You are currently viewing a new version of our website. To view the old version click .
MDPIMDPI
Search all articles
Diagnostics
Download PDF
  • Systematic Review
  • Open Access

30 April 2025

Advanced Deep Learning Approaches in Detection Technologies for Comprehensive Breast Cancer Assessment Based on WSIs: A Systematic Literature Review

,
,
and
1
Center for Artificial Intelligence Technology, Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Bangi 43600, Selangor, Malaysia
2
College of Intelligent Manufacturing, Zibo Vocational Institute, Zibo 255314, China
3
Department of Pathology, Faculty of Medicine, Universiti Kebangsaan Malaysia, Jalan Yaacob Latif, Bandar Tun Razak, Cheras 56000, Wilayah Persekutuan Kuala Lumpur, Malaysia
*
Author to whom correspondence should be addressed.
Diagnostics2025, 15(9), 1150;https://doi.org/10.3390/diagnostics15091150
This article belongs to the Special Issue Artificial Intelligence for Health and Medicine
Version Notes
Order Reprints

Abstract

Background: Breast cancer is one of the leading causes of death among women worldwide. Accurate early detection of lymphocytes and molecular biomarkers is essential for improving diagnostic precision and patient prognosis. Whole slide images (WSIs) are central to digital pathology workflows in breast cancer assessment. However, applying deep learning techniques to WSIs presents persistent challenges, including variability in image quality, limited availability of high-quality annotations, poor model interpretability, high computational demands, and suboptimal processing efficiency. Methods: This systematic review, guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), examines deep learning-based detection methods for breast cancer published between 2020 and 2024. The analysis includes 39 peer-reviewed studies and 20 widely used WSI datasets. Results: To enhance clinical relevance and guide model development, this study introduces a five-dimensional evaluation framework covering accuracy and performance, robustness and generalization, interpretability, computational efficiency, and annotation quality. The framework facilitates a balanced and clinically aligned assessment of both established methods and recent innovations. Conclusions: This review offers a comprehensive analysis and proposes a practical roadmap for addressing core challenges in WSI-based breast cancer detection. It fills a critical gap in the literature and provides actionable guidance for researchers, clinicians, and developers seeking to optimize and translate WSI-based technologies into clinical workflows for comprehensive breast cancer assessment.

1. Introduction

Breast cancer remains one of the most prevalent and deadly malignancies affecting women worldwide, posing a significant threat to both health outcomes and quality of life [,,]. Despite remarkable advances in medical diagnostics and therapeutic strategies, the molecular heterogeneity and diverse clinical manifestations of breast cancer continue to complicate its detection and management []. Biomarker and lymphocyte detection are critical components in breast cancer assessment, offering essential guidance for diagnosis, classification, and prognosis [].
Among these, the estrogen receptor (ER) serves as a key molecular marker for subtype classification and supports the development of personalized treatment regimens, directly influencing patient outcomes []. The proliferation index Ki-67 Antigen (Ki67) provides independent predictive value for treatment response, while the joint evaluation of ER and progesterone receptor (PR) further refines hormone therapy decisions []. Overexpression of human epidermal growth factor receptor 2 (HER2) correlates with increased tumor aggressiveness and recurrence risk, making HER2-positive cases suitable for targeted therapy. In addition, tumor-infiltrating lymphocytes (TILs) are considered reliable prognostic indicators, especially for triple-negative breast cancer (TNBC) patients, where TIL density correlates with both disease-free and overall survival []. Accurate identification of these biomarkers and immune features forms the foundation for effective clinical decision-making.
Recent advancements in whole-slide imaging (WSI) have enabled digital pathology to visualize entire tissue sections at ultra-high resolution, capturing both the spatial distribution of biomarkers and the microenvironmental context of lymphocytes [,]. These large-format digitized slides—typically prepared using Hematoxylin and Eosin (H&E) staining for structural visualization and Immunohistochemistry (IHC) for highlighting specific molecular targets—provide pathologists with richer information but also introduce significant technical challenges. The sheer volume of image data—often exceeding one billion pixels per slide—and the complex spatial patterns of diagnostic features render manual analysis time-consuming, subjective, and difficult to scale [,]. Traditional workflows depend heavily on expert interpretation, leading to potential variability and limited reproducibility in large-scale clinical studies.
To address these limitations, deep learning has emerged as a promising approach for automating WSI analysis. While segmentation tasks aim to label each pixel according to tissue type or structure [], detection-based techniques are particularly well-suited for clinical applications. They generate interpretable, quantitative outputs such as cell counts, biomarker localization, and lymphocyte spatial distributions, which align more directly with diagnostic workflows [,]. This review focuses on detection-oriented methods in WSI analysis and their applications across four critical clinical tasks: diagnosis, classification, grading, and prognosis.
Nonetheless, the integration of deep learning into WSI-based breast cancer assessment remains nontrivial. The ultra-high resolution of WSIs demands substantial computational resources and presents design challenges for conventional model architectures [,]. Furthermore, WSIs exhibit a multi-scale nature—ranging from microscopic cell morphology to macroscopic tissue organization—necessitating models capable of learning across spatial hierarchies []. The analysis is further complicated by sparse diagnostic features, inter-sample heterogeneity, and staining variations. Compounding these issues is the scarcity of large, well-annotated datasets, which limits the scalability of supervised learning approaches []. Interpretability also remains a major concern, as the opaque decision-making processes of deep learning models reduce clinical trust []. Effective detection algorithms must therefore balance sensitivity to subtle morphological cues with robustness against noise and artifacts—especially when identifying key features such as biomarker expression and lymphocyte infiltration.
Despite a growing body of literature in this domain, existing reviews often remain narrowly focused. Many concentrate on segmentation techniques or algorithmic performance metrics, without sufficiently addressing the clinical integration of detection methods or their practical utility [,,]. Additionally, critical issues such as dataset bias, computational burden, and feedback mechanisms from clinical deployment are frequently overlooked. This gap highlights the need for a more holistic and clinically grounded synthesis of detection-focused research in breast cancer WSIs.
To this end, this review aims to address three key research questions:
1.
What types of datasets are used for comprehensive breast cancer assessment using WSIs?
2.
What are the main challenges associated with comprehensive breast cancer assessment using WSIs?
3.
How do WSIs impact the accuracy and reliability of advanced deep learning approaches for comprehensive breast cancer assessment?
Figure 1 summarizes the task types, datasets, AI models, and evaluation criteria that frame this review of deep learning-based breast cancer assessment using WSIs. This study systematically reviews the landscape of relevant approaches published between 2020 and 2024, following PRISMA guidelines []. It synthesizes recent methodological advancements, identifies unresolved challenges, and outlines future research directions. In particular, we emphasize the clinical significance of detection tasks, the scalability of AI-based diagnostic tools, and the potential for deep learning to transform biomarker discovery and personalized treatment planning in digital pathology.
Figure 1. Overview of detection tasks, datasets, models, and evaluation criteria in WSI-based breast cancer assessment.

2. Methods

The study followed the PRISMA guideline [] in this systematic literature review. PRISMA provides a consistent, repeatable process for locating, assessing, and selecting pertinent research. It also provides guidance on how to choose, recognize, and evaluate studies [,].
Figure 2 shows the PRISMA procedure used for this systematic literature review. The next subsection provides details on the review process.
Figure 2. PRISMA flow diagram of the SLR.

2.1. Data Sources and Search Strategy

To identify relevant academic publications, 8 major bibliographic databases were used, including Scopus, IEEE Xplore Library, Web of Science, SpringerLink, ACM Digital Library, and ScienceDirect. Consequently, the systematic literature review focused on articles published between 2020 and 2024 to ensure the inclusion of the most recent and relevant research findings. The search strategy employed combinations of key terms like the following:
  • “breast cancer detection” AND “deep learning”
  • “breast cancer diagnosis” AND “deep learning”
  • “convolutional neural networks” AND “breast cancer”
  • “lymphocytes detection” OR “biomarkers detection” AND “breast cancer”
  • “H&E stained images” AND “deep learning” AND “breast cancer”
  • “immunohistochemistry” AND “deep learning” AND “breast cancer”
  • “automated breast cancer diagnosis” OR “AI in breast cancer screening”
These search parameters were created to encompass the depth of knowledge regarding deep learning applications in the detection and analysis of breast cancer.

