levels, tumor-stroma ratio, and PD-L1 (22C3) status (CPS). mIHC, including CK, CD4, CD8, CD20, and FOXP3, was performed for tumor cells (TC) and immune cells (IC). Spatial metrics, including the cellular densities of TC and IC, and cell-cell distances between IC, were analyzed. Results: A total of 1,183,644 cells (761,976 TC and 421,668 IC) from sixteen patients (six pCR [4 high- and 2 low TIL] and ten non-pCR [4 high- and 6 low TIL]) were analyzed. There were no significant differences in baseline characteristics or IC densities between the pCR and non-pCR groups. In the pCR group, CD20+IC and CD8+IC were predominant IC subtypes, particularly in high TIL cases, whereas CD4+IC was the dominant fraction in the non-pCR group. Density plots of different IC types showed dense clustering of CD8+IC and CD20+IC near the TC in the pCR group. Regarding IC-IC interaction, the CD4 IC pair was the largest fraction overall. Within the pCR group, high TIL cases had a high proportion of CD20 IC pairs, whereas low TIL cases had IC pairs, including CD8+IC. Conclusion: This study observed differences in IC composition and their distribution in pretreated TNBC samples based on pCR status and TIL levels. The findings suggest that tumoral CD8+IC and CD20+IC play crucial roles in determining the response to NAC in eTNBC. Particularly, significant infiltration of CD20+IC in high TIL conditions suggests the presence of B-T cell interaction. Citation Format: Jee Hung Kim, Inho Park, Su-Jin Shin, Heounjeong Go, Jiwon KO, Yangkyu Lee, Soong June Bae, Sung Gwe Ahn, Joon Jeong, Yoon Jin Cha. Comparison of tumor immune microenvironment of TNBC with different treatment outcome with neoadjuvant chemotherapy: Spatial analysis using multiplex fluorescent immunohistochemistry [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P1-04-28.

Abstract P1-05-23: An observational cohort study of Multiplex8+: a spatially informed assay that uses multiplexed RNA-FISH guided laser capture microdissection followed by total RNA-sequencing

Abstract We previously developed and validated a breast cancer (BCa) diagnostic test called mFISHseq on a retrospective cohort of 1,082 FFPE breast tissues. The test utilizes a multiplexed RNA-FISH panel consisting of estrogen (ESR1), progesterone (PGR), and HER2 (ERBB2) receptors and Ki67 (MKI67) to characterize regions of interest that are subsequently captured by laser capture microdissection for total RNA-SEQ. Here, we report the results from 53 patients that underwent a research use only (RUO) version of the test called Multiplex8+, which provides information about the four main BCa biomarkers, intrinsic molecular and TNBC subtypes, prognostic risk, and expression of 40 genes and 28 gene signatures that inform about cancer hallmark pathways and treatment response to chemotherapies as well as targeted therapies (e.g., immunotherapy, CDK4/6 inhibitors, and antibody drug-conugates (ADCs)). Three patients had two specimens (primary and metastatic tumor specimens) processed yielding a total of 56 patient specimens, which consisted of hormone receptor positive (HR+)/HER2− (n=11), HR+/HER2+ (n=9), HR−/HER2+ (n=4), and HR−/HER2− (n=30) as determined by IHC. Two patients had unavialable IHC data. The median age at surgery was 46 yrs and the patients comprised various node statuses (negative=17, positive=18, unavailable=21), tumor sizes (pT1=22, pT2-4=17, unavailable=17), and grades (G1=1, G2=16, G3=30, unavailable=9). Multiplex8+ showed high agreement with 90.5% of IHC reported results. Agreement was highest for HER2-ERBB2 (96.4%) followed by PR-PGR (92.7%), ER-ESR1 (85.5%), and Ki67-MKI67 (85.3%). Most discordant results for hormone receptors were in gray zone areas for IHC (i.e., ER/PR low ≤10%) and Multiplex8+ classified them as negative. Notably, two patients were classified as basal-like by intrinsic subtyping (both mesenchymal (M) TNBC subtype) and had high expression of proliferation, immune, and/or angiogenesis signatures. Another patient was classified as HER2 overexpressed and had elevated HER2 amplicon signatures and low luminal pathway signatures. This suggests that Multiplex8+ reflected the underlying tumor biology better than IHC. Follow up data for 20 patients revealed insights into the ultility of genes and gene signatures in predicting risk of relapse and response to treatment. Multiplex8+ was correct in predicting treatment sensitivity or resistance in 77% (17/22) of instances, which spanned chemotherapy, endocrine, anti-Her2, and novel targeted therapies. For example, one metastatic TNBC patient who had an excellent response to bevacizumab had two heterogeneous regions dissected using LCM and both regions had high levels of a hypoxia/inflammation/angiogenesis signature previously shown to predict sensitivity to bevacizumab. This patient also had high TACSTD2 and TOP1 the antigen and payload targets of the ADC, sacituzumab govitecan (SG). Notably, after progressing on bevacizumab, this patient responded well to SG, highlighting the value of this test in predicting response to more than one line of therapy. Another patient with metastatic HR+/HER2− BCa had a low (neo)adjuvant chemotherapy response signature as well as a high E2F4 signature that was previously shown to predict resistance to aromatase inhibitors and sensitivity to CDK4/6 inhibitors. Indeed, this patient had a poor response to aromatase inhibitors and chemotherapy, relapsing 2 months after surgery, but then had a beneficial response to CDK4/6 inhibitors. These examples, and other patient vignettes that will be discussed, demonstrate the clinical potential of Multiplex8+ by providing unique insights for each patient that could identify potential treatments (including subsequent lines of therapy) and explain why prior treatments performed poorly. The specimens in this observational study were obtained with the patients informed consent and the study was approved by the Ethics Committee of the Bratislava Self-Governing Region (Ref. No. 05320/2020/HF). Citation Format: Evan Paul, Barbora Huraiová, Natália Matyašovská, Daniela Gábrišová, Soňa Gubová, Tomáš Ondris, Michal Gala, Liliane Barroso, Helena Ignačáková, Natália Valková, Fresia Pareja, Jakob N. Kather, Pavol Čekan. An observational cohort study of Multiplex8+: a spatially informed assay that uses multiplexed RNA-FISH guided laser capture microdissection followed by total RNA-sequencing [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P1-05-23.

