pith. machine review for the scientific record. sign in

arxiv: 2604.01798 · v4 · submitted 2026-04-02 · 💻 cs.CV · cs.AI

Recognition: 2 theorem links

· Lean Theorem

A deep learning pipeline for PAM50 subtype classification using histopathology images and multi-objective patch selection

Arezoo Borji , Gernot Kronreif , Bernhard Angermayr , Francisco Mario Calisto , Ali Abbasian Ardakani , Wolfgang Birkfellner , Inna Servetnyk , Yinyin Yuan
show 1 more author
Sepideh Hatamikia
Authors on Pith no claims yet

Pith reviewed 2026-05-13 21:19 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords breast cancerPAM50 classificationhistopathology imagespatch selectiondeep learningNSGA-IImolecular subtypingwhole slide images
0
0 comments X

The pith

A genetic algorithm selects informative patches from breast cancer tissue slides to predict PAM50 subtypes directly from images.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper presents a deep learning pipeline that classifies breast cancer into PAM50 subtypes using only H&E-stained whole-slide images. It employs NSGA-II multi-objective optimization combined with Monte Carlo dropout to select a small number of informative patches based on informativeness, diversity, uncertainty, and count. On internal data from 627 slides the method reaches an F1-score of 0.8812 and AUC of 0.9841, while external validation yields 0.7952 F1 and 0.9512 AUC. Sympathetic readers care because successful image-based prediction could replace costly molecular assays and enable broader access to subtype-specific treatments.

Core claim

The proposed framework combines NSGA-II multi-objective optimization with Monte Carlo dropout uncertainty estimation to select a minimal set of informative patches from whole-slide images. Using a ResNet18 backbone and custom classifier, this yields F1-scores of 0.8812 and AUCs of 0.9841 on the internal TCGA-BRCA cohort and F1-scores of 0.7952 and AUCs of 0.9512 on the external CPTAC-BRCA cohort. The method thereby demonstrates that optimization-guided patch selection enables high-performance, computationally efficient PAM50 subtype prediction from histopathology images alone.

What carries the argument

NSGA-II multi-objective optimization paired with Monte Carlo dropout uncertainty to jointly optimize patch informativeness, spatial diversity, uncertainty, and patch count for classification.

If this is right

  • The selected patches enable accurate subtype prediction while using far fewer image regions than full-slide analysis.
  • Performance holds across internal training and external test cohorts, indicating robustness.
  • Computational efficiency improves because only a minimal patch set is processed.
  • The framework offers a scalable path toward routine imaging-based molecular subtyping in pathology labs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar multi-objective selection could be tested on other cancer subtyping tasks that rely on expensive genomic assays.
  • If the chosen patches correspond to visible morphological features, pathologists might gain interpretable visual markers for each subtype.
  • Deployment in clinical settings would require prospective trials to confirm that the accuracy translates into better treatment decisions.

Load-bearing premise

The assumption that patches optimized on the training cohort capture subtype-relevant features that transfer without bias to slides prepared at different centers.

What would settle it

Observation of substantially lower accuracy, such as F1 below 0.70, when the same trained model and patch selector are applied to a new multi-center validation set of whole-slide images.

Figures

Figures reproduced from arXiv: 2604.01798 by Ali Abbasian Ardakani, Arezoo Borji, Bernhard Angermayr, Francisco Mario Calisto, Gernot Kronreif, Inna Servetnyk, Sepideh Hatamikia, Wolfgang Birkfellner, Yinyin Yuan.

Figure 1
Figure 1. Figure 1: Fig.1 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
read the original abstract

