CoChat

Digital image analysis and machine learning-assisted prediction of neoadjuvant chemotherapy response in triple-negative breast cancer

Open access

Published version of a preprint

The DOI matches its Crossref record; 78 of the 79 references it cites were found in a registry or an index; 13 of the 18 sources its claims cite were checked for support.

Abstract

BACKGROUND: Pathological complete response (pCR) is associated with favorable prognosis in patients with triple-negative breast cancer (TNBC). However, only 30-40% of TNBC patients treated with neoadjuvant chemotherapy (NAC) show pCR, while the remaining 60-70% show residual disease (RD). The role of the tumor microenvironment in NAC response in patients with TNBC remains unclear. In this study, we developed a machine learning-based two-step pipeline to distinguish between various histological components in hematoxylin and eosin (H&E)-stained whole slide images (WSIs) of TNBC tissue biopsies and to identify histological features that can predict NAC response. METHODS: H&E-stained WSIs of treatment-naïve biopsies from 85 patients (51 with pCR and 34 with RD) of the model development cohort and 79 patients (41 with pCR and 38 with RD) of the validation cohort were separated through a stratified eightfold cross-validation strategy for the first step and leave-one-out cross-validation strategy for the second step. A tile-level histology label prediction pipeline and four machine-learning classifiers were used to analyze 468,043 tiles of WSIs. The best-trained classifier used 55 texture features from each tile to produce a probability profile during testing. The predicted histology classes were used to generate a histology classification map of the spatial distributions of different tissue regions. A patient-level NAC response prediction pipeline was trained with features derived from paired histology classification maps. The top graph-based features capturing the relevant spatial information across the different histological classes were provided to the radial basis function kernel support vector machine (rbfSVM) classifier for NAC treatment response prediction. RESULTS: The tile-level prediction pipeline achieved 86.72% accuracy for histology class classification, while the patient-level pipeline achieved 83.53% NAC response (pCR vs. RD) prediction accuracy of the model development cohort. The model was validated with an independent cohort with tile histology validation accuracy of 83.59% and NAC prediction accuracy of 81.01%. The histological class pairs with the strongest NAC response predictive ability were tumor and tumor tumor-infiltrating lymphocytes for pCR and microvessel density and polyploid giant cancer cells for RD. CONCLUSION: Our machine learning pipeline can robustly identify clinically relevant histological classes that predict NAC response in TNBC patients and may help guide patient selection for NAC treatment.

Overview

CoChat Summary

## TL;DR

This study developed a two-step machine learning pipeline on H&E whole-slide images from triple-negative breast cancer biopsies to predict neoadjuvant chemotherapy response (pCR vs residual disease). In independent validation, the model achieved 83.53% patient-level training/cohort performance in development and 81.01% accuracy in the validation cohort, with the strongest predictive histology pairs linked to pCR and RD.

## Key findings

  • The tile-level histology classifier achieved 86.72% accuracy in cross-validation, and 83.59% validation accuracy in the independent cohort.
  • The patient-level NAC response model achieved 83.53% accuracy in the development cohort and 81.01% accuracy in the validation cohort.
  • In validation, the NAC response pipeline correctly predicted 33/41 pCR cases and 31/38 RD cases, with specificity 81.58%, sensitivity 80.49%, and AUC 0.83.
  • The most predictive histology pairs were tumor + tumor-infiltrating lymphocytes (tTILs) for pCR and microvessel density (MVD) + polyploid giant cancer cells (PGCCs) for RD; among 120 histology feature pairs, the authors selected the top eight as most predictive.

## Methods

  • Retrospective analysis of H&E-stained pre-NAC core needle biopsy WSIs from treatment-naïve TNBC patients: 85 patients in the model development cohort (51 pCR, 34 RD) and 79 patients in the validation cohort (41 pCR, 38 RD).
  • Two-step pipeline: first, tile-level histology classification of 468,043 tiles using texture features (reduced from 80 to 55) and four classifiers; second, patient-level NAC response prediction using graph-based spatial TME features from paired histology classification maps.
  • Main analysis used stratified 8-fold cross-validation for tile classification and leave-one-out cross-validation (LOO-CV) for patient-level prediction, with rbfSVM performing best at both stages.

## Limitations

  • The paper states limitations of small sample sizes, slide quality issues, and expensive computational processes.
  • It also notes class imbalance across histology classes and that more histology classes and more expert annotators would likely improve performance.
  • Evident from the text: the study is retrospective and drawn from only two institutions, which may limit generalizability.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1186/s13058-023-01752-y