Abstract PS17-01: Spatial transcriptomics identifies molecular patterns predictive of response to neoadjuvant chemotherapy in triple-negative breast cancer

Abstract Background: Triple-Negative Breast Cancer (TNBC) is a highly aggressive subtype of breast cancer and only ∼40% of TNBC patients achieve pathological complete response (pCR) following neoadjuvant chemotherapy (NACT). Since molecular heterogeneity of TNBC contributes to NACT response, we investigated their histology and spatial transcriptomics (ST) patterns. Methods: In total, 97 core biopsies, taken prior to (n = 92) or during (n = 5) treatment, and 44 post-treatment surgical resections were collected from 96 women with TNBC receiving NACT from archival samples and the ongoing FORCE (NCT03238144) clinical trial at Guy’s Hospital, London, UK. Each tissue section was H&amp;E stained, imaged, and annotated by a pathologist. A 6-digit classifier was designed to capture the histology of each region of interest (ROI), characterising the tumour region, epithelial type and immune cell localisation, population, distribution, and abundance. The annotations were used to guide ST profiling of serial sections, using the GeoMx Digital Spatial Profiler. Tissue sections were stained with fluorescent antibodies against PanCK and CD45. Tumour and immune-enriched compartments, defined as PanCK+/CD45- and PanCK-/CD45+, respectively, were acquired. Data analysis was performed in R 4.3.1, including quality control, batch correction, differential and topographical gene expression, immune cell deconvolution and gene set enrichment analysis (GSEA). Epithelial states (ES) were defined by identifying gene modules of co-expressed genes and annotating these with enriched pathways. RESULTS: To date, we have performed ST profiling of 120 TNBCs from 85 patients (40 Non-Responders; 45 Responders), totalling 4506 tumour or immune-cell enriched regions. Genes signatures known to predict pCR of NACT in women with TNBC were first assessed by generating a pseudo-bulk dataset produced in silico by combining gene counts from all regions. Two gene signatures, namely the PAM50 proliferation signature and the GeparSixto Immune signature, were found to be predictive (PAM50: OR = 0.12, p-value = 0.06; GeparSixto: OR= 0.19, p-value = 0.03) in our data. Seven transcriptionally distinct epithelial states (ES), each underpinning unique molecular functions, were defined by identifying gene modules of tumour-enriched regions. The relative abundance of ES2 (DNA repair mechanisms) and ES4 (deregulated ECM, increased epithelial-mesenchymal transition and TGF-b signaling) in pre-NACT core biopsies was significantly associated with pCR (OR=1.03, p-value = 0.02) and residual disease (OR = 0.98, p-value = 0.05), respectively. Within the immune compartment, seven clusters of tumour-immune microenvironments (TIMEs), were obtained by unsupervised hierarchical clustering of immune cell proportions. A high relative abundance of TIME1 (naïve CD4 T cells and plasma cells) and TIME5 (naïve CD4 T cells, naïve B cells, memory B cells and plasma cells) were significantly associated with pCR (TIME1 OR = 1.05, p-value = 0.01; TIME5 OR = 1.04, p-value = 0.05). We further investigated rates of co-occurrence between each ES and TIME, identifying combinations that were statistically most probable to occur and identified TIME as a confounding factor in the relative risk associated with the presence of individual ES. Moreover, we identify potential ligand-receptor interactions between distinct ES and TIMEs which may provide new insights into the mechanisms driving differential responses to NACT in TNBC. Conclusion: To date, we have collected one of the largest spatial transcriptomics data sets with detailed histological annotation. The abundance of epithelial states and tumour-immune microenvironments is predictive for pCR in NACT TNBC. The spatial location, colocalization and ligand-receptor interactions in ES and TIME provide novel insights into the molecular heterogeneity of TNBCs and mechanisms regulating responses to NACT. Citation Format: Isobelle Wall, Jelmar Quist, Sarah Pinder, KCL Biobank, Sheeba Irshad, Cheryl Gillet, Victoria Seewaldt, David Frankhouser, Shankar Subramanium, Maddy Parsons, and Anita Grigoriadis. Spatial transcriptomics identifies molecular patterns predictive of response to neoadjuvant chemotherapy in triple-negative breast cancer [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr PS17-01.