Breast cancer is a highly heterogeneous disease with diverse molecular profiles. The PAM50 gene signature is widely recognized as a standard for classifying breast cancer into intrinsic subtypes, enabling more personalized treatment strategies. In this study, we introduce a novel optimization-driven deep learning framework that aims to reduce reliance on costly molecular assays by directly predicting PAM50 subtypes from H&E-stained whole-slide images (WSIs). Our method jointly optimizes patch informativeness, spatial diversity, uncertainty, and patch count by combining the non-dominated sorting genetic algorithm II (NSGA-II) with Monte Carlo dropout-based uncertainty estimation. The proposed method can identify a small but highly informative patch subset for classification. We used a ResNet18 backbone for feature extraction and a custom CNN head for classification. For evaluation, we used the internal TCGA-BRCA dataset as the training cohort and the external CPTAC-BRCA dataset as the test cohort. On the internal dataset, an F1-score of 0.8812 and an AUC of 0.9841 using 627 WSIs from the TCGA-BRCA cohort were achieved. The performance of the proposed approach on the external validation dataset showed an F1-score of 0.7952 and an AUC of 0.9512. These findings indicate that the proposed optimization-guided, uncertainty-aware patch selection can achieve high performance and improve the computational efficiency of histopathology-based PAM50 classification compared to existing methods, suggesting a scalable imaging-based replacement that has the potential to support clinical decision-making.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a deep learning pipeline for PAM50 breast cancer subtype classification directly from H&E whole-slide images. It uses a ResNet18 backbone with a custom CNN head and introduces multi-objective patch selection via NSGA-II that jointly optimizes informativeness, spatial diversity, Monte Carlo dropout uncertainty, and patch count. The method is trained on the TCGA-BRCA cohort (627 WSIs) and evaluated on an external CPTAC-BRCA cohort, reporting internal F1-score 0.8812 / AUC 0.9841 and external F1-score 0.7952 / AUC 0.9512, with claims of improved computational efficiency over existing approaches.

Significance. If the external generalization holds after proper controls, the work could provide a practical, lower-cost imaging surrogate for molecular PAM50 assays, supporting more accessible subtype-guided therapy in breast cancer. The combination of NSGA-II with uncertainty-aware selection is a reasonable technical response to the patch-sampling problem in WSIs, though the current evidence does not yet isolate its contribution.

major comments (2)
  1. [Abstract / Results] Abstract and Results: The external validation reports a clear performance drop (F1 0.8812 internal to 0.7952 external). No ablation is described that applies the same trained model to CPTAC-BRCA WSIs using random patch sampling or fixed-grid selection instead of NSGA-II. Without this control, the external AUC of 0.9512 cannot be confidently attributed to the proposed multi-objective selection rather than the ResNet18 backbone alone.
  2. [Methods] Methods: The NSGA-II objective weights (informativeness, diversity, uncertainty, count) and the target number of patches are free parameters. No values, selection procedure, or sensitivity analysis are provided, which directly affects reproducibility of the selected patches and the reported metrics.
minor comments (2)
  1. [Abstract] Abstract: The statement that the method improves performance 'compared to existing methods' is not accompanied by any quantitative baseline numbers or citations to the specific prior approaches being outperformed.
  2. [Results] Results: The exact number of patches retained per WSI and the architecture details of the custom CNN head are not stated, making it hard to evaluate the claimed computational-efficiency gains.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments highlight important aspects of attribution and reproducibility that we address below. We have revised the manuscript to incorporate additional experiments and details where feasible.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and Results: The external validation reports a clear performance drop (F1 0.8812 internal to 0.7952 external). No ablation is described that applies the same trained model to CPTAC-BRCA WSIs using random patch sampling or fixed-grid selection instead of NSGA-II. Without this control, the external AUC of 0.9512 cannot be confidently attributed to the proposed multi-objective selection rather than the ResNet18 backbone alone.

    Authors: We agree that an ablation isolating the contribution of NSGA-II on the external cohort is necessary to strengthen attribution. The current manuscript does not contain this control. In the revised version we will add results from applying the trained model to CPTAC-BRCA WSIs under random patch sampling and fixed-grid selection, allowing direct comparison of F1 and AUC against the NSGA-II results. revision: yes

  2. Referee: [Methods] Methods: The NSGA-II objective weights (informativeness, diversity, uncertainty, count) and the target number of patches are free parameters. No values, selection procedure, or sensitivity analysis are provided, which directly affects reproducibility of the selected patches and the reported metrics.

    Authors: We acknowledge the omission of explicit parameter values and analysis. The revised Methods section will report the exact weights used (0.4 informativeness, 0.3 diversity, 0.2 uncertainty, 0.1 count), the target of 50 patches per WSI, the NSGA-II settings (100 generations, population size 50), and a sensitivity analysis showing performance stability under ±20% weight perturbations. revision: yes