2.2. Selection Criteria

An exhaustive search strategy combining both automated and manual methods initially retrieved 417 academic publications. Throughout the entire screening process, the inclusion and exclusion criteria outlined in Table 1 were applied consistently. After eliminating 254 duplicate records, 163 unique publications were screened based on their titles, abstracts, and keywords. At this stage, 115 papers were excluded due to a lack of domain relevance or methodological inadequacy.
Table 1. Inclusion and exclusion criteria.
Following this, 48 full-text articles were retrieved for in-depth evaluation. One article was excluded due to inaccessibility, and the remaining 47 underwent full eligibility assessment. Nine additional studies were excluded for the following reasons: three did not focus specifically on breast cancer or its primary clinical tasks (e.g., detection, classification, prognosis); three lacked concrete details regarding deep learning algorithms or implementation strategies; two relied solely on conventional pathology without computational approaches; and one failed to meet the minimum quality threshold on the Standard Quality Checklist (SCQ).
This systematic and criteria-driven selection process resulted in 39 high-quality studies being included in the final review. The characteristics of these selected publications are summarized in Table 2, providing a foundation for subsequent quality assessment and synthesis.
Table 2. Summary of included studies with evaluation and data scale (LD = lymphocyte detection; BD = biomarker detection; - = not specified in the article).

2.3. Quality Assessment

Assessing the quality of evidence is crucial in a systematic literature review (SLR), as methodological biases may influence outcomes and lead to misinterpretation. To ensure the reliability and rigor of the included studies, this review adopted the Standard SCQ proposed by [], which comprises ten evaluation items. Following the approach of [], only studies that provided a “yes” response to at least seven SCQ items were included.
Among the 39 studies selected after quality filtering, the SCQ score distribution was as follows: 10 studies received a full score (10/10), 15 studies scored 9/10, 11 studies scored 8/10, and 3 studies met the minimum threshold with 7/10. Studies scoring below this threshold were excluded to maintain methodological rigor. This distribution reflects the overall quality consistency of the selected literature and ensures that only robust and dependable studies were included in the final synthesis. The SCQ evaluation and data extraction processes were closely integrated to enhance the validity and significance of the review outcomes. Table 3 outlines the SCQ criteria used in this study.
Table 3. Checklist for quality assessment.

2.4. Data Extraction and Synthesis

The study noted the pertinent information from each study, such as the publisher, authors, and year of publication, and gathered data for our SLR on the deep learning techniques used, reported accuracy, and assessment criteria. To answer the study’s research questions, the authors particularly examined the results during the data synthesis phase. The authors used a variety of visualization approaches and tools, including tables and diagrams, to make this analysis easier.

3. Results and Meta-Analysis

The meta-analysis of the search results from our systematic literature review is shown in this section. It starts with a summary of the chosen articles before going into each of the study questions that were developed and stated in the introduction.

3.1. Overview of Selected Studies

Figure 3 provides a chronological overview of the articles selected for this SLR, detailing the number of publications related to advanced deep learning methods for breast cancer cell detection using digital pathology images from 2020 to 2024. The diagram indicates a growing trend in this research area over the past few years, particularly from 2023 to 2024, when the number of published articles significantly increased. Most of the articles considered for this study were published after 2023. Specifically, the highest number of articles published in a single year was 13 in 2023 and 9 in 2024, followed by 10 papers in 2021 and 4 papers in 2020. In contrast, only 2 papers were reviewed in 2022, which some literature attributes to the impact of the COVID-19 pandemic on research output.
Figure 3. Diagram showing the number of articles published per year from 2020 to 2024.

3.2. Research Question 1: What Types of Datasets Are Employed for Comprehensive Breast Cancer Assessment Using WSIs?

This research evaluates 20 key WSIs datasets pivotal to advancing deep learning applications in breast cancer detection, diagnosis, classification, grading, and prognosis. These datasets, each contributing uniquely to comprehensive breast cancer assessment, reflect the rapid evolution and diversification of digital pathology resources.
The following attributes define the state of the breast cancer WSI datasets:
1.
Scale and Origin:
  • The basic resources are large-scale public repositories such as TCGA-BRCA (3111 WSIs) [].
  • Focused, high-resolution data are available from specialized datasets like PanNuke (200,000 nuclei across 19 tissue types) and BACH (400 patches, 30 WSIs) [,].
  • For particular study goals, derivative datasets (MoNuSeg, BCSS, TIGER) expand on primary sources [].
2.
Research Focus and Annotation Granularity:
  • Datasets cover a wide range of anatomical structures, such as nuclei in MoNuSeg and tumor-infiltrating lymphocytes in PanopTILs, to specific cellular structures.
  • Increasing model sophistication is reflected in the variation in annotation detail from whole-slide to pixel-level [].
3.
Multi-modal Integration:
  • Datasets that integrate clinical and genetic information with histopathological images, such as TCGA-BRCA, provide opportunities for more thorough and comprehensive analysis [,].
4.
Ethical Considerations and Diversity:
  • More recent datasets, such as AI-TUMOR, highlight the diversity of patient demographics and the use of ethical data collection techniques [].
Together, these datasets enable a wide range of deep learning uses, ranging from simple cell identification to intricate tumor categorization and prognostic modeling. In addition to driving innovation in deep learning architectures suited to the particular difficulties of breast cancer analysis, they act as benchmarks for model development and validation [,].
These datasets development follows the field’s move toward more extensive, fully annotated resources on a wider scale. This pattern suggests a developing sector ready to use cutting-edge computational techniques to address the intricate, multidimensional character of breast cancer research, along with the integration of multimodal data and a growing emphasis on ethical issues.
Table 4 lists the various datasets and highlights their unique properties that were employed for WSI-based comprehensive breast cancer assessment. Each dataset plays a crucial role in advancing deep learning methodologies by providing diverse and detailed data for model training and validation.
Table 4. Key WSI datasets relevant to comprehensive breast cancer assessment.

3.3. Research Question 2: What Are the Main Challenges Associated with Comprehensive Breast Cancer Assessment Using WSIs?

WSI has emerged as a pivotal component in the realm of digital pathology, particularly for the application of deep learning methodologies in the detection of breast cancer []. Nonetheless, numerous substantial obstacles hinder its optimal implementation. These obstacles encompass a spectrum of technical challenges associated with image processing and analytical procedures, as well as overarching issues pertaining to data integrity, and annotation practices. Table 5 summarizes the main issues raised in recent research papers, together with their descriptions, implications for breast cancer detection, and relevant research references.
Table 5. Main challenges in WSIs for comprehensive breast cancer assessment.

3.4. Research Question 3: How Do WSIs Affect the Accuracy and Reliability of Advanced Deep Learning Approaches for Comprehensive Breast Cancer Assessment?

WSI plays a critical role in enhancing the accuracy and reliability of deep learning techniques for comprehensive assessments of breast cancer []. Deep learning algorithms can examine complicated tissue structures and cellular configurations with more ease thanks to the high-resolution and finely detailed features of WSIs. This is crucial for distinguishing between cancerous and benign cells. However, as Table 6 summarizes, several critical variables related to WSIs significantly impact the accuracy and reliability of deep learning models when it comes to breast cancer diagnosis.
Table 6. The impact of WSIs on the accuracy and reliability of deep learning methods for comprehensive breast cancer assessment.
As Table 6 shows, these factors are interrelated and impact model performance. Reliability in various circumstances is contingent upon constant image quality and standardization, whereas high resolution improves detection accuracy but raises processing demands. Strong annotations and a wide range of training data are essential for the resilience and generalizability of the model. WSI resolution makes it easier to obtain the necessary detail for accurate analytical evaluations, but it also creates problems with data processing and standardization methods. It is crucial to guarantee that WSIs are of superior quality, coherent, and comprehensively annotated, while concurrently tackling the intricacies linked to data heterogeneity and standardization since it is essential to the creation of reliable models that may be successfully applied in clinical settings.
In summary, WSIs have a major influence on the precision and dependability of deep learning techniques used in the detection and assessment of breast cancer. Developing deep learning models that are therapeutically useful requires addressing these variables simultaneously. In order to further improve the accuracy and dependability of deep learning techniques in breast cancer evaluation utilizing WSIs, future research should concentrate on maximizing these factors.