Cites

78 of 78 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Overall schema of the developed NAC response prediction pipeline. The tile-level histology classification module (first step) consists of A training WSI annotation; B definition of histology classes of interest; C tile preprocessing; D feature extraction and selection; E classifier training, testing, and validation; and F generation of histology classification map. The patient-level NAC response prediction module (second step) consists of G graph node identification; H TME spatial descriptor computation; I graph construction and graph feature selection; J machine learning model training, testing, and validation; and K generation of an attention map with highlighted tissue regions with full feature set. Abbreviations: FE, feature extraction; sTILs, stromal TILs; tTILs, tumor TILs; Feat, feature; MVD, microvessel; PGCC, polyploid giant cancer cell
Fig. 1 Overall schema of the developed NAC response prediction pipeline. The tile-level histology classification module (first step) consists of A training WSI annotation; B definition of histology classes of interest; C tile preprocessing; D feature extraction and selection; E classifier training, testing, and validation; and F generation of histology classification map. The patient-level NAC response prediction module (second step) consists of G graph node identification; H TME spatial descriptor computation; I graph construction and graph feature selection; J machine learning model training, testing, and validation; and K generation of an attention map with highlighted tissue regions with full feature set. Abbreviations: FE, feature extraction; sTILs, stromal TILs; tTILs, tumor TILs; Feat, feature; MVD, microvessel; PGCC, polyploid giant cancer cellCC BY · PubMed Central
Fig. 2 Representative image tiles of distinct histological classes and a tile-level histology classification map. A Typical examples of image tiles (224 × 224) capturing stroma, tTILs, benign tissue, and vessels are presented. B A histology classification map is presented to visualize the spatial organization of TME components related to stroma (green), benign tumor (blue), tTILs (purple), and blood vessels (red)
Fig. 2 Representative image tiles of distinct histological classes and a tile-level histology classification map. A Typical examples of image tiles (224 × 224) capturing stroma, tTILs, benign tissue, and vessels are presented. B A histology classification map is presented to visualize the spatial organization of TME components related to stroma (green), benign tumor (blue), tTILs (purple), and blood vessels (red)CC BY · PubMed Central
Fig. 3 Testing tile-level histology classification performance in the model development cohort. Confusion matrix showing the aggregated performance of the rbfSVM model for tile-level histology class prediction (i.e., 0, stroma; 1, tumor; 2, tertiary TILs; 3, stroma TILs; 4, normal tissue; 5, PGCCs; 6, blood vessels; 7, necrosis; 8, microvessel; 9, benign tumor; 10, tumor TILs; 11, in situ carcinoma; 12, hemorrhage; 13, adipocytes; 14, apocrine change; 15 mucinous change; and 16, background). Abbreviations: 1NN, 1-nearest neighbor; linSVM, linear support vector machine SVM; PGCC: polyploid giant cancer cells; rbfSVM, radial basis function SVM
Fig. 3 Testing tile-level histology classification performance in the model development cohort. Confusion matrix showing the aggregated performance of the rbfSVM model for tile-level histology class prediction (i.e., 0, stroma; 1, tumor; 2, tertiary TILs; 3, stroma TILs; 4, normal tissue; 5, PGCCs; 6, blood vessels; 7, necrosis; 8, microvessel; 9, benign tumor; 10, tumor TILs; 11, in situ carcinoma; 12, hemorrhage; 13, adipocytes; 14, apocrine change; 15 mucinous change; and 16, background). Abbreviations: 1NN, 1-nearest neighbor; linSVM, linear support vector machine SVM; PGCC: polyploid giant cancer cells; rbfSVM, radial basis function SVMCC BY · PubMed Central
Fig. 4 Testing tile-level histology classification performance of the rbfSVM classifier with the Emory Hospital development cohort. Each bar represents the weighted average of tiles and their predicted probabilities during testing for a histology class. Specificity (green) measures how often the rbfSVM classifier correctly predicted true negatives. AUC (blue) reflects the model's ability to distinguish between positive and negative classes. Accuracy (yellow) indicates the proportion of correct predictions out of total predictions. F1-score (gray) presents a balanced view of rbfSVM classifier performance. Sensitivity (orange) suggests how often the rbfSVM classifier correctly identifies positive instances. Precision (blue) indicates how often the rbfSVM classifier correctly predicts true positives. Error bars represent the 95% confidence interval in each case
Fig. 4 Testing tile-level histology classification performance of the rbfSVM classifier with the Emory Hospital development cohort. Each bar represents the weighted average of tiles and their predicted probabilities during testing for a histology class. Specificity (green) measures how often the rbfSVM classifier correctly predicted true negatives. AUC (blue) reflects the model's ability to distinguish between positive and negative classes. Accuracy (yellow) indicates the proportion of correct predictions out of total predictions. F1-score (gray) presents a balanced view of rbfSVM classifier performance. Sensitivity (orange) suggests how often the rbfSVM classifier correctly identifies positive instances. Precision (blue) indicates how often the rbfSVM classifier correctly predicts true positives. Error bars represent the 95% confidence interval in each caseCC BY · PubMed Central
Fig. 5 TME graph feature selection by ReliefF. TME graph features histology class pairs sorted by their importance weights from the ReliefF algorithm in the model development cohort. Abbreviations: TF, Texture Feature; sTILs, stromal TILs; tTILs, tumor TILs; PGCCs, polyploid giant cancer cells; GLCM, gray-level co-occurrence matrix
Fig. 5 TME graph feature selection by ReliefF. TME graph features histology class pairs sorted by their importance weights from the ReliefF algorithm in the model development cohort. Abbreviations: TF, Texture Feature; sTILs, stromal TILs; tTILs, tumor TILs; PGCCs, polyploid giant cancer cells; GLCM, gray-level co-occurrence matrixCC BY · PubMed Central
Fig. 6 Patient-level NAC response prediction performance of the best classifier by LOO-CV in the model development cohort. A Confusion matrix showing performance of the rbfSVM model. B ROC curve of the best NAC response prediction pipeline. Abbreviations: LOO-CV, leave-one-out cross-validation; rbfSVM, radial basis function SVM; ROC, receiver operating characteristic; TN, true negative; FP, false positive; FN, false negative
Fig. 6 Patient-level NAC response prediction performance of the best classifier by LOO-CV in the model development cohort. A Confusion matrix showing performance of the rbfSVM model. B ROC curve of the best NAC response prediction pipeline. Abbreviations: LOO-CV, leave-one-out cross-validation; rbfSVM, radial basis function SVM; ROC, receiver operating characteristic; TN, true negative; FP, false positive; FN, false negativeCC BY · PubMed Central
Fig. 7 The NAC prediction model's performance in the validation cohort. A Confusion matrix showing performance of the best NAC response prediction model. B ROC curve of the best NAC response prediction pipeline. Abbreviations: LOO-CV, leave-one-out cross-validation; rbfSVM, radial basis function SVM; ROC, receiver operating characteristic; TN, true negative; FP, false positive; FN, false negative; TP, true positive
Fig. 7 The NAC prediction model's performance in the validation cohort. A Confusion matrix showing performance of the best NAC response prediction model. B ROC curve of the best NAC response prediction pipeline. Abbreviations: LOO-CV, leave-one-out cross-validation; rbfSVM, radial basis function SVM; ROC, receiver operating characteristic; TN, true negative; FP, false positive; FN, false negative; TP, true positiveCC BY · PubMed Central