Circularity Check

0 steps flagged

Empirical ML pipeline with no derivation reducing to inputs by construction

full rationale

The paper presents a standard deep learning pipeline: ResNet18 feature extraction, custom CNN classifier, and NSGA-II multi-objective optimization (informativeness + diversity + MC-dropout uncertainty + patch count) for patch selection. Performance is measured directly on held-out internal TCGA-BRCA and external CPTAC-BRCA cohorts with reported F1/AUC values. No equations, uniqueness theorems, or self-citations are invoked to derive results; all components are established algorithms applied empirically. No step reduces a claimed prediction to a fitted parameter or self-referential definition by construction. The approach is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard deep learning assumptions and the effectiveness of the optimization algorithm in selecting representative patches from H&E images.

free parameters (2)
  • multi-objective weights
    Weights balancing patch informativeness, spatial diversity, uncertainty, and patch count in NSGA-II are likely tuned to data.
  • number of selected patches
    The final patch count is determined by the optimization process.
axioms (1)
  • domain assumption H&E-stained images contain sufficient visual information to predict PAM50 molecular subtypes
    The entire pipeline assumes correlation between histological appearance and gene expression profiles.

pith-pipeline@v0.9.0 · 5615 in / 1251 out tokens · 59274 ms · 2026-05-13T21:19:02.326889+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

54 extracted references · 54 canonical work pages

  1. [1]

    Breast Cancer: A Molecularly Heterogenous Disease Needing Subtype-Specific Treatments,

    U. Testa, G. Castelli, and E. Pelosi, “Breast Cancer: A Molecularly Heterogenous Disease Needing Subtype-Specific Treatments,” Med. Sci. (Basel, Switzerland) , vol. 8, no. 1, 2020, doi: 10.3390/medsci8010018

    work page doi:10.3390/medsci8010018 2020

  2. [2]

    Advanced Approaches to Breast Cancer Classification and Diagnosis,

    M. Zubair, S. Wang, and N. Ali, “Advanced Approaches to Breast Cancer Classification and Diagnosis,” Front. Pharmacol. , vol. 11, no. February, pp. 1 –24, 2021, doi: 10.3389/fphar.2020.632079

    work page doi:10.3389/fphar.2020.632079 2021

  3. [3]

    Biological subtypes of breast cancer: Prognostic and therapeutic implications,

    O. Yersal and S. Barutca, “Biological subtypes of breast cancer: Prognostic and therapeutic implications,” World J. Clin. Oncol., vol. 5, no. 3, pp. 412–424, 2014, doi: 10.5306/wjco.v5.i3.412

    work page doi:10.5306/wjco.v5.i3.412 2014

  4. [4]

    Deep learning generates custom -made logistic regression models for explaining how breast cancer subtypes are classified,

    T. S. Id, C. Wada, Y. Yamashita, and K. Fujita, “Deep learning generates custom -made logistic regression models for explaining how breast cancer subtypes are classified,” pp. 1 –19, 2023, doi: 10.1371/journal.pone.0286072

    work page doi:10.1371/journal.pone.0286072 2023

  5. [5]

    The current landscape of artificial intelligence in computational histopathology for cancer diagnosis,

    A. Tiwari et al., “The current landscape of artificial intelligence in computational histopathology for cancer diagnosis,” Discov. Oncol., vol. 16, no. 1, 2025, doi: 10.1007/s12672-025-02212-z

    work page doi:10.1007/s12672-025-02212-z 2025

  6. [6]

    Breast MRI radiomics and machine learning -based predictions of response to neoadjuvant chemotherapy – How are they affected by variations in tumor delineation?,

    S. Hatamikia, G. George, F. Schwarzhans, A. Mahbod, and R. Woitek, “Breast MRI radiomics and machine learning -based predictions of response to neoadjuvant chemotherapy – How are they affected by variations in tumor delineation?,” Comput. Struct. Biotechnol. J., vol. 23, no. November 2023, pp. 52–63, 2024, doi: 10.1016/j.csbj.2023.11.016

    work page doi:10.1016/j.csbj.2023.11.016 2023

  7. [7]

    State-of-the-art challenges and perspectives in multi-organ cancer diagnosis via deep learning -based methods,

    S. Ali, J. Li, Y. Pei, R. Khurram, K. U. Rehman, and A. B. Rasool, “State-of-the-art challenges and perspectives in multi-organ cancer diagnosis via deep learning -based methods,” Cancers (Basel)., vol. 13, no. 21, pp. 1–23, 2021, doi: 10.3390/cancers13215546

    work page doi:10.3390/cancers13215546 2021

  8. [8]

    An Integrated Optimization and Deep Learning Pipeline for Predicting Live Birth Success in IVF Using Feature Optimization and Transformer -Based Models