4. Criteria for Comprehensive Breast Cancer-Assessment WSI Algorithms

A number of important factors must be considered when evaluating detection criteria for a thorough assessment of breast cancer based on WSI algorithms to guarantee efficacy, consistency, and clinical applicability. These criteria can be divided into two factors: technological and clinical. Each of these factors is vital in establishing the technology’s overall value and suitability.
Technical Criteria:
1.
Accuracy and Performance Metrics: According to [], sensitivity, specificity, accuracy, recall, AUC, and F1 score are essential for accurately identifying cancer cells while reducing false positives.
2.
Robustness and Generalizability: Algorithms must manage common problems like noise and artifacts while operating consistently over a range of datasets, scanners, and staining processes [].
3.
Interpretability and Explainability: Clinical trust depends on model openness and error analysis capabilities [].
4.
Computational Efficiency: Algorithms should be appropriate for a range of computational contexts, with respectable processing speeds and efficient resource consumption [].
Clinical Criteria:
Annotation Quality and Requirements: Preference for algorithms that reduce reliance on resource-intensive annotations by performing well with minimum or semi-supervised learning [].
Together, these criteria ensure that the algorithms chosen are not only sound from a methodological standpoint but also practical, significant, and useful in clinical settings. Through a methodical approach to these components, scientists and medical professionals can evaluate and choose algorithms that are suitable for the complex problems involved in WSIs-based breast cancer cell detection, improving diagnostic accuracy and patient outcomes. To drive progress in breast cancer-detection techniques, researchers must prioritize optimizing baseline models while also carrying out thorough assessments of those models. These models serve as essential baselines to assess the effectiveness of WSI-based detection techniques in clinical settings. Targeted optimization techniques can significantly enhance important characteristics like clinical relevance, accuracy, and robustness, addressing the many problems that come with assessing breast cancer. This systematic approach guarantees that the chosen algorithms enhance patient outcomes and boost diagnostic accuracy.

4.1. Baseline Models for Detection Technologies Applied in Comprehensive Breast Cancer Assessment Based on WSI Algorithms

To better visualize the methodological progression in WSI-based breast cancer detection over recent years, Figure 4 presents a Sankey diagram capturing the dynamic interplay between publication year, detection task, and model architecture from 2020 to 2024. Each stream represents a flow of research attention, where the thickness of the connection reflects the frequency of model usage for specific tasks. As shown in the figure, two major trends emerge: first, a shift in detection focus—from early emphasis on biomarkers to increasing attention on lymphocyte detection, and more recently, to frameworks addressing both targets simultaneously; and second, a transition in model design—from dominant use of Convolutional Neural Network (CNN) and U-Net toward more sophisticated or hybrid approaches involving Transformers and Generative Adversarial Network (GAN), reflecting growing demands for richer spatial modeling and generalization across WSI domains.
Figure 4. Evolution of baseline-detection models for breast cancer assessment in WSI from 2020 to 2024. (Flow width indicates the relative usage frequency of each model type, with thicker streams showing more widely adopted models. Different colors represent distinct model architectures tracked across the study period).
These evolving trends are closely reflected in the selection and adaptation of baseline model architectures for different detection tasks. Over the five-year span, CNN and U-Net remained dominant. CNNs were predominantly applied to biomarkers detection, where classification or sparse detection was needed to identify ER/PR/HER2-positive cells. In contrast, U-Net architectures were widely adopted for lymphocyte detection, due to their pixel-level precision and strong segmentation capabilities—critical for detecting densely distributed immune cells with indistinct boundaries. Notably, in tasks combining biomarkers and lymphocyte detection, U-Net-based frameworks were still preferred for their ability to support multi-task outputs, such as simultaneous localization and segmentation.
From 2022 onward, architectural diversification accelerated. Transformer-based models gained traction, especially in biomarker detection, by leveraging self-attention mechanisms to capture long-range contextual dependencies in high-resolution WSI data. Hybrid approaches, such as CNN+Transformer and GAN+CNN+U-Net, emerged around 2023, integrating the spatial locality of CNNs, the generative robustness of GANs, and the global modeling power of Transformers—enabling more adaptive and domain-generalizable detection systems. Meanwhile, lighter or exploratory models like Multilayer Perceptron (MLP), Multiple Instance Learning (MIL), and You Only Look Once (YOLO) appeared after 2021, though their use remained limited due to challenges in dense detection and precise localization on WSIs.
Evaluation metrics across these models varied with task and output granularity. Classification and sparse detection tasks typically employed AUC, accuracy, and F1-score, while segmentation-oriented models were assessed using Dice coefficient, IoU, and boundary-aware metrics. For multi-output detection models, task-specific metrics were reported independently, reflecting the complexity of comprehensive breast cancer assessment.
Overall, segmentation-centric models have remained the backbone of WSI-based detection. Their ability to handle high-resolution images through patch-based processing, preserve pixel-level detail, and support dense prediction tasks makes them especially suited for WSI applications, where targets are often overlapping, small, and structurally complex. The increasing use of hybrid and Transformer-based models reflects a broader trend toward unifying global and local representation learning for improved clinical utility.

4.2. Optimizing and Improving Existing Baselines Based on Evaluation Criteria

A multidisciplinary approach is required to optimize current baseline models for breast cancer cell detection using WSI, considering technological and clinical factors. Several important categories can be used to group this optimization process:

4.2.1. Enhancing Model Performance

In the realm of deep learning-based detection techniques for computer-aided diagnosis (CAD) of breast cancer, strategies to enhance model performance have become increasingly diverse and sophisticated. Primarily, ensemble learning frameworks effectively improve detection accuracy and generalization by integrating various deep learning architectures such as U-Net, GANs, and CNNs [,,,,,,,,,,,,,,,,]. Additionally, the application of multi-task learning paradigms, which allow models to jointly learn multiple related tasks such as lesion segmentation, classification, and malignancy grading within a unified architecture, has gained attention for its ability to leverage shared representations, reduce overfitting, and improve generalization across heterogeneous lesions. This approach not only provides more comprehensive pathological information but also enhances the model’s robustness and diagnostic performance [,,,,,]. Furthermore, multimodal data fusion strategies integrate genomic data with WSIs [,], or combine multi-level data from cellular and tissue levels [,,], significantly boosting diagnostic precision and interpretability by capturing complementary biological features. Training on cross-institutional, multi-center datasets enhances the model’s domain adaptability, mitigating the adverse effects of data distribution shifts on performance [].
Regarding model architectures, increasing network depth equips models with stronger feature extraction capabilities, enabling the capture of more complex pathomorphological features []. The introduction of diverse convolutional modules, such as residual convolutional blocks [], parallel convolutional blocks [], dilated convolutional blocks [], and color deconvolution [], further enhances the model’s feature representation ability and multi-scale information capture. Recently, the integration of multiple attention mechanisms, including combinations of spatial attention, channel attention, and self-attention, has gained widespread application in medical image analysis, improving detection precision and efficiency by enhancing the model’s focus on key regions and features.
Lastly, post-processing optimization techniques, including morphological opening operations, watershed algorithms [,], and advanced methods like HoVer-Net [], further elevate the precision and consistency of detection outcomes, particularly in cell instance segmentation and separation of adhered structures. The synergistic effect of these methods is driving the gradual implementation and application of breast cancer CAD systems in clinical practice.
While recent advances have significantly improved detection accuracy, they often come at the expense of interpretability and computational efficiency. In deep learning-based CAD systems, increasing model complexity to boost performance tends to reduce transparency and raises deployment barriers. For example, models like CB-HVT Net, which integrate PVT, ResNet variants, and attention mechanisms, feature high parameter counts and computational demands that hinder real-time deployment [,]. These challenges highlight the need for balanced approaches that maintain strong accuracy while enhancing explainability and deployability.

4.2.2. Improving Robustness and Generalizability

In the realm of deep learning-based breast cancer-detection research, robustness enhancement strategies predominantly converge on two pivotal methodologies. Cross-disease data integration and validation, referring to the inclusion of datasets from multiple cancer types to expose the model to a broader range of pathological variations, substantially augments model robustness through the incorporation of multi-disease datasets and the execution of pan-cancer experiments [,,,,,]. This approach facilitates the model’s acquisition of heterogeneous pathological features, thereby enhancing its discriminative capacity across diverse cancer types and elevating its adaptability within complex clinical milieus. Concurrently, the implementation of semi-supervised learning paradigms efficaciously addresses the constraints imposed by the paucity of annotated data []. By synergistically leveraging limited labeled datasets in conjunction with voluminous unlabeled data, this technique not only mitigates the reliance on extensive manual annotation processes but also markedly amplifies the model’s generalization prowess in data-constrained environments.
However, while cross-disease integration can enhance the model’s adaptability to diverse histopathological morphologies, it may also introduce label noise and inconsistencies in annotation standards. This, in turn, can lead to domain shift and reduce detection specificity for certain cancer types, such as breast cancer. For example, when independent models were constructed for five cancer types in TCGA and pan-cancer training was applied, the performance of pan-cancer models declined in certain tasks (e.g., PD-L1 prediction in STAD), suggesting that disease-specific signals may be diluted in a multi-cancer setting []. Moreover, many semi-supervised frameworks rely on heuristic thresholds or consistency regularization strategies, which often require task-specific tuning, thereby limiting their generalizability and scalability in real-world clinical applications.