Abstract PS1-07: Al-Derived Tumor-Infiltrating Lymphocytes Enhance the Model with Baseline Stromal Tumor-Infiltrating Lymphocytes and Ki-67 in Predicting Pathologic Complete Response in an Early-Stage Triple-Negative Breast Cancer Prospective Trial

Abstract Background: The tumor immune microenvironment is crucial in shaping the response to neoadjuvant chemotherapy in triple-negative breast cancer (TNBC). Stromal tumor-infiltrating lymphocytes (sTILs) and Ki-67 are important biomarkers to predict treatment response in TNBC (Abuhadra et al., 2023). However, manual assessment of sTILs faces challenges due to variability in interpretation across different observers (Van Bockstal et al. 2021). To obtain a consistent predictive model for treatment response with chemotherapy in TNBC, we utilized artificial intelligence (AI) methods for enhancement. Material &amp; methods: We collected 408 hematoxylin and eosin (H&amp;E)-stained images from ARTEMIS (NCT02276443) arm 1 pretreatment core biopsies, divided into discovery and validation cohorts as per a previous work (Abuhadra et al., 2023). We also used the same criteria to evaluate pathologic complete response (pCR), Ki-67, and sTILs. H&amp;E images exhibiting metaplasia and giant cells were excluded, resulting in 201 slides for the discovery cohort and 193 slides for the validation cohort, with one slide per case. To calculate the TILs score, we applied AI pipelines on H&amp;E images to identify epithelial, lymphoid, stromal, and other cells (AbdulJabbar et al., 2020), and to segment tissue into tumor, stroma, parenchyma, necrosis/hemorrhage, and adipose tissue (unpublished). To ensure the accuracy of cell classification, we combined the two pipelines to refine cell recognition. Specifically, cell types identified outside the tumor area but misclassified as epithelial cells underwent a secondary prediction. AI-derived TILs (AI-TILs) was calculated as the proportion of identified lymphoid cells within the tumor and stromal tissues, excluding other cell types. Results: To evaluate the AI pipelines, 20 images from discovery cohort were manually annotated by three pathologists, resulting in 8010 cells. We achieved an average balanced accuracy of 91.2% across the identified cell types. In the discovery cohort (n=201; pCR rate 42%, 85/201), AI-TILs was notably associated with manual sTILs (Spearman’s rho=0.49, P&amp;lt;0.001). In a multivariable model consisting of AI-TILs and Ki-67 to predict pCR, with AI-TILs cutoff (0.173, range: 0.03-0.73) established through recursive partitioning analysis, we found that higher AI-TILs levels were significantly associated with pCR (Odds ratio, 4.57, 95% CI=1.63-12.80, P=0.004). Manual sTILs showed a similar pattern (Odds ratio, 3.71, 95% CI=1.97-6.99, P&amp;lt;0.001) in a multivariable model with Ki-67. When combining Ki-67, sTILs and AI-TILs, the significant association between AI-TILs and pCR was retained (Odds ratio: 3.01, 95% CI=1.04-8.72, P=0.042). Using the same model, we achieved an AUC of 0.74 and a precision of 0.68, marginally improving the model consisting of Ki-67 and sTILs (AUC 0.71, precision 0.68). In the validation cohort (n=193; pCR rate 42%, 82/193), the AI-TILs score performed consistently, maintaining correlation with sTILs (rho=0.39, P&amp;lt;0.001). By employing the model constructed in the discovery dataset with Ki-67, sTILs, and AI-TILs to predict pCR, we achieved an AUC of 0.69 and a precision of 0.67, comparable to the model with Ki-67 and sTILs alone (AUC 0.68, precision 0.66). Calibration plots and Hosmer-Lemeshow test indicated that the model including AI-TILs aligned better with the actual data compared to the model without AI-TILs. Conclusion: This study demonstrates that AI-TILs from baseline biopsies can be a promising biomarker to enhance the prediction for neoadjuvant chemotherapy response in TNBC, alongside existing predictors such as manual sTILs, underscoring the potential clinical value of AI-TILs due to its reproducibility and objectivity. Citation Format: Xiaoxi Pan, Caner Ercan, Zhongya Wang, Roland Bassett Jr., Karina B Pinao Gonzales, Clinton Yam, Lei Huo, Yinyin Yuan. Al-Derived Tumor-Infiltrating Lymphocytes Enhance the Model with Baseline Stromal Tumor-Infiltrating Lymphocytes and Ki-67 in Predicting Pathologic Complete Response in an Early-Stage Triple-Negative Breast Cancer Prospective Trial [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr PS1-07.