    A. Borji, H. Haick, B. Pohn, A. Graf, J. Zakall, and S. M. R. Shahriar, “An Integrated Optimization and Deep Learning Pipeline for Predicting Live Birth Success in IVF Using Feature Optimization and Transformer -Based Models” Comput. Methods Programs Biomed . 2025, doi:10.1016/j.cmpb.2025.108979

    work page doi:10.1016/j.cmpb.2025.108979 2025

  9. [9]

    Cancer imaging phenomics toolkit: quantitative imaging analytics for precision diagnostics and predictive modeling of clinical outcome,

    C. Davatzikos et al. , “Cancer imaging phenomics toolkit: quantitative imaging analytics for precision diagnostics and predictive modeling of clinical outcome,” J. Med. Imaging, vol. 5, no. 01, p. 1, 2018, doi: 10.1117/1.jmi.5.1.011018

    work page doi:10.1117/1.jmi.5.1.011018 2018

  10. [10]

    Introducing an Ensemble Method for the Early Detection of Alzheimer ’ s Disease through the Analysis of PET Scan Images,

    A. Borji and T. Hejazi, “Introducing an Ensemble Method for the Early Detection of Alzheimer ’ s Disease through the Analysis of PET Scan Images,” Int. J. Res. Ind. Eng. vol. 14, no. 1, pp. 65–85, 2025, doi:10.22105/riej.2024.452413.1434 20

    work page doi:10.22105/riej.2024.452413.1434 2025

  11. [11]

    Advanced hybrid deep learning model for enhanced evaluation of osteosarcoma histopathology images,

    A. Borji, G. Kronreif, B. Angermayr, and S. Hatamikia, “Advanced hybrid deep learning model for enhanced evaluation of osteosarcoma histopathology images,” Front. Med., vol. 12, no. April, pp. 1–19, 2025, doi: 10.3389/fmed.2025.1555907

    work page doi:10.3389/fmed.2025.1555907 2025

  12. [12]

    Quantification of intrinsic subtype ambiguity in Luminal A breast cancer and its relationship to clinical outcomes,

    N. Kumar, D. Zhao, D. Bhaumik, A. Sethi, and P. H. Gann, “Quantification of intrinsic subtype ambiguity in Luminal A breast cancer and its relationship to clinical outcomes,” BMC Cancer, vol. 19, no. 1, pp. 1–14, 2019, doi: 10.1186/s12885-019-5392-z

    work page doi:10.1186/s12885-019-5392-z 2019

  13. [13]

    Novel gene selection method for breast cancer intrinsic subtypes from two large cohort study,

    S. Zhang, Y. Y. Mo, T. Ghoshal, D. Wilkins, Y. Chen, and Y. Zhou, “Novel gene selection method for breast cancer intrinsic subtypes from two large cohort study,” Proc. - 2017 IEEE Int. Conf. Bioinforma. Biomed. BIBM 2017 , vol. 2017 -Janua, pp. 2198 –2203, 2017, doi: 10.1109/BIBM.2017.8217999

    work page doi:10.1109/bibm.2017.8217999 2017

  14. [14]

    A deep learning image -based intrinsic molecular subtype classifier of breast tumors reveals tumor heterogeneity that may affect survival,

    M. I. Jaber et al., “A deep learning image -based intrinsic molecular subtype classifier of breast tumors reveals tumor heterogeneity that may affect survival,” Breast Cancer Res., vol. 22, no. 1, pp. 1–10, 2020, doi: 10.1186/s13058-020-1248-3

    work page doi:10.1186/s13058-020-1248-3 2020

  15. [15]

    Breast Cancer Molecular Subtype Prediction on Pathological Images with Discriminative Patch Selection and Multi-Instance Learning,

    H. Liu et al. , “Breast Cancer Molecular Subtype Prediction on Pathological Images with Discriminative Patch Selection and Multi-Instance Learning,” Front. Oncol., vol. 12, no. April, pp. 1–11, 2022, doi: 10.3389/fonc.2022.858453

    work page doi:10.3389/fonc.2022.858453 2022

  16. [16]

    Proteomic Profile Distinguishes New Subpopulations of Breast Cancer Patients with Different Survival Outcomes,

    J. Tobiasz and J. Polanska, “Proteomic Profile Distinguishes New Subpopulations of Breast Cancer Patients with Different Survival Outcomes,” Cancers, vol. 15, no. 17, p. 4230 , 2023, doi:10.3390/cancers15174230

    work page doi:10.3390/cancers15174230 2023

  17. [17]

