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

Analysis

Why This Paper Matters

Breast cancer remains a leading cause of death among women, and accurate early detection is critical. Whole slide images (WSIs) are central to digital pathology, but applying deep learning to them is fraught with challenges: image quality variability, scarce high-quality annotations, poor interpretability, high computational costs, and suboptimal processing efficiency. This systematic review directly addresses these issues by synthesizing the state of the art from 2020 to 2024 and proposing a structured evaluation framework.

The paper's significance lies in its comprehensive, PRISMA-guided approach. By analyzing 39 peer-reviewed studies and 20 WSI datasets, it offers a broad yet detailed landscape of current deep learning detection technologies. More importantly, it introduces a five-dimensional evaluation framework that goes beyond mere accuracy, encompassing robustness, interpretability, computational efficiency, and annotation quality. This holistic perspective is crucial for translating research into clinical practice, where these factors are often as important as raw performance.

Technical Contributions

• Systematic Review with PRISMA: A rigorous methodology ensures reproducibility and comprehensiveness, covering studies from 2020 to 2024.
• Five-Dimensional Evaluation Framework: The core contribution is a novel framework that assesses methods across accuracy/performance, robustness/generalization, interpretability, computational efficiency, and annotation quality. This enables a balanced comparison of methods.
• Dataset Compilation: The review catalogs 20 widely used WSI datasets, providing a valuable resource for benchmarking and future research.
• Practical Roadmap: The paper proposes actionable guidance for addressing core challenges, bridging the gap between research and clinical deployment.

Results

The paper does not report quantitative performance metrics (e.g., accuracy, AUC) from individual studies, as it is a systematic review. Instead, its main result is the development and application of the five-dimensional framework. The framework reveals that no single method excels across all dimensions; for instance, high-accuracy models often suffer from poor interpretability or high computational demands. The review also identifies common limitations in existing studies, such as reliance on single datasets and lack of external validation, which the framework helps to highlight. The practical roadmap synthesizes these findings into a step-by-step guide for developing clinically viable WSI-based detection systems.

Significance

This review fills a critical gap in the literature by providing a structured, clinically aligned evaluation methodology for WSI-based breast cancer detection. For AI practitioners, the five-dimensional framework offers a template for evaluating and comparing deep learning models in medical imaging, encouraging a shift from purely accuracy-driven development to a more holistic consideration of real-world constraints. The roadmap can accelerate the translation of research into clinical workflows, potentially improving diagnostic precision and patient prognosis. Moreover, the systematic compilation of datasets and challenges serves as a foundation for future research, fostering reproducibility and collaboration in the field.