Abstract P4-03-17: Enhanced spatial detection of post-neoadjuvant breast cancer minimal residual disease by tissue-based Cancer Personalized Profiling by Deep Sequencing

Abstract Background: Circulating tumor DNA (ctDNA) is undetectable after neoadjuvant chemotherapy (NAC) in most breast cancer patients with residual disease, limiting its utility to inform adjuvant treatment. The current standard-of-care for assessing NAC response is pathologic evaluation of resection tissues. Many patients without histologically detected post-NAC disease, however, go on to recur, and many with significant gross residual disease are cured, highlighting the need to more accurately quantify low-burden post-NAC MRD, when adjuvant treatment decisions are made. Methods: We describe spatial post-NAC MRD detection by tissue-based Cancer Personalized Profiling by Deep Sequencing (t-CAPP-Seq), profiling tissue-based MRD (tMRD) in breast and lymph node resection tissues. Using personalized oligonucleotide hybrid capture panels derived from whole exome sequencing (WES) along with a fixed panel of recurrently mutated and biologically relevant breast cancer genes, we established spatial tMRD detection across numerous spatially catalogued samples per case using barcode mediated error suppression and a Monte Carlo statistical framework. We compared tMRD to pathologic detection in matched samples and described patterns of quantitative tMRD measurement that predict recurrence risk. Results: We followed a median of 36 (range: 1-112) single nucleotide variants (SNVs)/case derived from WES and a fixed panel of 56 common breast cancer genes spanning 225kb across 29 tumors representing a range of breast cancer subtypes. Nine (33%) and 10 (37%) tumors harbored TP53 and PIK3CA mutations, respectively. We genotyped 797 individual spatially resolved post-surgical specimens spanning these cases (median 24 post-treatment blocks/case, range: 12-46) and established strong concordance between genomic and histologic MRD detection (considering histology as the gold standard, sensitivity=78.8%, specificity=83.3%), with excellent classification of histologic status by tMRD (AUC=0.92). tMRD positive samples undetected by pathology had lower mean allele frequencies (AFs) than those detected by both methods (p&amp;lt;0.001), suggesting a superior limit of detection by tMRD. With a cohort median progression-free survival (PFS) of 140 months, we identified 8 progression events, including two local recurrences adjacent to a tMRD-positive surgical margin that was histologically negative for invasive carcinoma. One of these recurrences was identified post-mastectomy in a pathologic complete response case. Both tMRD positive margins contained missense PIK3CA SNVs that were absent from the pretreatment tumor, representing identification of targetable (alpelisib, capivasertib) post-NAC driver lesions that were selected under the pressure of chemotherapy. In the overall cohort, higher mean tMRD AF across the resection tissues was associated with inferior PFS (HR 1.29, 95% CI 1.08-1.55, p = 0.004). We defined tMRD high status as a mean case AF &amp;gt;3%, which predicted inferior PFS (p=0.034) and identified four genes (CSPP1, POLE, DNAAF4, PCNT) containing SNVs in post-NAC tissues that were significantly correlated with inferior PFS across the cohort (Cox proportional hazards model, p&amp;lt;0.05). Conclusions: Here we introduce t-CAPP-Seq, a novel tMRD detection platform that outperforms surgical pathology for MRD assessment, quantitatively and spatially profiling patterns of genomic lesions under the selection pressure of treatment that are associated with clinical progression. We anticipate tMRD-based interventional adjuvant approaches will motivate movement of molecularly targeted therapies into the adjuvant setting, leading to superior survival outcomes and challenging current treatment paradigms. Citation Format: Julia Ransohoff , Mia A. Carleton, Sofia Miron Barroso, Gregory Bean, Alisha Maltos, James Ford, George Duran, Michael Khodadoust, Melinda L. Telli, Ash A. Alizadeh, David M. Kurtz. Enhanced spatial detection of post-neoadjuvant breast cancer minimal residual disease by tissue-based Cancer Personalized Profiling by Deep Sequencing [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P4-03-17.