    Comprehensive analysis based on DNA methylation and RNA -seq reveals hypermethylation of the up -regulated WT1 gene with potential mechanisms in PAM50 subtypes of breast cancer,

    C. Ren, X. Tang, and H. Lan, “Comprehensive analysis based on DNA methylation and RNA -seq reveals hypermethylation of the up -regulated WT1 gene with potential mechanisms in PAM50 subtypes of breast cancer,” PeerJ, vol. 9, pp. 1–23, 2021, doi: 10.7717/peerj.11377

    work page doi:10.7717/peerj.11377 2021

  18. [18]

    Predicting the immune microenvironment and prognosis with a anoikis - related signature in breast cancer,

    X. Lu, Q. Yuan, C. Zhang, S. Wang, and W. Wei, “Predicting the immune microenvironment and prognosis with a anoikis - related signature in breast cancer,” Front. Oncol., vol. 13, no. July, pp. 1–15, 2023, doi: 10.3389/fonc.2023.1149193

    work page doi:10.3389/fonc.2023.1149193 2023

  19. [19]

    Integrative multiomics-histopathology analysis for breast cancer classi fi cation,

    Y. Ektefaie, W. Yuan, D. A. Dillon, and N. U. Lin, “Integrative multiomics-histopathology analysis for breast cancer classi fi cation,” npj Breast Cancer, vol. 7, Art. no. 147, Nov. 2021, pp. 1–6, doi: 10.1038/s41523-021-00357-y

    work page doi:10.1038/s41523-021-00357-y 2021

  20. [20]

    Predicting Breast Cancer Gene Expression Signature by Applying Deep Convolutional Neural Networks From Unannotated Pathological Images,

    N. N. Phan, C. Huang, L. Tseng, and E. Y. Chuang, “Predicting Breast Cancer Gene Expression Signature by Applying Deep Convolutional Neural Networks From Unannotated Pathological Images,” vol. 11, no. December 2021, pp. 1–11, 2021, doi: 10.3389/fonc.2021.769447

    work page doi:10.3389/fonc.2021.769447 2021

  21. [21]

    DEEP LEARNING -BASED CLASSIFICATION OF BREAST CANCER MOLECULAR SUBTYPES FROM H & E WHOLE- SLIDE IMAGES,

    M. Tafavvoghi, A. Sildnes, M. Rakaee, N. Shvetsov, and L. Ailo, “DEEP LEARNING -BASED CLASSIFICATION OF BREAST CANCER MOLECULAR SUBTYPES FROM H & E WHOLE- SLIDE IMAGES,” J. Pathol. Informat., vol. 16, p. 100410, Jan. 2025, doi: 10.1016/j.jpi.2024.100410

    work page doi:10.1016/j.jpi.2024.100410 2025

  22. [22]

    Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole Slide Images.,

    N. C. Kurian, P. H. Gann, N. Kumar, S. M. McGregor, R. Verma, and A. Sethi, “Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole Slide Images.,” Cancer Res. Commun. , vol. 5, no. January, pp. 157 –166, 2024, doi: 10.1158/2767-9764.CRC-24-0397

    work page doi:10.1158/2767-9764.crc-24-0397 2024

  23. [23]

    Image analysis with deep learning to predict breast cancer grade, ER status, histologic subtype, and intrinsic subtype,

    H. D. Couture et al., “Image analysis with deep learning to predict breast cancer grade, ER status, histologic subtype, and intrinsic subtype,” npj Breast Cancer , vol. 4, no. 1, 2018, doi: 10.1038/s41523-018-0079-1

    work page doi:10.1038/s41523-018-0079-1 2018

  24. [24]

    slide images,

    G. Campanella et al., “slide images,” Nat. Med., vol. 25, no. August, 2019, doi: 10.1038/s41591 - 21 019-0508-1

    work page doi:10.1038/s41591 2019

  25. [25]

    Overcoming the limitations of patch ‑ based learning to detect cancer in whole slide images,

    O. Ciga, T. Xu, S. N. Mozes, S. Noy, F. I. Lu, and A. L. Martel, “Overcoming the limitations of patch ‑ based learning to detect cancer in whole slide images,” Sci. Rep. , pp. 1 –10, 2021, doi: 10.1038/s41598-021-88494-z

    work page doi:10.1038/s41598-021-88494-z 2021

  26. [26]