4.2.3. Increasing Interpretability and Explainability

Interpretability techniques in deep learning have emerged as essential components for fostering trust, transparency, and clinical acceptability in breast cancer assessment. The Human-Interpretable Features (HIF) paradigm aims to bridge the semantic gap between model outputs and clinical understanding by aligning predictions with visually and diagnostically meaningful image features [], whereas saliency-based visualization methods offer intuitive heatmaps that localize regions contributing most to the model’s decision-making process, thereby enhancing interpretability for end-users []. While the integration of such methods has demonstrably improved the transparency and perceived reliability of AI-assisted diagnostic systems, several limitations persist. HIF-based strategies typically rely on predefined, handcrafted feature sets, which may insufficiently capture the complex and abstract representations encoded by deep neural networks, thus constraining their explanatory power and generalizability across datasets or imaging modalities. In parallel, saliency visualizations are often susceptible to input perturbations and architectural variations, producing unstable and sometimes misleading attributions. Moreover, the post hoc nature of most interpretability tools, coupled with a lack of standardized validation protocols, raises concerns regarding their clinical robustness and reproducibility. These limitations underscore an urgent need for the development of principled, rigorously evaluated interpretability frameworks that can yield consistent, meaningful, and clinically actionable explanations.

4.2.4. Optimizing Computational Efficiency

In research applying deep learning techniques to breast cancer assessment, various strategies have been employed to enhance model efficiency. The primary approach involves precise localization of ROI through methods such as Gaussian kernel annotation [] and micro-block selection techniques [,], enabling models to focus on key pathological features. This not only improves accuracy but also reduces computational costs. Additionally, researchers have utilized pre-training strategies to optimize model architecture [,], accelerating model convergence, improving task initialization, and reducing parameter count and architectural complexity, thereby lowering computational demands while maintaining high performance. However, these efficiency-oriented strategies are not without limitations. ROI localization methods often rely on heuristic rules or expert-defined annotations, which may introduce bias and reduce scalability across datasets with varying staining characteristics or image resolutions. Moreover, such narrowly focused techniques risk omitting relevant contextual cues essential for accurate diagnosis. Similarly, the effectiveness of pre-training heavily depends on the relevance and quality of the source domain; mismatched pre-training can result in suboptimal initialization and diminished downstream performance. Furthermore, existing studies rarely provide systematic evaluations of the trade-offs between architectural simplification and diagnostic accuracy, leaving the optimal balance between efficiency and performance largely unexplored.

4.2.5. Addressing Data Quality and Annotation Challenges

In deep learning-based breast cancer-detection research, data quality and annotation challenges are primarily mitigated through two methodologies: weak supervision learning and segmentation map generation. Weak supervision enables models to extract features from limited or imprecise annotations, reducing dependence on fully annotated datasets [,,]. Segmentation map generation techniques create synthetic annotations, providing finer-grained information on regions of interest, thus compensating for incomplete or noisy labels []. These approaches synergistically enhance model robustness and performance in the face of data quality and annotation challenges in breast cancer-detection tasks. However, both methods present notable limitations. Weakly supervised models are vulnerable to label noise and may overfit to coarse annotations, limiting their generalizability. Meanwhile, segmentation map generation often relies on heuristic or rule-based pseudo-labels that may introduce bias, particularly in complex tumor microenvironments. The lack of standardized validation procedures for these synthetic annotations also raises concerns regarding their clinical reliability and reproducibility.
Table 7 concisely summarizes the baseline models and their optimization strategies, aiding in the understanding of how these models can be improved to enhance the efficiency of comprehensive breast cancer assessment based on detection technologies.
Table 7. Baseline models for WSI detection technologies and the strategies for optimizing and improving these models based on specific criteria.

5. Discussion and Potential Solutions for Improving WSIs for Breast Cancer Cell Detection

The application of advanced deep learning methodologies to WSIs for comprehensive breast cancer assessment represents a significant advancement in digital pathology. A systematic analysis of WSI datasets, associated challenges, and their impact on deep learning models has yielded five critical evaluation metrics: model performance, data integration and preprocessing, architectural optimization, robustness and generalizability, and interpretability and clinical application. This analytical framework provides a comprehensive evaluation of current technologies while illuminating future research trajectories.
The diversity of WSIs datasets, ranging from large-scale public repositories such as TCGA-BRCA to specialized high-resolution datasets like PanNuke and BACH, significantly influences model performance and data integration strategies. The multimodal nature of datasets such as TCGA-BRCA offers a robust foundation for developing comprehensive assessment models. However, dataset heterogeneity presents substantial challenges, particularly in terms of model generalizability. Innovative approaches have emerged to address these challenges. Notable among these is the integration of U-Net, GANs, and CNNs [], which demonstrates exceptional performance in handling diverse data. Additionally, the multi-task learning paradigm described in [] enhances model adaptability to varied WSIs data types through simultaneous segmentation, classification, and grading tasks.
The primary challenges in WSIs applications—including image size, quality variability, and annotation difficulties—directly impact data preprocessing strategies and model architecture design. These challenges have catalyzed innovative solutions. The panoramic segmentation method proposed by Liu et al. effectively addresses large-sized WSIs processing, markedly improving computational efficiency. In data integration and preprocessing, [] showcases a multimodal fusion strategy combining genomic data with WSIs, enhancing diagnostic accuracy while partially mitigating data heterogeneity issues. Nevertheless, the efficient processing of large-scale WSIs data remains a significant challenge, necessitating further research into advanced preprocessing techniques.
The impact of WSIs on deep learning model accuracy and reliability reveals a dichotomy: high-resolution data enhances detection precision while simultaneously presenting computational and standardization challenges. This contradiction has driven innovations in model architectures. The introduction of residual convolutional blocks [] significantly enhances feature extraction capabilities, while the attention mechanism described in [] improves the model’s capacity to identify key regions. These optimizations directly address the computational challenges posed by high-resolution WSIs while enhancing the model’s ability to capture complex pathological features.
Regarding robustness and generalizability, cross-disease dataset integration [] and semi-supervised learning methods [] demonstrate significant advantages in managing diverse WSIs datasets and limited annotated data. These approaches enhance model adaptability across varied clinical settings and effectively address the challenge of WSIs data annotation difficulties.
Interpretability and clinical application remain significant hurdles for WSI-based deep learning models. HIF methods [] and saliency visualization techniques [] have advanced the interpretability of model outputs, crucial for fostering clinical trust and adoption. However, the seamless integration of these technologies into clinical workflows requires further investigation. The identified evaluation metrics not only provide a comprehensive assessment of existing models but also delineate key directions for future research. Significant opportunities persist in addressing WSIs dataset diversity, processing large-scale high-resolution data, and improving model robustness and interpretability.

6. Conclusions, Implication, and Recommendations for Future Research

This review has systematically examined deep learning-based detection methods for breast cancer using WSIs, highlighting substantial progress in model accuracy, robustness, and data efficiency. Nonetheless, significant challenges remain in computational scalability, interpretability, annotation quality, and clinical applicability. To address these issues, future research should be guided by both technical feasibility and clinical relevance. In the short term, efforts should focus on developing lightweight, interpretable architectures optimized for WSI-scale processing to support real-time, resource-aware deployment. Enhancing weakly supervised and semi-supervised learning frameworks through uncertainty modeling and confidence-guided label refinement represents a technically viable strategy for improving annotation robustness. Medium-term priorities include designing domain-adaptive and resolution-consistent models to address data heterogeneity across institutions and staining variations. In the long term, the establishment of clinically validated interpretability protocols and the construction of large-scale, standardized WSI datasets should be pursued to support reproducibility, benchmarking, and translational impact. Prioritizing these directions will facilitate a more effective alignment between algorithmic innovation and real-world clinical integration, advancing the role of AI in precision breast cancer diagnostics.