Verification

Claims the paper makes

5 Valid · 2 Not valid · 0 Not checked · 52 Own results and interpretations

32 more claims

Create a free account to see all 59 claims and to verify the 2 references still open.

See all claims — free account
  • Probably not a substantial claim. 14
  • It may state two separate claims as one. 11

Create a free account to see these claims.

Citations

Cites

Show all 78 references

Research works printed but not linked: 3

  1. Howlader NNA, Krapcho M, Miller D, Brest A, Yu M, Ruhl J, Tatalovich Z, Mariotto A, Lewis DR, Chen HS, Feuer EJ, Cronin KA (eds. SEER Cancer Statistics Review. 1975–2017).
  2. Zgura A, Galesa L, Bratila E, Anghel R. Not available. Maedica (Bucur). 2018;13(4):317–20.
  3. Robnik-Sikonja M, Kononenko I. An adaptation of Relief for attribute estimation in regression. In: Proceedings of the fourteenth international conference on machine learning: Morgan Kaufmann Publishers Inc.; 1997. pp. 296–304.

Cited by

See all 42

Metadata

Authors
Fisher, T. B., Saini, G., Rekha, T. S., Krishnamurthy, J., Bhattarai, S., Callagy, G., Webber, M., Janssen, E. A. M., Kong, J., & Aneja, R.
Affiliations
  • Georgia State University (US)
  • University of Alabama at Birmingham (US)
  • JSS Academy of Higher Education and Research (IN)
  • Ollscoil na Gaillimhe – University of Galway (IE)
  • Stavanger University Hospital (NO)
Corresponding author
  • Kong, Jun
  • Aneja, Ritu
Venue
Breast Cancer Research
Published
2024-01-18
Volume
26
Issue
1
Pages
12-12
ISSN
  • 1465-5411 (ISSN-L)
  • 1465-542X
OpenAlex
W4390971849
Cited by
42
References
79
Citations by year
2026: 17 · 2025: 18 · 2024: 7
FWCI
6.19
Type
journal article
Language
en
Open access
Licence
Locations
  • Breast Cancer Research (publishedVersion, cc-by)
  • Research Square (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Institutes of Health (1U01CA242936)
  • National Institutes of Health,United States (R01CA239120)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › AI in cancer detection
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Radiomics and Machine Learning in Medical Imaging
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Cancer Research › Breast Cancer Treatment Studies
Keywords
triple-negative breast cancer, neoadjuvant chemotherapy response, pathological complete response, whole slide images, histological classification, tumor-infiltrating lymphocytes, microvessel density, polyploid giant cancer cells, residual disease, tumor microenvironment, texture features, treatment response prediction, patient selection, digital pathology, support vector machine, graph features

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Digital image analysis and machine learning-assisted prediction of neoadjuvant chemotherapy response in triple-negative breast cancer. https://app.cochat.ai/papers/10.1186/s13058-023-01752-y
The paper
Fisher, T. B., Saini, G., Rekha, T. S., Krishnamurthy, J., Bhattarai, S., Callagy, G., Webber, M., Janssen, E. A. M., Kong, J., & Aneja, R. (2024). Digital image analysis and machine learning-assisted prediction of neoadjuvant chemotherapy response in triple-negative breast cancer. Breast Cancer Research. https://doi.org/10.1186/s13058-023-01752-y

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.