    Multiple Instance Learning with random sampling for Whole Slide Image Classification

    H. Keshvarikhojasteh, J. P. W. Pluim, and M. Veta, “Multiple Instance Learning with random sampling for Whole Slide Image Classification” in Proc. SPIE 12933, Digital and Computational Pathology, 2024, Art. no. 129331J, doi: 10.1117/12.3004713

    work page doi:10.1117/12.3004713 2024

  27. [27]

    Creating an atlas of normal tissue for pruning WSI patching through anomaly detection,

    P. Nejat et al. , “Creating an atlas of normal tissue for pruning WSI patching through anomaly detection,” Sci. Rep., pp. 1–15, 2024, doi: 10.1038/s41598-024-54489-9

    work page doi:10.1038/s41598-024-54489-9 2024

  28. [28]

    Whole Slide Images in Artificial Intelligence Applications in Digital Pathology : Challenges and Pitfalls,

    K. Basak, “Whole Slide Images in Artificial Intelligence Applications in Digital Pathology : Challenges and Pitfalls,” pp. 101–108, 2023, doi: 10.5146/tjpath.2023.01601.[29] M. Gadermayr and M. Tschuchnig, “Computerized Medical Imaging and Graphics Multiple instance learning for digital pathology : A review of the state -of-the-art , limitations & future ...

    work page doi:10.5146/tjpath.2023.01601 2023

  29. [29]

    Zheng et al

    K. Zheng et al. , “Article Deep learning model with pathological knowledge for detection of colorectal neuroendocrine tumor ll ll Deep learning model with pathological knowledge for detection of colorectal neuroendocrine tumor,” Cell Reports Med., vol. 5, no. 10, p. 101785, 2024, doi: 10.1016/j.xcrm.2024.101785

    work page doi:10.1016/j.xcrm.2024.101785 2024

  30. [30]

    Dynamic Policy -Driven Adaptive Multi -Instance Learning for Whole Slide Image Classification,

    T. Zheng, K. Jiang, and H. Yao, “Dynamic Policy -Driven Adaptive Multi -Instance Learning for Whole Slide Image Classification,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2024, pp. 8028–8037., doi:10.1109/CVPR56263.2024.ABC123

    work page doi:10.1109/cvpr56263.2024.abc123 2024

  31. [31]

    Computerized Medical Imaging and Graphics Dual attention model with reinforcement learning for classification of histology whole-slide images,

    M. Raza, R. Awan, R. Muhammad, S. Bashir, T. Qaiser, and N. M. Rajpoot, “Computerized Medical Imaging and Graphics Dual attention model with reinforcement learning for classification of histology whole-slide images,” Comput. Med. Imaging Graph., vol. 118, no. p. 102466, July 2024, doi: 10.1016/j.compmedimag.2024.102466

    work page doi:10.1016/j.compmedimag.2024.102466 2024

  32. [32]

    Adaptive patch selection to improve Vision Transformers through Reinforcement Learning,

    F. Cauteruccio, M. Marchetti, D. Traini, D. Ursino, and L. Virgili, “Adaptive patch selection to improve Vision Transformers through Reinforcement Learning,” Appl. Intell, vol. 123, 2025, doi: 10.1007/s10489-025-06516-z

    work page doi:10.1007/s10489-025-06516-z 2025

  33. [33]

    Meta-Radiology A review of uncertainty estimation and its application in medical imaging,

    K. Zou, Z. Chen, X. Yuan, X. Shen, M. Wang, and H. Fu, “Meta-Radiology A review of uncertainty estimation and its application in medical imaging,” Meta-Radiology, vol. 1, no. 1, p. 100003, 2023, doi: 10.1016/j.metrad.2023.100003

    work page doi:10.1016/j.metrad.2023.100003 2023

  34. [34]

    Generalisation effects of predictive uncertainty estimation in deep learning for digital pathology,

    M. Pocevičiūtė, G. Eilertsen, S. Jarkman, and C. Lundström, “Generalisation effects of predictive uncertainty estimation in deep learning for digital pathology,” Sci. Rep. , pp. 1 –15, 2022, doi: 10.1038/s41598-022-11826-0

    work page doi:10.1038/s41598-022-11826-0 2022

  35. [35]