Author Contributions

Writing—original draft preparation and editing, Q.X.; Supervision, review, and editing, A.A. (Afzan Adam) and A.A. (Azizi Abdullah); Medical terminology and case validation, N.B. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Ministry of Higher Education Malaysia through the Fundamental Research Grant Scheme (FRGS), grant code: FRGS/1/2024/ICT02/UKM/02/5.

Institutional Review Board Statement

Not applicable.

Data Availability Statement

Not applicable.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
PRISMPreferred Reporting Items for Systematic Reviews and Meta-Analyses
WSIWhole Slide Image
WSIsWhole Slide Images
EREstrogen Receptor
PRrogesterone Receptor
HER2Human Epidermal Growth Factor Receptor 2
Ki-67Ki-67 Antigen
TILTumor-Infiltrating Lymphocyte
H&EHematoxylin and Eosin
IHCImmunohistochemistry
TNBCTriple-Negative Breast Cancer
LDLymphocyte Detection
BDBiomarker Detection
SCQSystematic Quality Criteria
SLRSystematic Literature Review
HPFHigh-Power Field
CNNConvolutional Neural Network
GANGenerative Adversarial Network
MLPMultilayer Perceptron
MILMultiple Instance Learning
YOLOYou Only Look Once
CADComputer-Aided Diagnosis
ROIRegion of Interest
HIFHuman-Interpretable Features

