Abstract
Diagnosis of breast carcinomas has so far been limited to the morphological interpretation of epithelial cells and the assessment of epithelial tissue architecture. Consequently, most of the automated systems have focused on characterizing the epithelial regions of the breast to detect cancer. In this paper, we propose a system for classification of hematoxylin and eosin (H&E) stained breast specimens based on convolutional neural networks that primarily targets the assessment of tumor-associated stroma to diagnose breast cancer patients. We evaluate the performance of our proposed system using a large cohort containing 646 breast tissue biopsies. Our evaluations show that the proposed system achieves an area under ROC of 0.92, demonstrating the discriminative power of previously neglected tumor-associated stroma as a diagnostic biomarker.
Introduction: Shifting Focus from Seeds to Soil
In traditional pathology, the "gold standard" for breast cancer diagnosis focuses on the "seeds": the epithelial cells. Pathologists look for abnormalities in cell shape, size, and the formation of structures like tubules. Consequently, most computer-aided design (CAD) systems have prioritized segmenting nuclei and glands within these epithelial regions.
However, cancer relies heavily on its "soil"—the stroma—which consists of connective tissue, blood vessels, and inflammatory cells. Recent biological research suggests that "cancer-associated fibroblasts" in the stroma drive tumor progression. Despite this, there is no universally agreed-upon definition of what "cancer-associated stroma" looks like under a microscope, making it difficult to design traditional hand-crafted algorithms for its detection. This study addresses this gap by using deep learning to automatically discover these subtle stromal patterns.
Methods: A Multi-Stage Deep Learning Pipeline
The researchers developed a multi-step pipeline to process Whole-Slide Images (WSIs) using convolutional neural networks (CNNs):
- Tissue Segmentation (CNN I): An 11-layer VGG-like network classifies the image into epithelium, stroma, and fat. This network is fully convolutional, allowing it to process images of any size.
- Stromal Classification (CNN II): A deeper, 16-layer VGG-net focuses specifically on the stromal regions identified in the first step, classifying them as either "normal stroma" or "tumor-associated stroma."
- Feature Extraction: 71 distinct features were extracted from the CNN outputs, including global tissue amounts, epithelial morphology, and spatial architecture (using Delaunay Triangulation and Voronoi diagrams to map tissue distribution).
- Final Classification: These features were processed by a Random Forest classifier to provide the final diagnosis: cancer or non-cancer.
To address the imbalance between normal and tumor-associated stroma, the team implemented a "gradually increasing misclassification loss," forcing the model to become more sensitive to rare tumor-associated regions during training.
Key Results
The system was evaluated on an independent test set of 296 WSIs. The results confirm that the stroma contains sufficient information to accurately diagnose invasive cancer, even without a primary focus on the cancer cells themselves.
- Overall Performance: The combined system achieved an Area Under the ROC Curve (AUC) of 0.921.
- Stroma vs. Epithelium: Features derived from the stroma-specific network (CNN II) were more discriminative (AUC 0.904) than those from the epithelium-focused network (CNN I, AUC 0.882).
- Pixel Accuracy: CNN I achieved 95.5% accuracy in tissue segmentation, while CNN II achieved 92.0% accuracy in distinguishing normal from tumor-associated stroma.
Discussion and Future Work
This research demonstrates that the stroma is a powerful, yet often neglected, diagnostic biomarker. Because pathologists struggle to define the exact morphology of cancer-associated stroma, the CNN acts as a feature discovery tool, identifying statistical variations in the tissue that humans might overlook.
While this study proves the stroma's value for diagnosis, future work will explore its prognostic potential—determining if "cancer-like" stroma in a currently benign biopsy can predict the future risk of cancer development. Additionally, further research is needed to "open the black box" of the CNN to identify the specific stromal textures and patterns it uses, potentially providing new biological insights for clinical pathology.
Summary updated