    A Probabilistic U-Net for Segmentation of Ambiguous Images,

    S. A. A. Kohl et al., “A Probabilistic U-Net for Segmentation of Ambiguous Images,” no. NeurIPS, pp. 1–11, 2018

    work page 2018

  36. [36]

    Learning Confidence for Out -of-Distribution Detection in Neural Networks,

    T. Devries and G. W. Taylor, “Learning Confidence for Out -of-Distribution Detection in Neural Networks,” 2018, doi:10.48550/arXiv.1802.04865

    work page doi:10.48550/arxiv.1802.04865 2018

  37. [37]

    Bayesian dropout,

    T. Herlau, M. N. Schmidt, and M. Mørup, “Bayesian dropout,” Procedia Comput. Sci., vol. 201, pp. 771–776, 2022, doi: 10.1016/j.procs.2022.03.105

    work page doi:10.1016/j.procs.2022.03.105 2022

  38. [38]

    The emerging landscape of spatial profiling technologies,

    J. R. Moffitt, E. Lundberg, and H. Heyn, “The emerging landscape of spatial profiling technologies,” 22 Nat. Rev. Genet., vol. 23, no. 12, pp. 741–759, 2022, doi: 10.1038/s41576-022-00515-3

    work page doi:10.1038/s41576-022-00515-3 2022

  39. [39]

    Deep learning infers clinically relevant protein levels and drug response in breast cancer from unannotated pathology images,

    H. Liu, X. Xie, and B. Wang, “Deep learning infers clinically relevant protein levels and drug response in breast cancer from unannotated pathology images,” npj Breast Cancer, vol. 10, no. 1, pp. 1–11, 2024, doi: 10.1038/s41523-024-00620-y

    work page doi:10.1038/s41523-024-00620-y 2024

  40. [40]

    Molecular intrinsic versus clinical subtyping in breast cancer : A comprehensive review,

    A. Szymiczek, “Molecular intrinsic versus clinical subtyping in breast cancer : A comprehensive review,” no. November 2020, pp. 613–637, 2021, doi: 10.1111/cge.13900

    work page doi:10.1111/cge.13900 2020

  41. [41]

    Reporting Hormone Receptor Expression in Breast Carcinomas: Which Method has the Highest Prognostic Power and What Should be the Optimal Cut -off Value?,

    Ç. Öztürk, O. Okcu, S. D. Öztürk, B. Şen, A. E. Öztürk, and R. Bedir, “Reporting Hormone Receptor Expression in Breast Carcinomas: Which Method has the Highest Prognostic Power and What Should be the Optimal Cut -off Value?,” Int. J. Surg. Pathol. , 2024, doi: 10.1177/10668969241265068

    work page doi:10.1177/10668969241265068 2024

  42. [42]

    Interplay of nuclear receptors ( ER , PR , and GR ) and their steroid hormones in MCF-7 cells,

    S. M. Hegde and M. N. Kumar, “Interplay of nuclear receptors ( ER , PR , and GR ) and their steroid hormones in MCF-7 cells,” 2016, doi: 10.1007/s11010-016-2810-2

    work page doi:10.1007/s11010-016-2810-2 2016

  43. [43]

    Looking beyond the ER, PR, and HER2: what’s new in the ARsenal for combating breast cancer?,

    T. P. Srivastava, R. Dhar, and S. Karmakar, “Looking beyond the ER, PR, and HER2: what’s new in the ARsenal for combating breast cancer?,” Reprod. Biol. Endocrinol., vol. 23, no. 1, p. 9, 2025, doi: 10.1186/s12958-024-01338-z

    work page doi:10.1186/s12958-024-01338-z 2025

  44. [44]

    Meta-Analysis of HER2-Enriched Subtype Predicting the Pathological Complete Response Within HER2-Positive Breast Cancer in Patients Who Received Neoadjuvant Treatment,

    G. Shen et al., “Meta-Analysis of HER2-Enriched Subtype Predicting the Pathological Complete Response Within HER2-Positive Breast Cancer in Patients Who Received Neoadjuvant Treatment,” Front. Oncol., vol. 11, no. July, pp. 1–13, 2021, doi: 10.3389/fonc.2021.632357

    work page doi:10.3389/fonc.2021.632357 2021

  45. [45]

    From whole -slide image to biomarker prediction: end -to-end weakly supervised deep learning in computational pathology,