References

  1. Faroughi, F.; Fathnezhad-Kazemi, A.; Sarbakhsh, P. Factors Affecting Quality of Life in Women With Breast Cancer: A Path Analysis. BMC Women’s Health 2023, 23, 578. [Google Scholar]
  2. Khairi, S.S.M.; Bakar, M.A.A.; Alias, M.A.; Bakar, S.A.; Liong, C.Y.; Rosli, N.; Farid, M. Deep learning on histopathology images for breast cancer classification: A bibliometric analysis. Healthcare 2021, 10, 10. [Google Scholar] [CrossRef] [PubMed]
  3. Saifullah, S.; Dreżewski, R. Enhancing breast cancer diagnosis: A CNN-based approach for medical image segmentation and classification. In International Conference on Computational Science; Springer: Berlin/Heidelberg, Germany, 2024; pp. 155–162. [Google Scholar]
  4. Mahmoud, R.; Ordóñez-Morán, P.; Allegrucci, C. Challenges for triple negative breast cancer treatment: Defeating heterogeneity and cancer stemness. Cancers 2022, 14, 4280. [Google Scholar] [CrossRef]
  5. El Bairi, K.; Haynes, H.R.; Blackley, E.; Fineberg, S.; Shear, J.; Turner, S.; De Freitas, J.R.; Sur, D.; Amendola, L.C.; Gharib, M.; et al. The tale of TILs in breast cancer: A report from the international immuno-oncology biomarker working group. NPJ Breast Cancer 2021, 7, 150. [Google Scholar] [PubMed]
  6. Pauzi, S.H.M.; Saari, H.N.; Roslan, M.R.; Azman, S.N.S.S.K.; Tauan, I.S.; Rusli, F.A.M.; Aizuddin, A.N. A comparison study of HER2 protein overexpression and its gene status in breast cancer. Malays. J. Pathol. 2019, 41, 133–138. [Google Scholar]
  7. Finkelman, B.S.; Zhang, H.; Hicks, D.G.; Turner, B.M. The evolution of Ki-67 and breast carcinoma: Past observations, present directions, and future considerations. Cancers 2023, 15, 808. [Google Scholar] [CrossRef]
  8. Zahari, S.; Syafruddin, S.E.; Mohtar, M.A. Impact of the Cancer Cell Secretome in driving breast Cancer progression. Cancers 2023, 15, 2653. [Google Scholar] [CrossRef]
  9. Gupta, R.; Le, H.; Van Arnam, J.; Belinsky, D.; Hasan, M.; Samaras, D.; Kurc, T.; Saltz, J.H. Characterizing immune responses in whole slide images of cancer with digital pathology and pathomics. Curr. Pathobiol. Rep. 2020, 8, 133–148. [Google Scholar]
  10. Roostee, S.; Ehinger, D.; Jönsson, M.; Phung, B.; Jönsson, G.; Sjödahl, G.; Staaf, J.; Aine, M. Tumour immune characterisation of primary triple-negative breast cancer using automated image quantification of immunohistochemistry-stained immune cells. Sci. Rep. 2024, 14, 21417. [Google Scholar] [CrossRef]
  11. Liu, Y.; Han, D.; Parwani, A.V.; Li, Z. Applications of artificial intelligence in breast pathology. Arch. Pathol. Lab. Med. 2023, 147, 1003–1013. [Google Scholar]
  12. Shakhawat, H.; Hossain, S.; Kabir, A.; Mahmud, S.H.; Islam, M.M.; Tariq, F. Review of artifact detection methods for automated analysis and diagnosis in digital pathology. In Artificial Intelligence for Disease Diagnosis and Prognosis in Smart Healthcare; CRC Press: Boca Raton, FL, USA, 2023; pp. 177–202. [Google Scholar]
  13. Li, X.; Li, C.; Rahaman, M.M.; Sun, H.; Li, X.; Wu, J.; Yao, Y.; Grzegorzek, M. A comprehensive review of computer-aided whole-slide image analysis: From datasets to feature extraction, segmentation, classification and detection approaches. Artif. Intell. Rev. 2022, 55, 4809–4878. [Google Scholar] [CrossRef]
  14. Wu, Y.; Cheng, M.; Huang, S.; Pei, Z.; Zuo, Y.; Liu, J.; Yang, K.; Zhu, Q.; Zhang, J.; Hong, H.; et al. Recent advances of deep learning for computational histopathology: Principles and applications. Cancers 2022, 14, 1199. [Google Scholar] [CrossRef] [PubMed]
  15. Carriero, A.; Groenhoff, L.; Vologina, E.; Basile, P.; Albera, M. Deep Learning in Breast Cancer Imaging: State of the Art and Recent Advancements in Early 2024. Diagnostics 2024, 14, 848. [Google Scholar] [CrossRef] [PubMed]
  16. Duggento, A.; Conti, A.; Mauriello, A.; Guerrisi, M.; Toschi, N. Deep computational pathology in breast cancer. Semin. Cancer Biol. 2021, 72, 226–237. [Google Scholar] [CrossRef]
  17. Luo, L.; Wang, X.; Lin, Y.; Ma, X.; Tan, A.; Chan, R.; Vardhanabhuti, V.; Chu, W.C.W.; Cheng, K.T.; Chen, H. Deep learning in breast cancer imaging: A decade of progress and future directions. IEEE Rev. Biomed. Eng. 2024, 18, 130–151. [Google Scholar] [CrossRef]
  18. Al-Thelaya, K.; Gilal, N.U.; Alzubaidi, M.; Majeed, F.; Agus, M.; Schneider, J.; Househ, M. Applications of discriminative and deep learning feature extraction methods for whole slide image analysis: A survey. J. Pathol. Inform. 2023, 14, 100335. [Google Scholar] [CrossRef]
  19. Hulsen, T. Explainable Artificial Intelligence (XAI): Concepts and Challenges in Healthcare. AI 2023, 4, 652–666. [Google Scholar] [CrossRef]
  20. Mahmood, T.; Li, J.; Pei, Y.; Akhtar, F.; Imran, A.; Rehman, K.U. A brief survey on breast cancer diagnostic with deep learning schemes using multi-image modalities. IEEE Access 2020, 8, 165779–165809. [Google Scholar] [CrossRef]
  21. Mridha, M.F.; Hamid, M.A.; Monowar, M.M.; Keya, A.J.; Ohi, A.Q.; Islam, M.R.; Kim, J.M. A comprehensive survey on deep-learning-based breast cancer diagnosis. Cancers 2021, 13, 6116. [Google Scholar] [CrossRef]
  22. Wen, Z.; Wang, S.; Yang, D.M.; Xie, Y.; Chen, M.; Bishop, J.; Xiao, G. Deep learning in digital pathology for personalized treatment plans of cancer patients. Semin. Diagn. Pathol. 2023, 40, 109–119. [Google Scholar] [CrossRef]
  23. Rabilloud, N.; Allaume, P.; Acosta, O.; De Crevoisier, R.; Bourgade, R.; Loussouarn, D.; Rioux-Leclercq, N.; Khene, Z.E.; Mathieu, R.; Bensalah, K.; et al. Deep learning methodologies applied to digital pathology in prostate cancer: A systematic review. Diagnostics 2023, 13, 2676. [Google Scholar] [CrossRef] [PubMed]
  24. Page, M.J.; McKenzie, J.E.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.; Brennan, S.E.; et al. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ 2021, 372, n71. [Google Scholar] [CrossRef] [PubMed]
  25. Knobloch, K.; Yoon, U.; Vogt, P.M. Preferred reporting items for systematic reviews and meta-analyses (PRISMA) statement and publication bias. J. Cranio-Maxillofac. Surg. 2011, 39, 91–92. [Google Scholar] [CrossRef] [PubMed]
  26. Tricco, A.C.; Lillie, E.; Zarin, W.; O’Brien, K.K.; Colquhoun, H.; Levac, D.; Moher, D.; Peters, M.D.; Horsley, T.; Weeks, L.; et al. PRISMA extension for scoping reviews (PRISMA-ScR): Checklist and explanation. Ann. Intern. Med. 2018, 169, 467–473. [Google Scholar] [CrossRef]
  27. Swiderska-Chadaj, Z.; Gallego, J.; Gonzalez-Lopez, L.; Bueno, G. Detection of Ki67 hot-spots of invasive breast cancer based on convolutional neural networks applied to mutual information of H&E and Ki67 whole slide images. Appl. Sci. 2020, 10, 7761. [Google Scholar] [CrossRef]
  28. George, K.; Faziludeen, S.; Sankaran, P.; Joseph, P.K. Breast cancer detection from biopsy images using nucleus guided transfer learning and belief based fusion. Comput. Biol. Med. 2020, 124, 103954. [Google Scholar] [CrossRef]
  29. Krithiga, R.; Geetha, P. Deep learning based breast cancer detection and classification using fuzzy merging techniques. Mach. Vis. Appl. 2020, 31, 63. [Google Scholar] [CrossRef]
  30. Evangeline, I.K.; Precious, J.G.; Pazhanivel, N.; Kirubha, S.P.A. Automatic Detection and Counting of Lymphocytes from Immunohistochemistry Cancer Images Using Deep Learning. J. Med Biol. Eng. 2020, 40, 735–747. [Google Scholar] [CrossRef]
  31. Geread, R.S.; Sivanandarajah, A.; Brouwer, E.R.; Wood, G.A.; Androutsos, D.; Faragalla, H.; Khademi, A. piNET: An automated proliferation index calculator framework for Ki67 breast cancer images. Cancers 2021, 13, 11. [Google Scholar] [CrossRef]
  32. Krijgsman, D.; van Leeuwen, M.B.; van der Ven, J.; Almeida, V.; Vlutters, R.; Halter, D.; Kuppen, P.J.K.; van de Velde, C.J.H.; Wimberger-Friedl, R. Quantitative whole slide assessment of tumor-infiltrating CD8-positive lymphocytes in ER-positive breast cancer in relation to clinical outcome. IEEE J. Biomed. Health Inform. 2021, 25, 381–392. [Google Scholar] [CrossRef]
  33. Gamble, P.; Jaroensri, R.; Wang, H.; Tan, F.; Moran, M.; Brown, T.; Flament-Auvigne, I.; Rakha, E.A.; Toss, M.; Dabbs, D.J.; et al. Determining breast cancer biomarker status and associated morphological features using deep learning. Commun. Med. 2021, 1, 14. [Google Scholar] [CrossRef] [PubMed]
