DEEP LEARNING-BASED ASSESSMENT OF TUMOR-ASSOCIATED STROMA FOR DIAGNOSING BREAST CANCER IN HISTOPATHOLOGY IMAGES

Area of Science:

• Oncology
• Computational Pathology
• Biomarker Discovery

Background:

• Current breast cancer diagnosis relies on epithelial cell morphology and tissue architecture.
• Automated systems predominantly focus on epithelial regions for cancer detection.
• Tumor-associated stroma, a crucial microenvironment component, is often overlooked in automated diagnostics.

Purpose of the Study:

• To develop and evaluate a novel system for breast cancer patient classification using convolutional neural networks.
• To investigate the diagnostic potential of tumor-associated stroma in hematoxylin and eosin (H&E) stained breast specimens.
• To establish stroma as a viable diagnostic biomarker for breast cancer.

Main Methods:

• Development of a classification system employing convolutional neural networks (CNNs).
• Primary focus on the assessment of tumor-associated stroma within H&E stained breast tissue.
• System validation on a large cohort of 646 breast tissue biopsies.

Main Results:

• The proposed system achieved an Area Under the Receiver Operating Characteristic curve (ROC) of 0.92.
• Demonstrated significant discriminative power in classifying breast cancer patients.
• Validated the effectiveness of analyzing tumor-associated stroma for diagnosis.

Conclusions:

• Tumor-associated stroma holds significant potential as a diagnostic biomarker for breast cancer.
• The developed CNN-based system effectively utilizes stromal features for accurate cancer classification.
• This approach offers a promising alternative or adjunct to traditional diagnostic methods.