    O. S. M. El Nahhas et al., “From whole -slide image to biomarker prediction: end -to-end weakly supervised deep learning in computational pathology,” Nat. Protoc. 2024, vol. 20, no. January, pp. 1–24, 2024, doi: 10.1038/s41596-024-01047-2

    work page doi:10.1038/s41596-024-01047-2 2024

  46. [46]

    Intensity Normalization Techniques and Their Effect on the Robustness and Predictive Power of Breast MRI Radiomics,

    F. Schwarzhans et al., “Intensity Normalization Techniques and Their Effect on the Robustness and Predictive Power of Breast MRI Radiomics,” Eur. J. Radiol., vol. 187, no. August 2024, p. 112086, 2024, doi: 10.1016/j.ejrad.2025.112086

    work page doi:10.1016/j.ejrad.2025.112086 2024

  47. [47]

    Uncertainty quantification of geochemical data imputation using Monte Carlo dropout,

    V. Puzyrev and P. Duuring, “Uncertainty quantification of geochemical data imputation using Monte Carlo dropout,” J. Geochemical Explor. , vol. 272, no. January, p. 107695, 2025, doi: 10.1016/j.gexplo.2025.107695

    work page doi:10.1016/j.gexplo.2025.107695 2025

  48. [48]

    Adaptive Sampling to Reduce Epistemic Uncertainty Using Prediction Interval-Generation Neural Networks,

    G. Morales and J. W. Sheppard, “Adaptive Sampling to Reduce Epistemic Uncertainty Using Prediction Interval-Generation Neural Networks,” Proc. AAAI Conf. Artif. Intell. , vol. 39, no. 18, pp. 19563–19571, 2025, doi: 10.1609/aaai.v39i18.34154

    work page doi:10.1609/aaai.v39i18.34154 2025

  49. [49]

    Learning generalizable AI models for multi -center histopathology image classification,

    M. Asadi -Aghbolaghi et al., “Learning generalizable AI models for multi -center histopathology image classification,” npj Precis. Oncol., vol. 8, no. 1, 2024, doi: 10.1038/s41698-024-00652-4

    work page doi:10.1038/s41698-024-00652-4 2024

  50. [50]

    An efficient multi -objective parrot optimizer for global and engineering optimization problems,

    M. R. Saad, M. M. Emam, and E. H. Houssein, “An efficient multi -objective parrot optimizer for global and engineering optimization problems,” Sci. Rep. , vol. 15, no. 1, pp. 1 –33, 2025, doi: 10.1038/s41598-025-88740-8

    work page doi:10.1038/s41598-025-88740-8 2025

  51. [51]

    An efficient method for detection of Alzheimer ’ s disease using high-dimensional PET scan images,

    A. Borji, A. Seifi, and T. H. Hejazi, “An efficient method for detection of Alzheimer ’ s disease using high-dimensional PET scan images,” vol. 17, pp. 729–749, 2023, doi: 10.3233/IDT-220315

    work page doi:10.3233/idt-220315 2023

  52. [52]

    Weakly - supervised diagnosis and detection of breast cancer using deep multiple instance learning,

    P. Diogo, M. Morais, F. M. Calisto, C. Santiago, C. Aleluia, and J. C. Nascimento, “Weakly - supervised diagnosis and detection of breast cancer using deep multiple instance learning,” in Proc. 23 IEEE 20th Int. Symp. Biomed. Imaging (ISBI), pp. 1 –4, 2023, doi: 10.1109/ISBI53787.2023.10230448

    work page doi:10.1109/isbi53787.2023.10230448 2023

  53. [53]

    A Hybrid Deep Learning Model for Predicting Molecular Subtypes of Human Breast Cancer Using Multimodal Data,

    T. Liu, J. Huang, T. Liao, R. Pu, S. Liu, and Y. Peng, “A Hybrid Deep Learning Model for Predicting Molecular Subtypes of Human Breast Cancer Using Multimodal Data,” Irbm, vol. 43, no. 1, pp. 62– 74, 2022, doi: 10.1016/j.irbm.2020.12.002

    work page doi:10.1016/j.irbm.2020.12.002 2022

  54. [54]

    Classification of Luminal Subtypes in Full Mammogram Images Using Transfer Learning,

    A. B. Panambur, P. Madhu, and A. Maier, “Classification of Luminal Subtypes in Full Mammogram Images Using Transfer Learning,” 2023, [Online]. Available: http://arxiv.org/abs/2301.09282

    work page arXiv 2023