  34. Lu, X.; You, Z.; Sun, M.; Wu, J.; Zhang, Z. Breast cancer mitotic cell detection using cascade convolutional neural network with U-Net. Math. Biosci. Eng. 2021, 18, 673–695. [Google Scholar] [CrossRef] [PubMed]
  35. Budginaite, E.; Morkunas, M.; Laurinavicius, A.; Treigys, P. Deep Learning Model for Cell Nuclei Segmentation and Lymphocyte Identification in Whole Slide Histology Images. Informatica 2021, 32, 23–40. [Google Scholar] [CrossRef]
  36. Narayanan, P.L.; Raza, S.E.A.; Hall, A.H.; Marks, J.R.; King, L.; West, R.B.; Hernandez, L.; Guppy, N.; Dowsett, M.; Gusterson, B.; et al. Unmasking the immune microecology of ductal carcinoma in situ with deep learning. NPJ Breast Cancer 2021, 7, 19. [Google Scholar] [CrossRef]
  37. Diao, J.A.; Wang, J.K.; Chui, W.F.; Mountain, V.; Gullapally, S.C.; Srinivasan, R.; Mitchell, R.N.; Glass, B.; Hoffman, S.; Rao, S.K.; et al. Human-interpretable image features derived from densely mapped cancer pathology slides predict diverse molecular phenotypes. Nat. Commun. 2021, 12, 1613. [Google Scholar] [CrossRef]
  38. Negahbani, F.; Sabzi, R.; Jahromi, B.P.; Firouzabadi, D.; Movahedi, F.; Shirazi, M.K.; Majidi, S.; Dehghanian, A. PathoNet introduced as a deep neural network backend for evaluation of Ki-67 and tumor-infiltrating lymphocytes in breast cancer. Sci. Rep. 2021, 11, 8489. [Google Scholar] [CrossRef]
  39. Diaz Guerrero, R.E.; Oliveira, J.L. Improvements in lymphocytes detection using deep learning with a preprocessing stage. In Proceedings of the 2021 IEEE 34th International Symposium on Computer-Based Medical Systems (CBMS), Aveiro, Portugal, 7–9 June 2021; pp. 178–182. [Google Scholar] [CrossRef]
  40. Schirris, Y.; Engelaer, M.; Panteli, A.; Horlings, H.M.; Gavves, E.; Teuwen, J. WeakSTIL: Weak whole-slide image level stromal tumor infiltrating lymphocyte scores are all you need. Proc. SPIE Med. Imaging: Digit. Comput. Pathol. 2022, 12039, 120390B. [Google Scholar] [CrossRef]
  41. Ektefaie, Y.; Yuan, W.; Dillon, D.A.; Lin, N.U.; Golden, J.A.; Kohane, I.S.; Yu, K.H. Integrative multiomics-histopathology analysis for breast cancer classification. NPJ Breast Cancer 2021, 7, 147. [Google Scholar] [CrossRef]
  42. Zafar, M.M.; Rauf, Z.; Sohail, A.; Khan, A.R.; Obaidullah, M.; Khan, S.H.; Lee, Y.S.; Khan, A. Detection of tumour infiltrating lymphocytes in CD3 and CD8 stained histopathological images using a two-phase deep CNN. Photodiagnosis Photodyn. Ther. 2022, 37, 102676. [Google Scholar] [CrossRef]
  43. Zhang, X.; Zhu, X.; Tang, K.; Zhao, Y.; Lu, Z.; Feng, Q. DDTNet: A dense dual-task network for tumor-infiltrating lymphocyte detection and segmentation in histopathological images of breast cancer. Med. Image Anal. 2022, 78, 102415. [Google Scholar] [CrossRef]
  44. Chen, Y.; Li, H.; Janowczyk, A.; Toro, P.; Corredor, G.; Whitney, J.; Lu, C.; Koyuncu, C.F.; Mokhtari, M.; Buzzy, C.; et al. Computational pathology improves risk stratification of a multi-gene assay for early stage ER+ breast cancer. NPJ Breast Cancer 2023, 9, 40. [Google Scholar] [CrossRef] [PubMed]
  45. Ali, M.L.; Rauf, Z.; Khan, A.; Sohail, A.; Ullah, R.; Gwak, J. CB-HVT Net: A Channel-Boosted Hybrid Vision Transformer Network for Lymphocyte Detection in Histopathological Images. IEEE Access 2023, 11, 115740–115750. [Google Scholar] [CrossRef]
  46. Huang, J.; Li, H.; Wan, X.; Li, G. Affine-Consistent Transformer for Multi-Class Cell Nuclei Detection. In Proceedings of the 2023 International Conference on Computer Vision (ICCV 2023), Paris, France, 2–3 October 2023; pp. 21327–21336. [Google Scholar] [CrossRef]
  47. Jiao, Y.; van der Laak, J.; Albarqouni, S.; Li, Z.; Tan, T.; Bhalerao, A.; Cheng, S.; Ma, J.; Pocock, J.; Pluim, J.P.W.; et al. LYSTO: The Lymphocyte Assessment Hackathon and Benchmark Dataset. IEEE J. Biomed. Health Inform. 2024, 28, 1161–1172. [Google Scholar] [CrossRef] [PubMed]
  48. Shah, H.A.; Kang, J.M. An Optimized Multi-Organ Cancer Cells Segmentation for Histopathological Images Based on CBAM-Residual U-Net. IEEE Access 2023, 11, 111608–111621. [Google Scholar] [CrossRef]
  49. Yosofvand, M.; Khan, S.Y.; Dhakal, R.; Nejat, A.; Moustaid-Moussa, N.; Rahman, R.L.; Moussa, H. Automated Detection and Scoring of Tumor-Infiltrating Lymphocytes in Breast Cancer Histopathology Slides. Cancers 2023, 15, 3635. [Google Scholar] [CrossRef]
  50. Huang, P.W.; Ouyang, H.; Hsu, B.Y.; Chang, Y.R.; Lin, Y.C.; Chen, Y.A.; Hsieh, Y.H.; Fu, C.C.; Li, C.F.; Lin, C.H.; et al. Deep-learning based breast cancer detection for cross-staining histopathology images. Heliyon 2023, 9, e13171. [Google Scholar] [CrossRef]
  51. Rauf, Z.; Sohail, A.; Khan, S.H.; Khan, A.; Gwak, J.; Maqbool, M. Attention-guided multi-scale deep object detection framework for lymphocyte analysis in IHC histological images. Microscopy 2023, 72, 27–42. [Google Scholar] [CrossRef]
  52. Wang, X.; Zhang, J.; Yang, S.; Xiang, J.; Luo, F.; Wang, M.; Zhang, J.; Yang, W.; Huang, J.; Han, X. A generalizable and robust deep learning algorithm for mitosis detection in multicenter breast histopathological images. Med. Image Anal. 2023, 84, 102703. [Google Scholar] [CrossRef]
  53. Ryu, J.; Puche, A.V.; Shin, J.; Park, S.; Brattoli, B.; Lee, J.; Jung, W.; Cho, S.I.; Paeng, K.; Ock, C.Y.; et al. OCELOT: Overlapped Cell on Tissue Dataset for Histopathology. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Paris, France, 2–3 October 2023; pp. 23902–23912. [Google Scholar] [CrossRef]
  54. Rauf, Z.; Khan, A.R.; Sohail, A.; Alquhayz, H.; Gwak, J.; Khan, A. Lymphocyte detection for cancer analysis using a novel fusion block based channel boosted CNN. Sci. Rep. 2023, 13, 14047. [Google Scholar] [CrossRef]
  55. Makhlouf, S.; Wahab, N.; Toss, M.; Ibrahim, A.; Lashen, A.G.; Atallah, N.M.; Ghannam, S.; Jahanifar, M.; Lu, W.; Graham, S.; et al. Evaluation of Tumour Infiltrating Lymphocytes in Luminal Breast Cancer Using Artificial Intelligence. Br. J. Cancer 2023, 129, 1747–1758. [Google Scholar] [CrossRef]
  56. Aswolinskiy, W.; Munari, E.; Horlings, H.M.; Mulder, L.; Bogina, G.; Sanders, J.; Liu, Y.H.; van den Belt-Dusebout, A.W.; Tessier, L.; Balkenhol, M.; et al. PROACTING: Predicting Pathological Complete Response to Neoadjuvant Chemotherapy in Breast Cancer from Routine Diagnostic Histopathology Biopsies with Deep Learning. Breast Cancer Res. 2023, 25, 142. [Google Scholar] [CrossRef] [PubMed]
  57. Genc-Nayebi, N.; Abran, A. A systematic literature review: Opinion mining studies from mobile app store user reviews. J. Syst. Softw. 2017, 125, 207–219. [Google Scholar] [CrossRef]
  58. Kabir, S.; Vranic, S.; Al Saady, R.M.; Khan, M.S.; Sarmun, R.; Alqahtani, A.; Abbas, T.O.; Chowdhury, M.E.H. The utility of a deep learning-based approach in Her-2/neu assessment in breast cancer. Expert Syst. Appl. 2024, 238, 122051. [Google Scholar] [CrossRef]
  59. Li, Z.; Li, W.; Mai, H.; Zhang, T.; Xiong, Z. Enhancing Cell Detection in Histopathology Images: A ViT-Based U-Net Approach. In Graphs in Biomedical Image Analysis, and Overlapped Cell on Tissue Dataset for Histopathology, 5th MICCAI Workshop, Proceedings of the 5th International Workshop on Graphs in Biomedical Image Analysis (GRAIL) / Workshop on Overlapped Cell on Tissue—Cell Detection from Cell-Tissue Interaction Challenge (OCELOT), MICCAI 2023, Vancouver, BC, Canada, 23 September–4 October 2023; Ahmadi, S.A., Pereira, S., Eds.; Lecture Notes in Computer Science; Springer: Berlin/Heidelberg, Germany, 2024; Volume 14373, pp. 150–160. [Google Scholar] [CrossRef]
  60. Millward, J.; He, Z.; Nibali, A. Dense Prediction of Cell Centroids Using Tissue Context and Cell Refinement. In Graphs in Biomedical Image Analysis, and Overlapped Cell on Tissue Dataset for Histopathology, 5th MICCAI Workshop, Proceedings of the 5th International Workshop on Graphs in Biomedical Image Analysis (GRAIL) and the Workshop on Overlapped Cell on Tissue (OCELOT), MICCAI 2023, Vancouver, BC, Canada, 23 September–4 October 2023; Ahmadi, S.A., Pereira, S., Eds.; Lecture Notes in Computer Science; Springer: Berlin/Heidelberg, Germany, 2024; Volume 14373, pp. 138–149. [Google Scholar] [CrossRef]
  61. Karol, M.; Tabakov, M.; Markowska-Kaczmar, U.; Fulawka, L. Deep Learning for Cancer Cell Detection: Do We Need Dedicated Models? Artif. Intell. Rev. 2024, 57, 53. [Google Scholar] [CrossRef]