Development and validation of an MRI spatiotemporal interaction model for early noninvasive prediction of neoadjuvant chemotherapy response in breast cancer: a multicentre study

Development and validation of an MRI spatiotemporal interaction model for early noninvasive prediction of neoadjuvant chemotherapy response in breast cancer: a multicentre study

Predictive power of tertiary lymphoid structure for prognosis and neoadjuvant chemotherapy response in HER2-positive breast cancer

This study evaluated the prognostic significance of tertiary lymphoid structures (TLS) in human epidermal growth factor receptor 2 (HER2)-positive breast cancer (BC), focusing on their associations with survival outcomes, response to neoadjuvant therapy, and potential as a biomarker for personalized treatment strategies. Data from patients with HER2-positive BC in the METABRIC and The Cancer Genome Atlas databases were analyzed. TLS expression scores were calculated using gene set variation analysis, and their associations with survival outcomes were assessed. Immune cell infiltration, immune checkpoint expression, tumor mutational burden, and pathway enrichment were also evaluated. Data from the I-SPY2 clinical trial and a clinicopathological cohort of 19 patients from Xiangya Hospital were used to assess the relationship between TLS expression and pathological complete response following neoadjuvant therapy. High TLS expression was associated with improved survival and increased infiltration of antitumor immune cells. TLS-high tumors were enriched in immune-related pathways, whereas TLS-low tumors showed activation of proliferation and metabolism pathways. Patients with high TLS expression had better responses to neoadjuvant therapy, while those with low TLS expression derived greater benefit from dual-targeted treatments. TLS represents a promising biomarker for predicting survival and response to neoadjuvant therapy in HER2-positive BC, with potential to support personalized treatment strategies.

Combination of ultrasound-based radiomics and deep learning with clinical data to predict response in breast cancer patients treated with neoadjuvant chemotherapy.

Accurate assessment of NAC efficacy is crucial for determining appropriate surgical strategies and guiding the extent of surgical resection in breast cancer. Therefore, this study aimed to design an integrated predictive model combining ultrasound imaging, deep learning features, and clinical characteristics to predict pCR in breast cancer patients undergoing NAC. A retrospective study was conducted, including 643 pathologically confirmed breast cancer patients who underwent NAC between January 2022 to February 2024 from two institutions (Center 1: 372 cases; Center 2: 271 cases). Ultrasound images before and after NAC were collected for each patient. A total of 2,920 radiomics features and 4,096 deep learning features were extracted from the ultrasound images. Multiple machine learning algorithms were employed to model and validate the diagnostic performance of different types of features. Finally, clinical data, radiomics, and deep learning features were integrated to form a fusion model, which was evaluated using receiver operating characteristic (ROC) analysis. The combined model achieved the highest predictive performance for pathological complete response (pCR) across both cohorts. In the internal validation cohort, it reached an accuracy of 0.892 (95% CI: 0.862-0.912) and an AUC of 0.901 (95% CI: 0.854-0.948). In the external cohort, it maintained strong performance with an accuracy of 0.857 (95% CI: 0.822-0.928) and an AUC of 0.891 (95% CI: 0.848-0.934), significantly outperforming the individual models (DeLong test, p < 0.01).The deep learning model showed solid performance with accuracies of 0.875 and 0.833 in the internal and external cohorts, respectively, and AUCs of 0.870 and 0.874. The radiomics model displayed moderate accuracy and AUC in both cohorts, while the clinical model showed the lowest predictive capability among the models, with accuracy and AUC values around 0.67 in both cohorts. The combined model, integrating clinical, radiomics, and deep learning features, demonstrated superior predictive accuracy for pCR following neoadjuvant chemotherapy (NAC) in breast cancer patients, outperforming individual models. This integrated approach highlights the value of combining diverse data types to improve prediction, offering a promising tool for guiding NAC response assessment and personalized treatment planning.