  62. Lakshmanan, B.; Anand, S.; Raja, P.S.V.; Selvakumar, B. Improved DeepMitosisNet Framework for Detection of Mitosis in Histopathology Images. Multimed. Tools Appl. 2023, 83, 43303–43324. [Google Scholar] [CrossRef]
  63. Hoerst, F.; Rempe, M.; Heine, L.; Seibold, C.; Keyl, J.; Baldini, G.; Ugurel, S.; Siveke, J.; Gruenwald, B.; Egger, J.; et al. CellViT: Vision Transformers for Precise Cell Segmentation and Classification. Med. Image Anal. 2024, 94, 103143. [Google Scholar] [CrossRef]
  64. Marzouki, A.; Guo, Z.; Zeng, Q.; Kurtz, C.; Loménie, N. Optimizing Lymphocyte Detection in Breast Cancer Whole Slide Imaging through Data-Centric Strategies. arXiv 2024, arXiv:2405.13710. [Google Scholar]
  65. Liu, S.; Amgad, M.; More, D.; Rathore, M.A.; Salgado, R.; Cooper, L.A.D. A panoptic segmentation dataset and deep-learning approach for explainable scoring of tumor-infiltrating lymphocytes. NPJ Breast Cancer 2024, 10, 52. [Google Scholar]
  66. Keele, S. Guidelines for Performing Systematic Literature Reviews in Software Engineering; Technical Report, Version 2.3, EBSE-2007-01; Software Engineering Group, School of Computer Science and Mathematics; Keele University: Keele, UK; Department of Computer Science, University of Durham: Durham, UK, 2007. [Google Scholar]
  67. Tafavvoghi, M.; Bongo, L.A.; Shvetsov, N.; Busund, L.T.R.; Møllersen, K. Publicly available datasets of breast histopathology H&E whole-slide images: A scoping review. J. Pathol. Inform. 2024, 15, 100363. [Google Scholar]
  68. Abdel-Nabi, H.; Ali, M.; Awajan, A.; Daoud, M.; Alazrai, R.; Suganthan, P.N.; Ali, T. A Comprehensive Review of the Deep Learning-Based Tumor Analysis Approaches in Histopathological Images: Segmentation, Classification and Multi-Learning Tasks. Clust. Comput. 2023, 26, 3145–3185. [Google Scholar] [CrossRef]
  69. Basu, A.; Senapati, P.; Deb, M.; Rai, R.; Dhal, K.G. A survey on recent trends in deep learning for nucleus segmentation from histopathology images. Evol. Syst. 2024, 15, 203–248. [Google Scholar]
  70. Wang, L.; Pan, L.; Wang, H.; Liu, M.; Feng, Z.; Rong, P.; Chen, Z.; Peng, S. DHUnet: Dual-branch hierarchical global–local fusion network for whole slide image segmentation. Biomed. Signal Process. Control 2023, 85, 104976. [Google Scholar]
  71. Liu, S.; Amgad, M.; Rathore, M.A.; Salgado, R.; Cooper, L.A.D. A panoptic segmentation approach for tumor-infiltrating lymphocyte assessment: Development of the MuTILs model and PanopTILs dataset. MedRxiv 2022. [Google Scholar] [CrossRef]
  72. Mondello, A.; Dal Bo, M.; Toffoli, G.; Polano, M. Machine learning in onco-pharmacogenomics: A path to precision medicine with many challenges. Front. Pharmacol. 2024, 14, 1260276. [Google Scholar]
  73. Rydzewski, N.R.; Shi, Y.; Li, C.; Chrostek, M.R.; Bakhtiar, H.; Helzer, K.T.; Bootsma, M.L.; Berg, T.J.; Harari, P.M.; Floberg, J.M.; et al. A platform-independent AI tumor lineage and site (ATLAS) classifier. Commun. Biol. 2024, 7, 314. [Google Scholar]
  74. Liu, Y.; He, X.; Yang, Y. Tumor immune microenvironment-based clusters in predicting prognosis and guiding immunotherapy in breast cancer. J. Biosci. 2024, 49, 19. [Google Scholar]
  75. Ahmed, F.; Abdel-Salam, R.; Hamnett, L.; Adewunmi, M.; Ayano, T. Improved Breast Cancer Diagnosis through Transfer Learning on Hematoxylin and Eosin Stained Histology Images. arXiv 2023, arXiv:2309.08745. [Google Scholar]
  76. Nasir, E.S.; Parvaiz, A.; Fraz, M.M. Nuclei and glands instance segmentation in histology images: A narrative review. Artif. Intell. Rev. 2023, 56, 7909–7964. [Google Scholar]
  77. National Cancer Institute. TCGA-BRCA Project. Available online: https://portal.gdc.cancer.gov/projects/TCGA-BRCA (accessed on 18 July 2024).
  78. Veta, M.; Van Diest, P.J.; Willems, S.M.; Wang, H.; Madabhushi, A.; Cruz-Roa, A.; Gonzalez, F.; Larsen, A.B.L.; Vestergaard, J.S.; Dahl, A.B.; et al. Assessment of algorithms for mitosis detection in breast cancer histopathology images. Med. Image Anal. 2015, 20, 237–248. [Google Scholar]
  79. Spanhol, F.A.; Oliveira, L.S.; Petitjean, C.; Heutte, L. A dataset for breast cancer histopathological image classification. IEEE Trans. Biomed. Eng. 2015, 63, 1455–1462. [Google Scholar] [CrossRef]
  80. Qaiser, T.; Mukherjee, A.; Reddy Pb, C.; Munugoti, S.D.; Tallam, V.; Pitkäaho, T.; Lehtimäki, T.; Naughton, T.; Berseth, M.; Pedraza, A.; et al. Her 2 challenge contest: A detailed assessment of automated HER 2 scoring algorithms in whole slide images of breast cancer tissues. Histopathology 2018, 72, 227–238. [Google Scholar] [CrossRef] [PubMed]
  81. Aresta, G.; Araújo, T.; Kwok, S.; Chennamsetty, S.S.; Safwan, M.; Alex, V.; Marami, B.; Prastawa, M.; Chan, M.; Donovan, M.; et al. Bach: Grand challenge on breast cancer histology images. Med. Image Anal. 2019, 56, 122–139. [Google Scholar] [CrossRef] [PubMed]
  82. Kumar, N.; Verma, R.; Anand, D.; Zhou, Y.; Onder, O.F.; Tsougenis, E.; Chen, H.; Heng, P.A.; Li, J.; Hu, Z.; et al. A Multi-Organ Nucleus Segmentation Challenge. IEEE Trans. Med. Imaging 2020, 39, 1380–1391. [Google Scholar] [CrossRef] [PubMed]
  83. Ortega-Ruiz, M.A.; Roman-Rangel, E.; Reyes-Aldasoro, C.C. Multiclass Semantic Segmentation of Immunostained Breast Cancer Tissue with a Deep-Learning Approach. medRxiv 2022. [Google Scholar] [CrossRef]
  84. Gamper, J.; Koohbanani, N.A.; Benes, K.; Graham, S.; Jahanifar, M.; Khurram, S.A.; Azam, A.; Hewitt, K.; Rajpoot, N. Pannuke dataset extension, insights and baselines. arXiv 2020, arXiv:2003.10778. [Google Scholar]
  85. Koohbanani, N.A.; Jahanifar, M.; Tajadin, N.Z.; Rajpoot, N. NuClick: A deep learning framework for interactive segmentation of microscopic images. Med. Image Anal. 2020, 65, 101771. [Google Scholar] [CrossRef]
  86. Jeretic, P.; Warstadt, A.; Bhooshan, S.; Williams, A. Are natural language inference models IMPPRESsive? Learning IMPlicature and PRESupposition. arXiv 2020, arXiv:2004.03066. [Google Scholar]
  87. Brancati, N.; Anniciello, A.M.; Pati, P.; Riccio, D.; Scognamiglio, G.; Jaume, G.; De Pietro, G.; Di Bonito, M.; Foncubierta, A.; Botti, G.; et al. BRACS: A Dataset for BReAst Carcinoma Subtyping in H&E Histology Images. arXiv 2021, arXiv:2111.04740. [Google Scholar]
  88. Grand Challenge. Tiger Challenge. 2022. Available online: https://tiger.grand-challenge.org/ (accessed on 18 July 2024).
  89. Aubreville, M.; Stathonikos, N.; Donovan, T.A.; Klopfleisch, R.; Ammeling, J.; Ganz, J.; Wilm, F.; Veta, M.; Jabari, S.; Eckstein, M.; et al. Domain generalization across tumor types, laboratories, and species—Insights from the 2022 edition of the Mitosis Domain Generalization Challenge. Med. Image Anal. 2024, 94, 103155. [Google Scholar] [CrossRef]
  90. Madoori, P.K.; Sannidhi, S.; Anandam, G. Artificial Intelligence and Computational Pathology: A comprehensive review of advancements and applications. Perspectives 2023, 11, 30. [Google Scholar] [CrossRef]
  91. Kumar, N.; Gupta, R.; Gupta, S. Whole slide imaging (WSI) in pathology: Current perspectives and future directions. J. Digit. Imaging 2020, 33, 1034–1040. [Google Scholar] [CrossRef] [PubMed]
  92. Wagner, S.J.; Matek, C.; Boushehri, S.S.; Boxberg, M.; Lamm, L.; Sadafi, A.; Winter, D.J.E.; Marr, C.; Peng, T. Built to last? Reproducibility and reusability of deep learning algorithms in computational pathology. Mod. Pathol. 2024, 37, 100350. [Google Scholar] [CrossRef] [PubMed]
  93. Nasarian, E.; Alizadehsani, R.; Acharya, U.R.; Tsui, K.L. Designing interpretable ML system to enhance trust in healthcare: A systematic review to proposed responsible clinician-AI-collaboration framework. Inf. Fusion 2024, 108, 102412. [Google Scholar] [CrossRef]
  94. Hossain, M.S.; Shahriar, G.M.; Syeed, M.M.M.; Uddin, M.F.; Hasan, M.; Hossain, M.S.; Bari, R. Tissue artifact segmentation and severity assessment for automatic analysis using WSI. IEEE Access 2023, 11, 21977–21991. [Google Scholar] [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Article Metrics

Citations

Article Access Statistics

Journal Statistics
Multiple requests from the same IP address are counted as one view.
Download PDF
Arrhythmic Risk Stratification in Patients with Arrhythmogenic Cardiomyopathy
Previous
Visualizing Aortic Inflammation by Diffusion-Weighted Whole-Body Imaging with Background Body Signal Suppression (DWIBS)
Next